Skip to content

M
mooc-rr-ressources
Commit 39a87534 authored by Arnaud Legrand's avatar Arnaud Legrand
Browse files
Options

This files have been copied in the orgmode_examples directory...

parent fcc23c8c
Show whitespace changes
Inline Side-by-side
Showing with 0 additions and 41445 deletions
module2/ressources/video_examples/README.org deleted 100755 → 0
View file @ fcc23c8c
# -*- coding: utf-8 -*-
# -*- mode: org -*-
#+TITLE: Org document examples
#+AUTHOR: Arnaud Legrand
#+STARTUP: overview indent inlineimages logdrawer
#+LANGUAGE: en
In the MOOC video, I quickly demo how org-mode can be used in various
contexts. Here are the (sometimes trimmed) corresponding
org-files. These documents depend on many other external data files
and are not meant to lead to reproducible documents but it will give
you an idea of how it can be organized:
1. [[file:journal.org][journal.org]]: an excerpt (I've only left a few code samples and links
to some resources on R, Stats, ...) from my own journal. This is a
personal document where everything (meeting notes, hacking, random
thoughts, ...) goes by default. Entries are created with the =C-c c=
shortcut.
2. [[file:labbook_single.org][labbook_single.org]]: this is an excerpt from the laboratory notebook
[[https://cornebize.net/][Tom Cornebize]] wrote during his Master thesis internship under my
supervision. This a personal labbook. I consider this notebook to be
excellent and was the ideal level of details for us to communicate
without any ambiguity and for him to move forward with confidence.
3. [[file:paper.org][paper.org]]: this is an ongoing paper based on the previous labbook of
Tom Cornebize. As such it is not reproducible as there are hardcoded
paths and uncleaned dependencies but writing it from the labbook was
super easy as we just had to cut and paste the parts we
needed. What may be interesting is the organization and the org
tricks to export to the right LaTeX style. As you may notice, in
the end of the document, there is a commented section with emacs
commands that are automatically executed when opening the file. It
is an effective way to depend less on the =.emacs/init.el= which is
generally customized by everyone.
4. [[file:labbook_several.org][labbook_several.org]]: this is a labbook for a specific project shared
by several persons. As a consequence it starts with information
about installation, common scripts, has section with notes about all
our meetings, a section with information about experiments and an
other one about analysis. Entries could have been labeled by who
wrote them but there were only a few of us and this information was
available in git so we did not bother. In such labbook, it is common
to find annotations indicating that such experiment was =:FLAWED:= as
it had some issues.
5. [[file:technical_report.org][technical_report.org]]: this is a short technical document I wrote
after a colleague sent me a PDF describing an experiment he was
conducting and asked me about how reproducible I felt it was. It
turned out I had to cut and paste the C code from the PDF, then
remove all the line numbers and fix syntax, etc. Obviously I got
quite different performance results but writing everything in
org-mode made it very easy to generate both HTML and PDF and to
explicitly explain how the measurements were done.
Here are a few links to other kind of examples:
- Slides: all my slides for a series of lectures is available here:
https://github.com/alegrand/SMPE. Here is a [[https://raw.githubusercontent.com/alegrand/SMPE/master/lectures/lecture_central_limit_theorem.org][typical source]] and the
[[https://raw.githubusercontent.com/alegrand/SMPE/master/lectures/lecture_central_limit_theorem.pdf][resulting PDF]]
- Lucas Schnorr, a colleague, maintains:
- a set of templates for various computer science
journals/conferences: [[https://github.com/schnorr/ieeeorg][IEEE]], [[https://github.com/schnorr/wileyorg][Wiley]], [[https://github.com/schnorr/acmorg][ACM]], [[https://github.com/schnorr/llncsorg][LNCS]]
- his lecture on programming languages for undergrads:
https://github.com/schnorr/mlp/tree/master/conteudo
module2/ressources/video_examples/journal.org deleted 100755 → 0
View file @ fcc23c8c
# -*- coding: utf-8 -*-
#+TITLE: Blog
#+AUTHOR: Arnaud Legrand
#+HTML_HEAD: <link rel="stylesheet" title="Standard" href="http://orgmode.org/worg/style/worg.css" type="text/css" />
#+STARTUP: overview indent inlineimages logdrawer
#+LANGUAGE: en
#+TAGS: Seminar(s)
#+TAGS: SG(s) WP1(1) WP2(2) WP3(3) WP4(4) WP5(5) WP6(6) WP7(7) WP8(8) WP0(0) Argonne(A)
#+TAGS: POLARIS(P) LIG(L) INRIA (I) HOME(H) Europe(E)
#+TAGS: twitter(t)
#+TAGS: Workload(w) BOINC(b) Blog noexport(n) Stats(S)
#+TAGS: BULL(B)
#+TAGS: autotuning(a)
#+TAGS: Epistemology(E) Vulgarization(V) Teaching(T)
#+TAGS: R(R) Python(p) OrgMode(O) HACSPECIS(h)
#+PROPERTY: header-args :eval never-export
#+EXPORT_SELECT_TAGS: Blog
#+OPTIONS: H:3 num:t toc:t \n:nil @:t ::t |:t ^:t -:t f:t *:t <:t
#+OPTIONS: TeX:t LaTeX:nil skip:nil d:nil todo:t pri:nil tags:not-in-toc
#+LATEX_HEADER: %\usepackage{palatino,a4wide,eurosym,graphicx}\usepackage[francais]{babel}
#+INFOJS_OPT: view:nil toc:nil ltoc:t mouse:underline buttons:0 path:http://orgmode.org/org-info.js
#+EXPORT_SELECT_TAGS: export
#+EXPORT_EXCLUDE_TAGS: noexport
#+EPRESENT_FRAME_LEVEL: 2
#+COLUMNS: %25ITEM %TODO %3PRIORITY %TAGS
#+SEQ_TODO: TODO(t!) STARTED(s!) WAITING(w!) APPT(a!) | DONE(d!) CANCELLED(c!) DEFERRED(f!) DELEGATED(D!)
* 2011
** 2011-02 février
*** 2011-02-08 mardi :R:
**** Pour apprendre:
- Pour les débutants:
http://wiki.stdout.org/rcookbook/
http://www.r-bloggers.com/
http://rstudio.org/ but emacs is just great too once ess is installed
- Indispensables:
+ http://had.co.nz/ggplot2/
+ http://plyr.had.co.nz/ et la démonstration par l'exemple
http://plyr.had.co.nz/09-user/
- Une intro pas trop mal faite:
- http://bioconnector.github.io/workshops/lessons/intro-r-lifesci/01-intro-r/
- Encore plus interactif: http://swirlstats.com/
- Plus avancé:
http://cran.r-project.org/doc/contrib/Paradis-rdebuts_fr.pdf
- Pour ceux qui veulent aller plus loin et coder
http://zoonek2.free.fr/UNIX/48_R/all.html
- Bien plus avancé pour les fans de sémantique et de ruses de
fou par Hadlay Wickam:
http://adv-r.had.co.nz/Computing-on-the-language.html
- Un [[http://ww2.coastal.edu/kingw/statistics/R-tutorials/dataframes.html][excellent tutorial on data frames]] (=attach=, =with=, =rownames=,
=dimnames=, notions of scope...)
**** R 101 :Blog:
[[file:public_html/blog/2012/09/12/R101.org][Moved to the blog]]
[[file:~/Work/SimGrid/infra-songs/slides/140422-compas-R101/R101.org][Compas tutorial]]
**** R tricks
***** Reshaping
http://www.statmethods.net/management/reshape.html
#+begin_src R :results output :session :exports both
# example of melt function
library(reshape)
mdata <- melt(mydata, id=c("id","time"))
#+end_src
***** somme d'éléments avec fenêtre glissante
#+begin_src R
filter(x, rep(1,4))
#+end_src
***** sorting a data frame
#+BEGIN_SRC R
dd[with(dd, order(-z, b)), ]
#+END_SRC
***** Capture output
#+begin_src R
sink("myfile.txt", append=TRUE, split=TRUE)
#+end_src
When redirecting output, use the cat() function to annotate the
output.
***** Batch processing
#+begin_src sh
R CMD BATCH [options] my_script.R [outfile]
#+end_src
***** Convenient commands
- describe
- structure
- ddply
- cbind/rbind
***** Labels and Factors
http://stackoverflow.com/questions/12075037/ggplot-legends-change-labels-order-and-title
#+begin_src R
dtt$model <- factor(dtt$model, levels=c("mb", "ma", "mc"), labels=c("MBB", "MAA", "MCC"))
#+end_src
here is another way of reordering factors:
#+begin_src R
dtt$model <- relevel(dtt$model, ref="MBB").
#+end_src
This puts the factor given by ref at the beginning.
***** "parallel" Prefix
#+BEGIN_SRC
cumsum
#+END_SRC
***** knitr preembule
check out "Tools for making a paper" in R-bloggers:
#+BEGIN_SRC
<<set-options, echo=FALSE, cache=FALSE>>=
opts_knit$set(stop_on_error=2L)
@
<<loadpackages,echo=FALSE>>=
suppressMessages(require(memisc))
@
#+END_SRC
***** Annotate in facet_wrap/facet_grid
http://www.ansci.wisc.edu/morota/R/ggplot2/ggplot2.html
***** Interactive plotting
http://rstudio.org/docs/advanced/manipulate
[[file:~/Work/SimGrid/infra-songs/WP4/R/Sweep3D_analysis/analyze.Rnw]]
#+begin_src R
GC <- function(df,start,end) {
ggplot(
df[(df$Start>=start & df$Start<=end)|(df$End>=start &
df$End<=end)|
(df$Start<=start & df$End>=end) ,],
aes(xmin=Start,xmax=End, ymin=ResourceId, ymax=ResourceId+1,
fill=Value))+
theme_bw()+geom_rect()+coord_cartesian(xlim = c(start, end))
}
GC(df_tau,1.1,1.2)
animate(GC(df_tau, start, end),start=slider...)
#+end_src
***** scoping issue with ggplot: mixing external variables with column names
There is a magical function designed for this: here()
#+BEGIN_EXAMPLE
ddply(df_native, c("ResourceId"), here(transform),
Chunk = compute_chunk(Start,End,Duration,min_time_pure))
#+END_EXAMPLE
Here, min_time_pure is an external variable, not a column name.
***** speeding things up with parallel plyR
#+BEGIN_SRC R
library(doMP)
library(plyr)
perf_win <- ddply(df_win,c("host_id"), summarize,
astro_avg=sum(et_avg*astro_win),
astro_var=sum(et_var*astro_win),
seti_avg=sum(et_avg*seti_win),
seti_var=sum(et_var*seti_win),
.parallel=TRUE, .progress = "text")
#+END_SRC
***** dessin de graphe avec courbes de bezier
https://gist.github.com/dsparks/4331058
***** Side effect in local functions
http://my.safaribooksonline.com/book/programming/r/9781449377502/9dot-functions/id3440389
***** Non-standard evaluation
http://adv-r.had.co.nz/Computing-on-the-language.html
***** Arrays of functions in for loops
http://stackoverflow.com/questions/26064649/enclosing-variables-within-for-loop
**** R weblinks/statistiques r-cran :WP8:
http://en.wikibooks.org/wiki/R_Programming/Graphics
http://freecode.com/articles/creating-charts-and-graphs-with-gnu-r
http://www.statmethods.net/graphs/density.html
http://addictedtor.free.fr/graphiques/RGraphGallery.php?graph=78
(scatterplot + histogram)
http://cran.r-project.org/doc/contrib/Ricci-distributions-en.pdf
(comparaison et fitting de distributions)
http://www.sr.bham.ac.uk/~ajrs/R/r-gallery.html
http://www.som.yale.edu/faculty/pks4/files/teaching/handouts/r2_tstat_explained.pdf
about t-values
http://www.statmethods.net/stats/anova.html
http://www.stat.wisc.edu/courses/st850-lindstro/handouts/blocking.pdf
(blocking dans un anova en R)
http://www-rocq.inria.fr/axis/modulad/archives/numero-34/Goupy-34/goupy-34.pdf
(tutorial on DOE in French)
file:/home/alegrand/Work/Documents/Enseignements/M2R_Mesures_Analyse_Eval_Perf_06/Intro_Statistics/doesimp2excerpt--chap3.pdf
file:/home/alegrand/Work/Documents/Enseignements/M2R_Mesures_Analyse_Eval_Perf_06/Intro_Statistics/doeprimer.pdf
DOE
http://cran.r-project.org/doc/contrib/Faraway-PRA.pdf
(gros bouquin de R sur l'ANOVA)
http://pages.cs.wisc.edu/~cyffka/R_regression-and-anova.pdf
https://marvelig.liglab.fr/doku.php/thematiques/methodologie/accueil
Documents de Nadine Mandran + pointeurs vers cours de stat
http://pbil.univ-lyon1.fr/R/pdf/bsa.pdf
http://grasland.script.univ-paris-diderot.fr/go303/ch5/doc_ch5.htm
Document sur l'analyse de données spatialisées
http://nsaunders.wordpress.com/2010/08/20/a-brief-introduction-to-apply-in-r/
Utilisation de Apply
http://zoonek2.free.fr/UNIX/48_R/all.html
Un utilisateur de R qui a noté tout un tas de choses utiles et
d'exemples, nottamment de programmation.
http://sharpstatistics.co.uk/r/ggplot/
http://rug.mnhn.fr/semin-r/PDF/INED-SFdS-MNHN_Sueur_280411.pdf
Tutorial ggplot2
https://catalyst.uw.edu/workspace/tbranch/24589/155528
Cours sur la visu en R à la Tufte
***** Linear regression and heteroscedasticity :ATTACH:
:PROPERTIES:
:Attachments: ModeleLineaireRegrDegerine.pdf Regression101R.pdf GLSHeteroskedasticity.pdf week2_ht.pdf
:ID: b3ced951-cda8-40ce-b281-cc71b55f1da9
:END:
- http://ljk.imag.fr/membres/Anatoli.Iouditski/cours/MLDESS.pdf (see
attachment) cours en français sur la régression linéaire, vision
proba.
- http://smat.epfl.ch/courses/Regression/Slides/week2_ht.pdf slides
sur la régression linéaire et le lien avec maximum likelihood
- http://www.r-tutor.com/elementary-statistics/simple-linear-regression/confidence-interval-linear-regression
#+begin_src R :results output :session :exports both
predict(eruption.lm, newdata, interval="confidence")
#+end_src
- http://www.princeton.edu/~otorres/Regression101R.pdf (std error et
confidence interval on parameters estimates + heteroscedasticity)
- http://www.econ.uiuc.edu/~wsosa/econ471/GLSHeteroskedasticity.pdf
Comment gérer l'hétéroscedastisité.
***** Time series :ATTACH:
:PROPERTIES:
:Attachments: SCBio.pdf
:ID: 40d5498d-e8b3-4c73-8722-7d0056667c15
:END:
http://ljk.imag.fr/membres/Serge.Degerine/Enseignement/SCBio.pdf
**** Quantile Regression and Bootstrap :Stats:ATTACH:
:PROPERTIES:
:Attachments: mcgill-r.pdf st-m-app-bootstrap.pdf stnews70.pdf
:ID: 8e3038dc-fa3e-4a7d-a4b1-216513e4359f
:END:
http://freakonometrics.hypotheses.org/date/2012/04 (open data and
ecological falacies (Simpson's paradox)).
http://freakonometrics.hypotheses.org/2396
(Talk-on-quantiles-at-the-R-Montreal-group)
http://www.cscu.cornell.edu/news/statnews/stnews70.pdf
**** Reproducible research :WP8:
Andrew Davison tutorial, which is full of interesting references:
http://rrcns.readthedocs.org/en/latest/index.html
***** org-mode
Une autre approche, uniquement en org.
http://orgmode.org/worg/org-contrib/babel/how-to-use-Org-Babel-for-R.html
***** R/Sweave/knitr
http://users.stat.umn.edu/~geyer//Sweave/
Sweave, exemple minimaux, emacs.
http://www.bepress.com/cgi/viewcontent.cgi?article=1001&context=bioconductor
Un article sur reproducible research et sweave
http://cran.r-project.org/web/packages/pgfSweave/vignettes/pgfSweave.pdf
Pgfsweave, un paquet latex qui améliore le look et la vitesse de
sweave. Le paquet est mort ceci dit et mes premiers essais n'étaient
pas concluants car tout convertir en pgf, c'est un peu bourrin.
http://yihui.name/knitr/
knitr, le dernier, bien à la mode, stable et très prometteur
http://www.stat.uiowa.edu/~rlenth/StatWeave/OLD/SRC-talk.pdf
StatWeave. Permet aussi de mettre du Maple.
***** Ipython notebook
https://osf.io/h9gsd/
sympa, facile à mettre en place
***** ActivePapers
- http://www.activepapers.org/
- https://bitbucket.org/khinsen/active_papers_py/wiki/Tutorial
***** Elsevier approach
http://www.elsevier.com/physical-sciences/computer-science/executable-papers
https://collage.elsevier.com/manual/
http://is.ieis.tue.nl/staff/pvgorp/research/?page=SCP11
***** Research Gate
https://www.researchgate.net/publicliterature.OpenReviewInfo.html
***** Conference or general discussions
http://reproducibleresearch.net/index.php/Main_Page
http://wiki.stodden.net/Main_Page
- [[http://www.eecg.toronto.edu/~enright/wddd/][Workshop on Duplicating, Deconstructing and Debunking (WDDD)]] ([[http://cag.engr.uconn.edu/isca2014/workshop_tutorial.html][2014
edition]])
- http://evaluate2010.inf.usi.ch
- [[http://www.stodden.net/AMP2011/][Reproducible Research: Tools and Strategies for Scientific Computing]]
- [[http://wssspe.researchcomputing.org.uk/][Working towards Sustainable Software for Science: Practice and
Experiences]]
- [[http://hunoldscience.net/conf/reppar14/pc.html][REPPAR'14: 1st International Workshop on Reproducibility in Parallel
Computing]]
- [[https://www.xsede.org/web/reproducibility][Reproducibility@XSEDE: An XSEDE14 Workshop]]
- [[http://www.occamportal.org/reproduce][Reproduce/HPCA 2014]]
- [[http://www.ctuning.org/cm/wiki/index.php?title%3DEvents:TRUST2014][TRUST 2014]]
- http://vee2014.cs.technion.ac.il/docs/VEE14-present602.pdf
http://www-958.ibm.com/software/data/cognos/manyeyes/visualizations
http://www.myexperiment.org/
http://wiki.galaxyproject.org/
http://www.runmycode.org/CompanionSite/
http://evaluate.inf.usi.ch/
github ?
workflow ?
vistrails ?
sumatra
vcr
***** Politics
http://michaelnielsen.org/blog/how-you-can-help-the-federal-research-public-access-act-frpaa-become-law/7
http://en.wikipedia.org/wiki/Federal_Research_Public_Access_Act
http://michaelnielsen.org/blog/on-elsevier/
**** General discussions about scientific practice :WP8:
http://michaelnielsen.org/blog/three-myths-about-scientific-peer-review*/
http://michaelnielsen.org/blog/some-garbage-in-gold-out/
http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1003285
**** Coursera
- https://www.coursera.org/course/compdata
- https://class.coursera.org/exdata-002/lecture
- https://class.coursera.org/repdata-002
**** ggplot2 cool examples
http://felixfan.github.io/rstudy/2014/02/28/ggplot2-cheatsheet/
http://blog.revolutionanalytics.com/graphics/
http://grrrraphics.blogspot.com.br/2012/05/ever-wanted-to-see-at-glance.html
http://www.ancienteco.com/2012/03/basic-introduction-to-ggplot2.html
http://sape.inf.usi.ch/quick-reference/ggplot2
http://www.r-bloggers.com/overplotting-solution-for-black-and-white-graphics/
http://stats.stackexchange.com/questions/12029/is-it-possible-to-create-parallel-sets-plot-using-r
http://novyden.blogspot.fr/2013/09/how-to-expand-color-palette-with-ggplot.html
https://gastonsanchez.wordpress.com/2012/08/27/scatterplot-matrices-with-ggplot/
**** Visualisations
http://www.visual-literacy.org/periodic_table/periodic_table.html
**** Design of Experiments (DoE)
- Montgommery book
- http://www.cs.wayne.edu/~hzhang/courses/7290/Lectures/4%20-%20Introduction%20to%20Experimental%20Design.pdf
- http://www.obgyn.cam.ac.uk/cam-only/statsbook/stexdes.html#3g
- http://mescal.imag.fr/membres/arnaud.legrand/teaching/2011/EP_czitrom.pdf
- http://www.basic.northwestern.edu/statguidefiles/oneway_anova_ass_viol.html
- http://techdigest.jhuapl.edu/TD/td2703/telford.pdf
*** 2011-02-15 mardi
**** Réunion CIGRI
***** Présents
- Olivier Richard, Mcf UJF/MESCAL, gestions de ressources,
initiateur de OAR et Cigri, G5K
- Bruno Bzeznik, Ingénieur CIMENT (admin, gestion clusters) et
MESCAL (dev OAR, outils pour CIMENT).
- Chislain Charrier, Ingénieur INRIA G5K à Rennes depuis quelques
mois. Mission: s'occuper des campagnes d'expérimentations.
- Philippe Leprouster, Ingénieur CDD UJF MESCAL pour bosser sur
l'optimisation d'OAR
- Bernard Boutherin, responsable info au LPSC, noeud Tier3 de la
grille EGI (600 coeurs de calcul, 700 To de stoquage,
précurseur autour du free-cooling, installation à moins de 60
kW depuis 2008).
- Catherine Biscarat, IR CNRS qui va s'occuper de la liaison
CIGRI/LPSC.
- Pierre Neyron, IR CNRS MESCAL/MOAIS, responsable de digitalis.
***** Point de Bruno sur l'état actuel de Cigri
Site web: https://ciment-grid.ujf-grenoble.fr
Logiciel principalement déployé dans CIMENT. Exploite
actuellement 3000 cores sur une vingtaine de machines.
R2D2 et fostino sont les plus grosses et gérées par un seul
serveur OAR.
Resources très faiblement utilisées (en général un ou deux
utilisateurs à un instant donnée). Besoin d'accompagner les
utilisateurs qui n'ont pas forcément conscience que CIGRI est
adapté à leurs besoins. Les utilisateur qui utilisent CIGRI
actuellement sont de gros consommateurs de ressources.
Collaboration CIGRI/LPSC initiée par un projet autour du
stoquage. Bruno a du coup équipé CIGRI de noeuds de stoquage et
a déployé Irods.
***** Plus d'infos sur : http://wiki-oar.imag.fr/index.php/CiGri-ng
Entered on [2011-02-15 mar. 09:41]
[[file:~/Liste.org]]
* 2012
** 2013-02 février
**** 2013-02-11 lundi
***** Reproducible research links :WP8:R:
http://wiki.stodden.net/ICERM_Reproducibility_in_Computational_and_Experimental_Mathematics:_Readings_and_References
http://www.rpubs.com/
An interesting article with a dissenting opinion on reproducible research:
http://cogprints.org/8675/
Entered on [2013-02-11 lun. 09:52]
***** Audio StarPU :WP4:
Lionel, Samuel, Paul, Luka, Brice.
****** Séquentialisation des comms
- Idées: faire des mesures automatiques
- Deux implems' (Sam & Paul), pas équivalentes, l'une modélisant
plus les communications synchrone et l'autre les
asynchrones. À creuser.
****** Petites macros pour mesurer/injecter le temps
- Temps injecté dans la version initiale de Sam = temps moyen
observé par StarPU .
- Une fois les problème de communication réglés (virer le
slow-start, séquentialiser ce qui doit l'être), les dernières
différences viennent de la variabilité vraie vie / simu
(surtout sur CPU).
- Objectif: insérer variabilité. C'est le même problème que pour
SMPI. Dans la version actuelle, on regarde le temps pris lors
de la simulation et on le réinjecte, d'où une très mauvaise portabili
- Idée: identifier les blocs, capturer les temps et utiliser en
simu un tirage à partir du profil capturé. C'est assez
"nouveau" car FSuter capturait une trace de niveau MPI donc
sans info sur quel bloc de code => pas d'information sur la
source de la variabilité.
- On commence par une approche basique: à la compilation, on
identifie un bloc par FILE,LINE, avec éventuellement une
extension via une annotation manuelle (c'est le cas pour
StarPU qui lance les calculs toujours au même endroit).
- Niveau workflow, première exécution pour avoir les timings,
puis R, puis réinsertion ds SG.
- La capture est pas compliquée et comme il y a le même besoin
pour SMPI, on factorise pour éviter les divergences. Ce code
est donc dans SG. Paul et Luka ont fait ça la semaine dernière
et Paul l'a utilisé dans *PU, reste à tester pour confirmer
- Luka essaie maintenant de mettre ça ds SMPI, c'est plus
difficile de savoir où mettre les benchmarks. L'ideal serait
de regarder dans la pile, c'est un peu compliqué donc on reste
sur notre approche simple pour l'instant et on raffinera plus
tard si c'est vraiment nécessaire. L'avantage escompté, c'est
sur les plates-formes Mt Blanc par exemples, on peut exécuter
une fois et utiliser ensuite les timings pour faire des tests
de scalabilité sur une vraie machine de brute qui va vite.
****** Objectifs des uns et des autres
- Lionel & Paul à Bdx: objectif = proposer des modèles, support
- Sam: objectif = bricoler DES ordonnanceurs, lancer vite sur
différentes et évaluer l'impact de tailles de blocs ou de la
taille d'une fenêtre glissante. C'est donc clairement un outil
de développement pour tester des choses et il faut donc que
l'outil soit un minimum stable. Rien de grave mais il faut
bien en être conscient en terme de développement. Il sera
important de propager les informations du genre "attention on
a corrigé un truc, ça risque d'invalider les expériences
précédentes".
- Arnaud rappelle que d'un point de vue développement, c'est
comme pour SMPI, il faut être conscient qu'il y a trois types
de tâches toutes aussi importantes les unes que les autres
(i.e., quand on en néglige une on s'en mord toujours les
doigts à un moment ou à un autre):
+ Exploration: le plus fun, de petites expériences pour voir
si ça marche. Pour moi, la démarche de Sam et ses
expériences faites rentrent dans cette catégorie.
+ Ingénierie: écriture de code, petites fonctions
techniques. Dans le contexte de starPU, typiquement, il
s'agit du travail initial de Sam mais aussi du codage de la
séquentialisation des communications ou bien des macros de
capture de traces.
+ Consolidation: moins drôle, mais il faut le faire pour
vérifier que tout le monde peut fairer ses mesures,
réutiliser, et qu'on puisse avancer en toute
confiance. Souvent, sur une nouvelle machine, de nouveaux
phénomènes apparaissent et ce n'est qu'avec des outils
d'exploration systématique et automatiques qu'on s'en sort.
Il faut donc mettre en place dès le début des outils de capture
d'information et d'analyse.
****** Roadmap
- Séquentialisation / Parallélisation: Paul à Bordeaux s'en
occupe. Il met en place le code qui crache une matrice
d'interférences et met en place dans StarPU/SG le code qui
l'exploite.
- Infrastructure de mesure / collecte de traces: Luka et Arnaud à
Grenoble s'en occupent. Réflexion sur un workflow qui va bien
pour garder de bonnes traces et pouvoir facilement tester de
nouvelles machines.
- Paul vient le 14 Mars à Grenoble et on en profitera pour faire
le point.
****** Divers
******* Campagne d'expériences pour valider le modèle / invalider les précédents. Comment faire ?
Il est difficile (impossible ?) de dire qu'un modèle est
valide. Il est plus raisonnable de montrer à quel point on a
essayer de l'invalider, ce qui permet à chacun d'évaluer à quel
point il fait confiance aux capacités d'extrapolation et
d'explication du modèle.
On peut donc montrer l'impact des améliorations successives du
modèle, soit sur le temps final soit sur des choses plus fines
de la trace. Il faut tester sur des cas de plus en plus
complexes, d'où la nécessité d'avoir une méthode un peu
automatique pour comparer des résultats. À titre d'illustration
voici le genre de choses que Martin a raconté à l'éval d'Héméra.
http://mescal.imag.fr/membres/arnaud.legrand/uss_simgrid/130211-HEMERA-eval.pdf
+ Mesure 0: makespan. C'est ce qui nous intéresse mais c'est
généralement très pauvre et on peut arriver par hasard à de
bons résultats ou avoir de mauvais résultats juste parce
qu'un paramètre a été mal mesuré. Même si c'est uniquement,
cette mesure là qui nous intéresse au final, il est
indispensable de comparer pour des mesures plus fines car
c'est ce qui permet de mettre une certaine confiance dans les
capacités d'extrapolation de l'outil.
+ Mesure 1: comparaison visuelle de gantt chartt, peut être
joli, facile et instructif avec R, mais difficilement
quantifiable.
+ Mesure 2: Regarder les distributions de temps passés dans
différents états. On peut faire ça en partie pour StarPU mais
uniquement pour les temps de calcul, pas pour les temps de
communication. En effet, on a peu de maîtrises sur les temps
de comms, on ne sait pas vraiment à quel moment la
communication s'est terminée ni quand elle a commencé.
+ Mesure 3: Comparer les schedules... C'est difficile car déjà
quand c'est stable, une métrique n'est pas évidente à définir
mais quand en plus c'est variable d'une fois sur
l'autre... Idéalement, il faudrait comparer plutôt la
distribution des schedules plutôt que les schedules
individuels... C'est difficile mais passionnant. Ce qui est
super, c'est qu'on a l'outil qui permet de générer les
traces.
******* Problème de main() entre *PU et SG
Passer systématiquement par le XML a résolu le problème de
l'initialisation et du lancement de simgrid. Ça reste gênant
parce qu'il faut recompiler l'appli Sam. aimerait pouvoir
changer l'appli en changeant le LD_LIBRARY_PATH, avec 2
versions de libstarpu.so => Mais alors comment passer les
arguments à SG ?
C'est l'appli qui donne argc,argv dans l'appel à *pu_init
L'appli risque de pas aimer les options de SG.
*pu passe par des variables d'environnements pour éviter ces soucis
Vision d'arnaud : --platform=toto.xml mangé par *pu, reste des
args mangés par SG Comment faire pour mettre les stats de temps
d'exécution adaptés au fichier XML Ptetre qu'on peut identifier
les GPUs par modèle plutôt que par numéro de GPU Possibilité de
vérifier les hostname (STARPU_HOSTNAME)
******* Comment faire pour que ça marche pour exécuter en local.
On peut pondre le .xml, tout classer par hostname. Tout est dans
.starpu/sampling, pour les différents hostnames et codelets, les
traces de perfs.
Du coup ce serait *pu qui génèrerait automatiquement le .xml ?
ça coûte pas cher. Mais les valeurs qu'on met dedans, comment
on les obtient ?
+ BP/latence du bus sont mesuréesde toutes façons au départ
+ Pour la matrice d'interférence de Paul, il faudrait aussi la
mettre dans le XML.
Du coup, il serait peut-être plus naturel que ce soit le script
de Paul qui ponde le .XML qui s'occupe de l'étalonnage
Entered on [2013-02-11 lun. 09:53]
** 2013-05 mai
*** 2013-05-21 mardi
**** BIS Workshop
#+BEGIN_SRC sh :results output raw :exports both
for i in gnome gnome-desktop-environment ifupdown iproute iproute-dev isc-dhcp-client libatm1 network-manager network-manager-gnome ; do
dir=`apt-cache showsrc $i | grep Directory | sed 's/.*: //'`
version=`apt-cache showsrc $i | grep ^Version | sed 's/.*: *//g'`
echo "wget http://http.us.debian.org/debian/$dir/$i""_$version"_amd64.deb
done
#+END_SRC
Entered on [2013-05-21 mar. 08:46]
**** Discussions avec Anne-Cécile à propos des timeouts TCP MPI :WP4:
I have some related news. I had the chance to chat with
Anne-Cecile and talked her about our timeout problem. After
digging a little, she was able to point me to related work:
- Understanding TCP Incast Throughput Collapse in Datacenter
Networks
- Safe and Effective Fine-grained TCP Retransmissions for
Datacenter Communication
- On the properties of an adaptive TCP minimum rto
- http://www.hjp.at/doc/rfc/rfc2988.txt
So the problem has a name (incast) and is linked to the following
TCP parameter:
| OS | TCP RTOmin |
|---------+-------------|
| Linux | 200ms |
| BSD | 200ms |
| Solaris | 400ms |
I haven't read the articles so I don't know the details. All I
can say so far is I don't know how to change this parameter
without recompiling the kernel...
#+BEGIN_SRC sh :results output raw :exports both
cd /usr/src/linux-headers-3.2.0-4-common
cg 'define *TCP_RTO_MIN' '*'
cg 'define *HZ' '*'
#+END_SRC
Basically, this value is good enough for wide area where RTT is
large but in our SAN setting, it's rather bad. Looking further,
http://comments.gmane.org/gmane.linux.network/162986, I learnt
that although this parameter cannot be modified through sysctl,
it could be overriden per route with iproute.
#+BEGIN_SRC sh :results output text :exports both
for i in ip ip-address ipcontroller iplogger ip-netns iproxy iptables-save ipython ip6tables ip-addrlabel ipcrm ip-maddress ip-ntable ip-rule iptables-xml ipython2.6 ip6tables-apply ipc ipcs ip-monitor ipptool iptables ip-tunnel ipython2.7 ip6tables-restore ipcluster ipengine ip-mroute ipptoolfile iptables-apply ipv6 ip6tables-save ipcmk ip-link ip-neighbour ip-route iptables-restore ip-xfrm ; do
man -T --troff-device=ascii $i | grep -i rto
done
#+END_SRC
#+RESULTS:
Entered on [2013-05-21 mar. 12:37]
**** Play with R xkcd :R:
#+begin_src R :results output :session :exports both
# install.packages('xkcd') # did not work so I made it manually
library(extrafont)
download.file("http://simonsoftware.se/other/xkcd.ttf", dest="xkcd.ttf")
system("mkdir ~/.fonts")
system("cp xkcd.tff -t ~/.fonts")
font_import()
loadfonts()
#+END_SRC
#+BEGIN_SRC R :results graphics :file /tmp/plot.png :exports results :width 600 :height 200 :session
library(xkcd)
theme_xkcd <- theme(
panel.background = element_rect(fill="white"),
axis.ticks = element_line(colour=NA),
panel.grid = element_line(colour="white"),
axis.text.y = element_text(colour=NA),
axis.text.x = element_text(colour="black"),
text = element_text(size=16, family="xkcd")
)
ggplot(data.frame(x=c(0, 10)), aes(x)) + theme_xkcd +
stat_function(fun=sin,position="jitter", color="red", size=2) +
stat_function(fun=cos,position="jitter", color="white", size=3) +
stat_function(fun=cos,position="jitter", color="blue", size=2) +
geom_text(family="xkcd", x=4, y=0.7, label="A SIN AND COS CURVE")+
xkcdaxis(c(0, 10),c(-1,1))
#+END_SRC
#+RESULTS:
[[file:/tmp/plot.png]]
Entered on [2013-05-21 mar. 15:49]
* 2015
** 2015-01 janvier
*** 2015-01-07 mercredi
**** Helping Martin with R :Teaching:R:
#+tblname: daily
| Date | exos_java | traces_java | exos_python | traces_python | exos_scala | traces_scala |
|------------+-----------+-------------+-------------+---------------+------------+--------------|
| 2014.9.2 | 6 | 1 | 0 | 0 | 0 | 0 |
| 2014.9.3 | 5 | 1 | 0 | 0 | 0 | 0 |
| 2014.9.4 | 8 | 2 | 0 | 0 | 0 | 0 |
| 2014.9.8 | 7 | 4 | 0 | 0 | 1290 | 86 |
| 2014.9.9 | 0 | 0 | 3 | 1 | 1615 | 86 |
| 2014.9.10 | 0 | 0 | 1 | 1 | 163 | 16 |
| 2014.9.11 | 3 | 2 | 0 | 0 | 999 | 63 |
| 2014.9.12 | 67 | 4 | 2 | 2 | 1149 | 67 |
| 2014.9.13 | 20 | 3 | 1 | 1 | 132 | 14 |
| 2014.9.14 | 7 | 1 | 0 | 0 | 170 | 12 |
| 2014.9.15 | 9 | 2 | 0 | 0 | 1112 | 73 |
| 2014.9.16 | 16 | 2 | 0 | 0 | 768 | 60 |
| 2014.9.17 | 36 | 3 | 0 | 0 | 274 | 40 |
| 2014.9.18 | 1 | 1 | 22 | 2 | 20 | 2 |
| 2014.9.19 | 1 | 1 | 18 | 2 | 10 | 2 |
| 2014.9.20 | 0 | 0 | 12 | 1 | 61 | 6 |
| 2014.9.21 | 0 | 0 | 6 | 2 | 36 | 6 |
| 2014.9.22 | 3 | 2 | 11 | 2 | 420 | 50 |
| 2014.9.23 | 1 | 1 | 0 | 0 | 218 | 31 |
| 2014.9.24 | 0 | 0 | 12 | 2 | 39 | 4 |
| 2014.9.25 | 0 | 0 | 1 | 1 | 220 | 30 |
| 2014.9.26 | 0 | 0 | 19 | 2 | 28 | 5 |
| 2014.9.27 | 0 | 0 | 10 | 1 | 17 | 4 |
| 2014.9.28 | 0 | 0 | 12 | 2 | 37 | 6 |
| 2014.9.29 | 26 | 3 | 8 | 1 | 509 | 81 |
| 2014.9.30 | 9 | 2 | 16 | 2 | 243 | 36 |
| 2014.10.1 | 1 | 1 | 26 | 14 | 99 | 16 |
| 2014.10.2 | 1 | 1 | 1 | 1 | 325 | 38 |
| 2014.10.3 | 26 | 15 | 52 | 16 | 22 | 4 |
| 2014.10.4 | 25 | 1 | 4 | 3 | 36 | 9 |
| 2014.10.5 | 10 | 1 | 2 | 1 | 49 | 9 |
| 2014.10.6 | 5 | 2 | 39 | 22 | 192 | 37 |
| 2014.10.7 | 24 | 4 | 17 | 7 | 143 | 25 |
| 2014.10.8 | 50 | 3 | 0 | 0 | 77 | 14 |
| 2014.10.9 | 24 | 2 | 11 | 3 | 48 | 9 |
| 2014.10.10 | 35 | 4 | 7 | 2 | 0 | 0 |
| 2014.10.11 | 0 | 0 | 9 | 3 | 3 | 1 |
| 2014.10.12 | 20 | 6 | 7 | 3 | 10 | 1 |
| 2014.10.13 | 32 | 4 | 18 | 4 | 0 | 0 |
| 2014.10.14 | 44 | 1 | 41 | 3 | 8 | 1 |
| 2014.10.15 | 5 | 3 | 64 | 10 | 6 | 2 |
| 2014.10.16 | 27 | 2 | 24 | 5 | 1 | 1 |
| 2014.10.17 | 43 | 3 | 14 | 4 | 0 | 0 |
| 2014.10.18 | 84 | 2 | 57 | 8 | 0 | 0 |
| 2014.10.19 | 10 | 2 | 86 | 11 | 0 | 0 |
| 2014.10.20 | 0 | 0 | 94 | 11 | 0 | 0 |
| 2014.10.21 | 15 | 1 | 67 | 8 | 10 | 2 |
| 2014.10.22 | 20 | 5 | 76 | 15 | 1 | 1 |
| 2014.10.23 | 33 | 3 | 12 | 5 | 0 | 0 |
| 2014.10.24 | 29 | 2 | 58 | 11 | 1 | 1 |
| 2014.10.25 | 33 | 8 | 38 | 8 | 1 | 1 |
| 2014.10.26 | 13 | 6 | 39 | 8 | 34 | 3 |
| 2014.10.27 | 13 | 4 | 49 | 12 | 15 | 1 |
| 2014.10.28 | 4 | 2 | 44 | 8 | 3 | 1 |
| 2014.10.29 | 0 | 0 | 28 | 9 | 13 | 2 |
| 2014.10.30 | 4 | 3 | 49 | 8 | 0 | 0 |
| 2014.10.31 | 3 | 2 | 58 | 14 | 7 | 1 |
| 2014.11.1 | 0 | 0 | 71 | 9 | 7 | 2 |
| 2014.11.2 | 23 | 2 | 57 | 6 | 0 | 0 |
| 2014.11.3 | 10 | 1 | 18 | 5 | 0 | 0 |
| 2014.11.4 | 19 | 1 | 49 | 10 | 3 | 1 |
| 2014.11.5 | 29 | 2 | 28 | 9 | 0 | 0 |
| 2014.11.6 | 86 | 3 | 142 | 19 | 0 | 0 |
| 2014.11.7 | 38 | 2 | 4 | 2 | 0 | 0 |
| 2014.11.8 | 0 | 0 | 18 | 4 | 6 | 1 |
| 2014.11.9 | 25 | 2 | 39 | 10 | 0 | 0 |
| 2014.11.10 | 16 | 1 | 17 | 3 | 0 | 0 |
| 2014.11.11 | 0 | 0 | 70 | 16 | 1 | 1 |
| 2014.11.12 | 0 | 0 | 4 | 3 | 0 | 0 |
| 2014.11.13 | 0 | 0 | 168 | 20 | 1 | 1 |
| 2014.11.14 | 0 | 0 | 18 | 2 | 0 | 0 |
| 2014.11.15 | 0 | 0 | 5 | 2 | 8 | 1 |
| 2014.11.16 | 16 | 2 | 16 | 4 | 4 | 1 |
| 2014.11.17 | 0 | 0 | 8 | 3 | 0 | 0 |
| 2014.11.18 | 4 | 1 | 7 | 3 | 0 | 0 |
| 2014.11.19 | 17 | 2 | 4 | 1 | 0 | 0 |
| 2014.11.20 | 0 | 0 | 102 | 13 | 0 | 0 |
| 2014.11.21 | 7 | 1 | 31 | 3 | 1 | 1 |
| 2014.11.22 | 1 | 1 | 17 | 4 | 0 | 0 |
| 2014.11.23 | 4 | 1 | 25 | 6 | 0 | 0 |
| 2014.11.24 | 0 | 0 | 2 | 1 | 3 | 1 |
| 2014.11.25 | 4 | 1 | 0 | 0 | 7 | 2 |
| 2014.11.26 | 0 | 0 | 4 | 1 | 0 | 0 |
| 2014.11.27 | 0 | 0 | 1 | 1 | 6 | 1 |
| 2014.11.28 | 0 | 0 | 6 | 3 | 1 | 1 |
| 2014.11.29 | 1 | 1 | 29 | 4 | 13 | 3 |
| 2014.11.30 | 3 | 1 | 57 | 10 | 15 | 2 |
| 2014.12.1 | 8 | 1 | 15 | 4 | 7 | 3 |
| 2014.12.2 | 8 | 3 | 17 | 5 | 0 | 0 |
| 2014.12.3 | 3 | 1 | 6 | 2 | 0 | 0 |
| 2014.12.4 | 4 | 3 | 1 | 1 | 1 | 1 |
| 2014.12.5 | 0 | 0 | 17 | 2 | 5 | 2 |
| 2014.12.6 | 0 | 0 | 6 | 2 | 3 | 1 |
| 2014.12.7 | 0 | 0 | 7 | 3 | 0 | 0 |
| 2014.12.8 | 11 | 3 | 0 | 0 | 0 | 0 |
| 2014.12.9 | 7 | 1 | 0 | 0 | 0 | 0 |
| 2014.12.10 | 27 | 2 | 0 | 0 | 0 | 0 |
| 2014.12.11 | 0 | 0 | 0 | 0 | 1 | 1 |
| 2014.12.13 | 17 | 3 | 0 | 0 | 0 | 0 |
| 2014.12.14 | 3 | 1 | 10 | 1 | 0 | 0 |
| 2014.12.15 | 25 | 3 | 1 | 1 | 9 | 2 |
| 2014.12.16 | 34 | 3 | 10 | 4 | 0 | 0 |
| 2014.12.17 | 11 | 2 | 3 | 2 | 1 | 1 |
| 2014.12.18 | 3 | 1 | 8 | 1 | 0 | 0 |
| 2014.12.19 | 7 | 1 | 1 | 1 | 9 | 1 |
| 2014.12.20 | 96 | 3 | 11 | 4 | 0 | 0 |
| 2014.12.21 | 1 | 1 | 17 | 4 | 12 | 3 |
| 2014.12.23 | 0 | 0 | 21 | 5 | 1 | 1 |
| 2014.12.24 | 5 | 1 | 11 | 4 | 0 | 0 |
| 2014.12.25 | 14 | 2 | 8 | 2 | 0 | 0 |
| 2014.12.26 | 0 | 0 | 13 | 4 | 0 | 0 |
| 2014.12.27 | 0 | 0 | 9 | 3 | 0 | 0 |
| 2014.12.28 | 0 | 0 | 24 | 4 | 0 | 0 |
| 2014.12.29 | 0 | 0 | 21 | 7 | 0 | 0 |
| 2014.12.30 | 0 | 0 | 34 | 6 | 0 | 0 |
| 2014.12.31 | 1 | 1 | 47 | 5 | 0 | 0 |
| 2015.1.1 | 0 | 0 | 33 | 4 | 0 | 0 |
| 2015.1.2 | 0 | 0 | 29 | 7 | 0 | 0 |
| 2015.1.3 | 0 | 0 | 25 | 4 | 12 | 1 |
| 2015.1.4 | 0 | 0 | 14 | 5 | 0 | 0 |
| 2015.1.5 | 12 | 2 | 0 | 0 | 0 | 0 |
#+tblname: idle_periods_mt
| Start | End |
|------------+------------|
| 2014.9.2 | 2014.9.20 |
| 2014.12.18 | 2014.12.31 |
#+begin_src R :exports both :results output graphics :var daily=daily :var idle=idle_periods_mt :file /tmp/daily.png :width 600 :height 600
library(reshape2)
library(ggplot2)
require(gridExtra)
daily$Date <- as.Date(daily$Date, "%Y.%m.%d")
data_long <- melt(daily, id.vars=c("Date"))
idle$Start <- as.Date(idle$Start, "%Y.%m.%d")
idle$End <- as.Date(idle$End, "%Y.%m.%d")
ymax1=200
p1 <- ggplot() +
geom_area(data=data_long[data_long$variable %in% c("exos_scala","exos_python","exos_java"),],
aes(x=Date, y=value, color=variable,fill=variable)) +
ggtitle("Daily activity (exercises)") +
geom_rect(data=idle,aes(xmin=Start, xmax=End, ymin=0, ymax=ymax1),alpha=.1,fill="red",color="blue") +
theme(legend.justification=c(1,0), legend.position=c(1,.6)) +
coord_cartesian(ylim = c(0,ymax1)) +
ylab("Exercises (#)")
ymax2 = 40
p2 <- ggplot() +
geom_area(data=data_long[data_long$variable %in% c("traces_scala","traces_python","traces_java"),],
aes(x=Date, y=value, color=variable,fill=variable)) +
ggtitle("Daily activity (users)") +
geom_rect(data=idle,aes(xmin=Start, xmax=End, ymin=0, ymax=ymax2),alpha=.1,fill="red",color="blue") +
theme(legend.justification=c(1,0), legend.position=c(1,.6)) +
coord_cartesian(ylim = c(0,ymax2)) + ### zoom with ggplot
ylab("Active Traces (#)")
grid.arrange(p1, p2)
#+end_src
#+RESULTS:
[[file:/tmp/daily.png]]
Entered on [2015-01-07 mer. 16:27]
[[file:/tmp/plm-iticse.org::*Data%20Analysis][Data Analysis]]
** 2015-07 juillet
*** 2015-07-31 vendredi
**** LOESS :WP8:Stats:R:
An involved lecture:
http://web.as.uky.edu/statistics/users/pbreheny/621/F10/notes/11-4.pdf
A few R examples illustrating the influence of bandwidth:
- http://research.stowers-institute.org/efg/R/Statistics/loess.htm
- http://www.duclert.org/Aide-memoire-R/Statistiques/Local-polynomial-fitting.php
Entered on [2015-07-31 ven. 09:35]
**** Harald Servat's Phd defense
Comments:
- I really enjoyed the very *clear presentation* of the document, of the
related work, etc.
- I particularly enjoyed the fact that *hypothesis are clearly stated*,
which I think is the sign of a true *scientific approach* in term of
methodology. I also think that moving to *continous approximations* as
you tried by using Kriging or segmented linear regressions is an
excellent idea.
- Last, *thanks for giving me the opportunity* to think more carefully
about the mathematical foundations of such tools and how they could
make sense or not. It actually raised a lots of questions.
Questions:
- You did not hesitate to use of *elaborate statistical tools*. Such
tools rely on probabilistic model and on of the interesting feature
is that they allow two things:
- Hypothesis testing
- Confidence interval calculation
Do you think it would be worth building on such features
- You demonstrated that your clustering methodology could be applied
to many use cases. Can tell me if you can think about situations
where it would not apply.
- 11: segmented linear regression seems more meaningful than krigging
here. Is it really the case ? Kriging does an interpolation and
maybe the nuggeting is unable to smooth things enough. But maybe the
different phases detected by segmented regression are not that
meaningful either ?
- Reuse of previous analysis to capture better traces with a lower
overhead ?
- 15: multi-dimensionnal segmented linear regression ?
- 28: you are annoyed because time is a random variable too. There is
uncertainty on it, which is why the classical technique (krigging or
segmented regression) do not apply
- Machine learning for pointing out to situations where correlation
make sense or not.
- 39: shouldn't the compiler have been able to do such kind of
optimization ?
- 49: L1-cache based sampling allowed to detect when MPI was receiving
a message
Entered on [2015-07-31 ven. 10:58]
** 2015-12 décembre
*** 2015-12-22 mardi
**** Programmation avec Clément: hexagone magique :Teaching:Python:
***** Génération de permutations toute simple
#+begin_src python :results output :exports both
N = 5
A = range(1,N)
def generate(tab,i):
if i>=len(tab):
print(tab)
else:
for j in range(i,len(tab)):
tab[i],tab[j] = tab[j],tab[i]
generate(tab,i+1)
tab[i],tab[j] = tab[j],tab[i]
generate(A,0)
#+end_src
#+RESULTS:
#+begin_example
[1, 2, 3, 4]
[1, 2, 4, 3]
[1, 3, 2, 4]
[1, 3, 4, 2]
[1, 4, 3, 2]
[1, 4, 2, 3]
[2, 1, 3, 4]
[2, 1, 4, 3]
[2, 3, 1, 4]
[2, 3, 4, 1]
[2, 4, 3, 1]
[2, 4, 1, 3]
[3, 2, 1, 4]
[3, 2, 4, 1]
[3, 1, 2, 4]
[3, 1, 4, 2]
[3, 4, 1, 2]
[3, 4, 2, 1]
[4, 2, 3, 1]
[4, 2, 1, 3]
[4, 3, 2, 1]
[4, 3, 1, 2]
[4, 1, 3, 2]
[4, 1, 2, 3]
#+end_example
***** Exploration comme un bourrin
On représente l'hexagone par un tableau numéroté comme ceci:
#+BEGIN_EXAMPLE
0 1 2
3 4 5 6
7 8 9 10 11
12 13 14 15
17 18 19
#+END_EXAMPLE
#+begin_src python :results output :exports both :tangle /tmp/test_bourrin.py
def check(tab):
start = 0
for r in [3,4,5,4]:
if sum(tab[start:(start+r)])!=38:
return False
start = start + r
for t in [[2,6,11],[1,5,10,15],[0,4,9,14,18],[3,8,13,17]]:
if sum([tab[i] for i in t])!=38:
return False
for t in [[7,3,0],[1,4,8,12],[2,5,9,13,16],[6,10,14,17]]:
if sum([tab[i] for i in t])!=38:
return False
return True
def generate(tab,i):
if i>=len(tab):
if check(tab):
print(tab)
else:
for j in range(i,len(tab)):
tab[i],tab[j] = tab[j],tab[i]
generate(tab,i+1)
tab[i],tab[j] = tab[j],tab[i]
generate([3, 17, 18, 19, 7, 1, 11, 16, 2, 5, 6, 9, 12, 4, 8, 14, 10, 13, 15],0)
#+end_src
Bon, sauf que ça va être monstrueusement long en fait. Sur ma machine:
#+begin_src sh :results output :exports both
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
#+end_src
#+RESULTS:
: 3300000
Donc, dans l'hypothèse, ultra optimiste où je serais capable de
vérifier une permutation par cycle d'horloge, il me faudrait:
#+begin_src R :results output :session :exports both
factorial(19)/3300000/24/3600/365
#+end_src
#+RESULTS:
: [1] 1168.891
Plus de 1000 ans. Bon, la loi de Moore finira par nous aider mais pas
des masses. :)
***** Génération de permutation avec coupe au plus tôt
De façon à éliminer les branches au plus tôt, on représente l'hexagone
par un tableau numéroté comme ceci:
#+BEGIN_EXAMPLE
0 1 2
11 12 13 3
10 17 19 14 4
9 16 15 5
8 7 6
#+END_EXAMPLE
Et on remarque qu'on n'a de choix de branchement que pour 0, 1, 3, 5,
7, 9, et 12. Tous les autres sont induits par les précédents.
#+begin_src python :results output :exports both :tangle /tmp/test_rapide.py
def assign(tab, i, x):
if x in tab[i:len(tab)]:
for j in range(i,len(tab)):
if(tab[j]==x):
tab[i],tab[j] = tab[j],tab[i]
generate(tab,i+1)
tab[i],tab[j] = tab[j],tab[i]
return
def generate(tab,i):
# print(i)
if i>=len(tab):
print(tab)
else:
if i in [0,1,3,5,7,9,12]:
for j in range(i,len(tab)):
tab[i],tab[j] = tab[j],tab[i]
generate(tab,i+1)
tab[i],tab[j] = tab[j],tab[i]
elif i in [2,4,6,8,10]:
x = 38 - (tab[i-1]+tab[i-2])
assign(tab,i,x)
elif i==11:
x = 38 - (tab[i-1]+tab[0])
assign(tab,i,x)
elif i==13:
x = 38 - (tab[11]+tab[12]+tab[3])
assign(tab,i,x)
elif i==14:
x = 38 - (tab[1]+tab[13]+tab[5])
assign(tab,i,x)
elif i==15:
x = 38 - (tab[3]+tab[14]+tab[7])
assign(tab,i,x)
elif i==16:
x = 38 - (tab[5]+tab[15]+tab[9])
assign(tab,i,x)
elif i==17:
x = 38 - (tab[7]+tab[16]+tab[11])
if x+tab[9]+tab[12]+tab[1]!=38:
return
assign(tab,i,x)
elif i==18:
if tab[10]+tab[17]+tab[18]+tab[14]+tab[4]==38 and \
tab[0]+tab[12]+tab[18]+tab[15]+tab[6]==38 and \
tab[2]+tab[13]+tab[18]+tab[16]+tab[8]==38:
generate(tab,i+1)
generate(range(1,20),0)
#+end_src
Et maintenant, combien de temps pour avoir la solution ?
#+begin_src sh :results output :exports both
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
time python /tmp/test_rapide.py 2>&1
#+end_src
#+RESULTS:
#+begin_example
performance
3300000
[3, 17, 18, 11, 9, 14, 15, 13, 10, 12, 16, 19, 7, 1, 6, 8, 4, 2, 5]
[3, 19, 16, 12, 10, 13, 15, 14, 9, 11, 18, 17, 7, 2, 4, 8, 6, 1, 5]
[9, 11, 18, 17, 3, 19, 16, 12, 10, 13, 15, 14, 6, 1, 7, 2, 4, 8, 5]
[9, 14, 15, 13, 10, 12, 16, 19, 3, 17, 18, 11, 6, 8, 4, 2, 7, 1, 5]
[10, 12, 16, 19, 3, 17, 18, 11, 9, 14, 15, 13, 4, 2, 7, 1, 6, 8, 5]
[10, 13, 15, 14, 9, 11, 18, 17, 3, 19, 16, 12, 4, 8, 6, 1, 7, 2, 5]
[15, 13, 10, 12, 16, 19, 3, 17, 18, 11, 9, 14, 8, 4, 2, 7, 1, 6, 5]
[15, 14, 9, 11, 18, 17, 3, 19, 16, 12, 10, 13, 8, 6, 1, 7, 2, 4, 5]
[16, 12, 10, 13, 15, 14, 9, 11, 18, 17, 3, 19, 2, 4, 8, 6, 1, 7, 5]
[16, 19, 3, 17, 18, 11, 9, 14, 15, 13, 10, 12, 2, 7, 1, 6, 8, 4, 5]
[18, 11, 9, 14, 15, 13, 10, 12, 16, 19, 3, 17, 1, 6, 8, 4, 2, 7, 5]
[18, 17, 3, 19, 16, 12, 10, 13, 15, 14, 9, 11, 1, 7, 2, 4, 8, 6, 5]
1.25user 0.00system 0:01.26elapsed 99%CPU (0avgtext+0avgdata 6668maxresident)k
0inputs+0outputs (0major+873minor)pagefaults 0swaps
#+end_example
***** Solution par d'autres personnes
Finalement, un peu de google nous donne ça:
http://codegolf.stackexchange.com/questions/6304/code-solution-for-the-magic-hexagon
La solution en C++ est fondamentalement la même mais sans utilisation
de récursion, i.e. en inlinant les 9 boucles et avec une macro pour
aléger l'écriture. Elle trouve les mêmes solutions que nous mais 60
fois plus vite:
#+begin_src cpp :results output :exports both :tangle /tmp/test_cpp.cpp
#include <stdio.h>
#define LOOP(V) for(int V=1;V<20;V++){if(m&1<<V){m&=~(1<<V);
#define ENDLOOP(V) m|=1<<V;}}
#define SET(V,e) int V=e;if(m&1<<V){m&=~(1<<V);
#define UNSET(V) m|=1<<V;}
int main() {
int m=1048574;
LOOP(A);
LOOP(B);
SET(C,38-A-B);
LOOP(D);
SET(H,38-A-D);
LOOP(G);
SET(L,38-C-G);
LOOP(E);
SET(F,38-D-E-G);
LOOP(I);
SET(M,38-B-E-I);
SET(Q,38-H-M);
LOOP(J);
SET(N,38-C-F-J-Q);
SET(R,38-D-I-N);
SET(S,38-Q-R);
SET(P,38-L-S);
SET(K,38-B-F-P);
SET(O,38-M-N-P);
printf("%d %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d\n",A,B,C,D,E,F,G,H,I,J,K,L,M,N,O,P,Q,R,S);
UNSET(O);
UNSET(K);
UNSET(P);
UNSET(S);
UNSET(R);
UNSET(N);
ENDLOOP(J);
UNSET(Q);
UNSET(M);
ENDLOOP(I);
UNSET(F);
ENDLOOP(E);
UNSET(L);
ENDLOOP(G);
UNSET(H);
ENDLOOP(D);
UNSET(C);
ENDLOOP(B);
ENDLOOP(A);
}
#+end_src
#+begin_src sh :results output :exports both
gcc -O3 /tmp/test_cpp.cpp -o test_cpp
time /tmp/test_cpp 2>&1
#+end_src
#+RESULTS:
#+begin_example
3 17 18 19 7 1 11 16 2 5 6 9 12 4 8 14 10 13 15
3 19 16 17 7 2 12 18 1 5 4 10 11 6 8 13 9 14 15
9 11 18 14 6 1 17 15 8 5 7 3 13 4 2 19 10 12 16
9 14 15 11 6 8 13 18 1 5 4 10 17 7 2 12 3 19 16
10 12 16 13 4 2 19 15 8 5 7 3 14 6 1 17 9 11 18
10 13 15 12 4 8 14 16 2 5 6 9 19 7 1 11 3 17 18
15 13 10 14 8 4 12 9 6 5 2 16 11 1 7 19 18 17 3
15 14 9 13 8 6 11 10 4 5 1 18 12 2 7 17 16 19 3
16 12 10 19 2 4 13 3 7 5 8 15 17 1 6 14 18 11 9
16 19 3 12 2 7 17 10 4 5 1 18 13 8 6 11 15 14 9
18 11 9 17 1 6 14 3 7 5 8 15 19 2 4 13 16 12 10
18 17 3 11 1 7 19 9 6 5 2 16 14 8 4 12 15 13 10
0.02user 0.00system 0:00.02elapsed 100%CPU (0avgtext+0avgdata 1284maxresident)k
0inputs+0outputs (0major+64minor)pagefaults 0swaps
#+end_example
Entered on [2015-12-22 mar. 17:59]
*** 2015-12-23 mercredi
**** Parrot :HOME:
Appel le 23/12/15 vers 17:35. No de dossier: 516 440
* 2016
** 2016-02 février
*** 2016-02-17 mercredi
**** Programmation pendu avec Rémi :Teaching:Python:ATTACH:
:PROPERTIES:
:Attachments: lst.txt
:ID: 0032c718-137f-464c-ab82-5cd3b378a222
:END:
***** Code
#+begin_src python :results output :exports both :tangle /tmp/pendu.py
from random import *
from sys import stdin
def valide(mot):
for l in mot:
ok = 0
for lp in ["a","b","c","d","e","f","g","h","i","j","k","l","m","n","o","p","q","r","s","t","u","v","w","x","y","z","-"]:
if(l==lp): ok=1
if(ok==0): return 0
return 1
def lit_dictionnaire(nom_fichier):
f = open(nom_fichier, 'r')
L = []
for line in f:
mot = line.rstrip()
if(valide(mot)):
L.append(mot)
# print "Il y a " + str(len(L)) + " mots dans mon dictionnaire."
# print "Le premier est '" + L[0] + "'."
# print "Le dixieme est '" + L[9] + "'."
return L
def choisit_mot(dico):
return dico[int(len(dico)*random())]
def trouve(lettre,lettres_autorisees):
for l in lettres_autorisees:
if(lettre==l):
return 1
return 0
def enleve(l,L):
for i in range(0,len(L)):
if(l==L[i]):
return(L[:i]+L[(i+1):])
def lit_lettre(lettres_autorisees):
print(lettres_autorisees)
input = stdin.readline().rstrip()
while (len(input)!=1 or (trouve(input,lettres_autorisees)!=1) ):
print trouve(input,lettres_autorisees)
print "Imbecile! Donne moi UNE lettre et qui soit autorisee!"
input = stdin.readline().rstrip()
return input
def remplace(mot_joueur, l, mot):
# print ">>> ("+mot_joueur+","+l+","+mot+")"
a_trouve = 0
for i in range(0,len(mot)):
if(mot[i] == l):
# print "Youpi!!! j'ai trouve: "+mot[i]
# print " "+mot_joueur
mot_joueur = mot_joueur[:i] + l + mot_joueur[(i+1):]
a_trouve = 1
# print " "+mot_joueur
return (mot_joueur,a_trouve)
def motif_ok(mot,motif):
if(len(mot)!=len(motif)):
return 0
for i in range(0,len(motif)):
if(motif[i]!="#"):
if(mot[i]!=motif[i]):
return 0;
return 1
def lettres_exclues_ok(mot,lettres_exclues):
for l in mot:
for le in lettres_exclues:
if l==le:
return 0
return 1
def filtre(dictionnaire,motif,lettres_exclues):
nouveau_dico = []
for mot in dictionnaire:
if(motif_ok(mot,motif) and lettres_exclues_ok(mot, lettres_exclues)):
nouveau_dico.append(mot)
return nouveau_dico
def conseille_stupide(mots_possibles, lettres_possibles):
return lettres_possibles[0]
def lettre_dans_mot(l,mot):
for lm in mot:
if(lm==l):
return 1
return 0
def conseille(mots_possibles, lettres_possibles):
nombre_mots = len(mots_possibles)
if nombre_mots==1:
for l in lettres_possibles:
if(lettre_dans_mot(l,mots_possibles[0])):
return l
def compte(l,mots_possibles):
num = 0
for mot in mots_possibles:
if lettre_dans_mot(l,mot)==1:
num=num+1
return num
nombre_mots_avec_la_bonne_lettre = []
score = []
for l in lettres_possibles:
num = compte(l,mots_possibles)
nombre_mots_avec_la_bonne_lettre.append(num)
score.append(abs(num-nombre_mots/2.0))
score_min = score[0]+.1
i_min = 0
for i in range(0,len(lettres_possibles)):
if(score[i]<score_min):
score_min = score[i]
i_min = i
### Brisons les cas d'egalite qui nous empecheraient de progresser
if(score_min==(nombre_mots/2.0)):
if(lettre_dans_mot(lettres_possibles[i_min],mots_possibles[0])==0):
score_min +=.1
# print mots_possibles
# print lettres_possibles
# print nombre_mots
# print nombre_mots_avec_la_bonne_lettre
# print score
# print i_min
return lettres_possibles[i_min]
def conseille_freq(mots_possibles, lettres_possibles):
nombre_mots = len(mots_possibles)
if nombre_mots==1:
for l in lettres_possibles:
if(lettre_dans_mot(l,mots_possibles[0])):
return l
def frequence(l,mots_possibles):
num = 0
for mot in mots_possibles:
for lm in mot:
if(lm==l):
num += 1;
return num
frequence_lettre = []
for l in lettres_possibles:
frequence_lettre.append(frequence(l,mots_possibles))
freq_max = 0
i_max = -1
for i in range(0,len(lettres_possibles)):
if(frequence_lettre[i]>freq_max):
freq_max = frequence_lettre[i]
i_max = i
# print lettres_possibles
# print nombre_mots
# print nombre_mots_avec_la_bonne_lettre
# print score
# print i_min
return lettres_possibles[i_max]
def jeu(dictionnaire,mot,mode):
mot_joueur = "#" * len(mot)
for i in range(0,len(mot)):
if mot[i]=="-": mot_joueur = mot_joueur[:i] + "-" + mot_joueur[(i+1):]
lettres_autorisees = ["a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z"]
lettres_exclues = []
max_erreur=18
erreur=0
mots_possibles=dictionnaire
# print mot
while(mot_joueur != mot):
if mode=="interactif": print mot_joueur + " | Nombre d'erreurs autorisees restant : " + str(max_erreur-erreur)
mots_possibles = filtre(mots_possibles,mot_joueur,lettres_exclues)
if mode=="interactif": print "Il reste " + str(len(mots_possibles)) + " mot(s)"
if mode=="interactif": lettre_conseillee = conseille(mots_possibles,lettres_autorisees)
if mode=="frequence": lettre_conseillee = conseille_freq(mots_possibles,lettres_autorisees)
if mode=="dichotomie": lettre_conseillee = conseille(mots_possibles,lettres_autorisees)
if mode=="interactif": print " Conseil: " + lettre_conseillee
if mode=="interactif": lettre = lit_lettre(lettres_autorisees)
else: lettre=lettre_conseillee
(mot_joueur,a_trouve) = remplace(mot_joueur, lettre, mot)
lettres_autorisees = enleve(lettre,lettres_autorisees)
if(a_trouve==0):
erreur += 1
lettres_exclues.append(lettre)
if(erreur==max_erreur):
if mode=="interactif": print "Tu as perdu!!!!"
if mode=="interactif": print "C'etait : " + mot
return erreur
if mode=="interactif": print mot_joueur + " | Nombre d'erreurs autorisees restant : " + str(max_erreur-erreur)
if mode=="interactif": print "Bravo!!!!"
return erreur
def main():
mon_dico = lit_dictionnaire("/home/alegrand/Hacking/boggle/Words.txt");
while(1):
mot = choisit_mot(mon_dico);
jeu(mon_dico,mot,"interactif")
def main2():
mon_dico = lit_dictionnaire("/home/alegrand/Hacking/boggle/Words.txt");
while(1):
mot = choisit_mot(mon_dico);
freq = jeu(mon_dico,mot,"frequence")
dicho = jeu(mon_dico,mot,"dichotomie")
print mot + " , " + str(freq) + " , " + str(dicho)
main2()
### Quelques constructions equivalentes
# i=0
# while(i<10):
# print i
# i=i+1
#
# for i in range(0,10):
# print i
# for i in range(0,len(liste)):
# print liste[i]
#
# for mot in liste:
# print mot
#+end_src
J'exécute et j'arrête au boût de 3 minutes
#+begin_src sh :results output :exports both
python pendu.py > lst.txt
#+end_src
#+begin_src R :results output graphics :file (org-babel-temp-file "figure" ".png") :exports both :width 400 :height 400 :session
library(ggplot2)
df=read.csv("data/00/32c718-137f-464c-ab82-5cd3b378a222/lst.txt",strip.white=T,header=F)
names(df)=c("mot","freq","dicho")
ggplot(data=df,aes(x=freq,y=dicho)) + geom_point(alpha=.3)
#+end_src
#+RESULTS:
[[file:/tmp/babel-9398SP8/figure9398caH.png]]
#+begin_src R :results output graphics :file (org-babel-temp-file "figure" ".png") :exports both :width 400 :height 400 :session
ggplot(data=df,aes(x=freq-dicho)) + geom_histogram()
#+end_src
#+RESULTS:
[[file:/tmp/babel-9398SP8/figure9398QDg.png]]
#+begin_src R :results output :session :exports both
summary(df)
#+end_src
#+RESULTS:
: mot freq dicho
: bionique : 2 Min. :0.000 Min. : 0.000
: crevassait: 2 1st Qu.:0.000 1st Qu.: 1.000
: pin : 2 Median :1.000 Median : 3.000
: primevere : 2 Mean :1.296 Mean : 3.087
: terrien : 2 3rd Qu.:2.000 3rd Qu.: 4.000
: abattement: 1 Max. :8.000 Max. :11.000
: (Other) :836
#+begin_src R :results output :session :exports both
X=df$freq-df$dicho
summary(X)
mean(X)
err = sd(X)/sqrt(length(X))
mean(X) - 2*err
mean(X) + 2*err
#+end_src
#+RESULTS:
: Min. 1st Qu. Median Mean 3rd Qu. Max.
: -10.000 -3.000 -1.000 -1.791 0.000 7.000
: [1] 2.304793
: [1] 0.07919362
***** Links
http://web.stanford.edu/class/cs106l/handouts/assignment-2-evil-hangman.pdf
http://www.sharkfeeder.com/hangman/
http://blog.wolfram.com/2010/08/13/25-best-hangman-words/
Entered on [2016-02-17 mer. 10:14]
*** 2016-02-25 jeudi
**** Hacking screenkey :Python:
#+begin_src sh :session foo :results output :exports both
diff -u /usr/share/pyshared/Screenkey/listenkdb.py_old /usr/share/pyshared/Screenkey/listenkdb.py
#+end_src
#+RESULTS:
#+begin_example
--- /usr/share/pyshared/Screenkey/listenkdb.py_old 2016-03-07 09:40:13.271193249 +0100
+++ /usr/share/pyshared/Screenkey/listenkdb.py 2016-03-07 09:42:41.216862924 +0100
@@ -230,7 +230,14 @@
mod = mod + "Alt+"
if self.cmd_keys['super']:
mod = mod + "Super+"
-
+
+ if self.cmd_keys['shift']:
+ if (len(key_shift)==1) and not(ord(key_normal) in range(97,123)) and not(ord(key_shift) in range(33,126)):
+ mod = mod + "Shift+"
65000):
+ mod = mod + "Shift+"
print "---------"
print key, key_shift, keysym
if self.cmd_keys['shift']:
key = key_shift
if self.cmd_keys['capslock'] \
#+end_example
Entered on [2016-02-25 jeu. 11:19]
** 2016-07 juillet
*** 2016-07-19 mardi
**** [[http://rmarkdown.rstudio.com/flexdashboard/][flexdashboard: Easy interactive dashboards for R]] :WP7:WP8:R:twitter:
Entered on [2016-07-19 mar. 09:04]
**** Steps toward reproducible research (Karl Broman) :WP8:twitter:R:
- https://github.com/kbroman/Talk_ReproRes (slides)
- Why
- I'm sorry but I think you haven't used the right data.
- The results in Table 1 don’t seem to correspond to those in
Figure 2.
- In what order do I run these scripts?
- Where did we get this data file?
- Why did I omit those samples?
- How did I make that figure?
- Important points
- Organize your data & code
#+BEGIN_QUOTE
Your closest collaborator is you six months ago,
but you don’t reply to emails.
(paraphrasing Mark Holder)
#+END_QUOTE
- Everything with a script
If you do something once, you’ll do it 1000 times.
- Automate the process as much as you can
In addition to automating a complex process, it also documents
the process, including the dependencies among data files and
scripts.
- Turn scripts into reproducible reports
- Use version control (git/GitHub)
#+BEGIN_QUOTE
The most important tool is the mindset,
when starting, that the end product
will be reproducible.
– Keith Baggerly
#+END_QUOTE
- [[http://kbroman.org/steps2rr/][initial steps toward reproducible research]] (lecture)
- https://github.com/kbroman/Tools4RR (Materials for a one-credit
course on reproducible research)
Entered on [2016-07-19 mar. 09:06]
**** [[http://michaellevy.name/blog/teaching-r-to-200-students-in-a-week/][Teaching R to 200 students in a week • Michael Levy]] :Teaching:R:
Link from Michael Blum
- Motivation precedes detail: “Here’s what you’re going to learn to do this week”
- Live coding: shows that I make mistakes, builds in flexibility, forces to slow down
- Live code piped to their browsers (dropbox? pad? floobits?)
- In-class exercises instead of lectures
- Stickies and good assistants
- Daily feedback: Each day, I asked the students to fill out a quick
survey: How well do you understand what was taught today, what’s
working for you, and what could use a change?
- Advanced exercises for experts.
Entered on [2016-07-19 mar. 09:36]
CEDEX 9
* 2017
** 2017-01 janvier
*** 2017-01-09 lundi
**** Ridge regression :Stats:R:
http://web.as.uky.edu/statistics/users/pbreheny/764-F11/notes/9-1.pdf
Ridge regression penalizes the size of the regression coefficients,
which is convenient in the presence of multicollinearity
- http://www.few.vu.nl/~wvanwie/Courses/HighdimensionalDataAnalysis/WNvanWieringen_HDDA_Lecture4_RidgeRegression_20162017.pdf
- https://arxiv.org/pdf/1509.09169.pdf
Entered on[2017-01-09 lun. 22:10]
*** 2017-01-19 jeudi
**** Groupe de lecture: [[file:public_html/readings/Jackson_Networks.pdf][Social and Economic Networks]] (session 1) :POLARIS:
***** Chapter 1: Introduction
- Example from the Medici graph
- degree can be seen as a measure of the influence of the node but a
probably more interesting notion is the betweenness which indicates
the probability the path a pair of peole have to go through you to
communicate:
b(k) = 1/((n-1)(n-2)/2) \sum_{i\neq j} (number of shortest paths from i to j going through
k)/(number ofr shortest paths from i to j)
***** Chapter 2: Basic notions
- notion of adjacency matrix with convenient properties.
- deg(i) = \sum_j g_{i,j}
- #triangles = tr(g^3)/6
- Clustering: "how close are you from a clique"
- local notion of connectivity
- Cl_i(G) = (#of pairs j,k connected to i s.t. j and k are also
connected )/(d(i).(d(i)-1)/2)
= \sum_{j,k s.t. i\ne j\ne k} g_{i,j} g_{j,k} g_{k,i} / \sum_{j,k s.t. i\ne j\ne k} g_{i,j} g_{i,k}
\approx (sum ith diag element of g^3) / (sum ith diag element of g^2)
- Cl(G) = 1/N \sum_i Cl_i(G) is then the /average clustering/ of the graph
if you average over nodes. If one consider a "star" of small
"cliques", all the nodes but the one in the center will have a
local clique of 1 (hence the average clustering tends to 1)
whereas it is far of being a clique itself.
- You could average over triples directly and get the /overall
clustering/ by considering directly:
\sum_{i\ne j\ne k} g_{i,j} g_{j,k} g_{k,i} / \sum_{i\ne j\ne k} g_{i,j} g_{i,k}
This may be a better notion as the overall clustering would go to
0... But we could totally have the reverse...
- Centrality:
- degree centrality is the average degree
- betweeness centrality is what we saw earlier
- Eigen centrality. Relates to a notion of prestige P:
P_i (g) = \sum_{j\ne i} g_{i,j} P_j(g)/\delta_g(j)
Hence P = \bar{g}.P with \bar{g} being the normalized graph (that
could be different if the graph is undirected or if one does not
consider \delta to be the degree d). Note that the Degree satisfies
such equation, hence you recove the degree centrality notion.
- In general, one may consider the larger eigenvalue of \bar{g} and
the Perron Frobenius tells that all components of the
corresponding eigenvector are positive. One may thus normalize by
the sum of the elements of the eigenvector P.
- Another notion is distance or decay centrality:
- \sum_j 1/d(i,j
- We don't really know whether there is a link between
Eigen-centrality and Betweenness centrality.
**** MOOC, Jean-Marc Hasenfratz :Vulgarization:Teaching:WP8:
Discussions sur l'opportunité de faire un MOOC Recherche reproductible
sur FUN. On fait le tour des choses existantes:
- Il y a des choses en stats, sur R
https://www.fun-mooc.fr/cours/#filter/subject/mathematiques-et-statistiques?page=1&rpp=50
mais rien sur les aspects recherche reproductible ou la
programmation littérale en tant que tel.
- Sur coursera, il y a
https://www.coursera.org/specializations/jhu-data-science. Il me
semble que les 5 premiers (intro/techno, R basis, data cleaning,
EDA, litterate programming) valent le coup mais c'est trop long.
- Lorena Barba's tutorial on RR ?
https://barbagroup.github.io/essential_skills_RRC/
- Si on reste sur l'idée que le "cahier de laboratoire" c'est la base,
il faut décider de quelle techno on pousse: knitr, jupyter ou
org-mode. D'où une discussion prévue avec Konrad et Christophe, ce
qui permettrait en plus de donnée une dimension pluri-disciplinaire
intéressante.
Entered on [2017-01-19 jeu. 14:29]
** 2017-12 décembre
*** 2017-12-07 jeudi
**** Refs Recherche reproductible intéressantes :WP8:
[[https://www.jove.com/blog/2017/10/27/reproducibility-librarian-yes-that-should-be-your-next-job/][Reproducible research librarian]]
- [[http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0038234][The Effects of FreeSurfer Version, Workstation Type, and Macintosh
Operating System Version on Anatomical Volume and Cortical Thickness
Measurements]]:
- No differences were detected between repeated single runs nor
between single runs and parallel runs on the same workstation and
for the same FreeSurfer and OS version. For the same OS version,
all Mac workstations produced identical results. However,
differences were revealed between:
- Mac and HP workstations
- FreeSurfer versions v4.3.1, v4.5.0, and v5.0.0
- OSX 10.5.8 and OSX 10.6.4/5
- Focus sur l'importance des erreurs mais pas sur leur origine car
tout est très closed-source.
- [[https://genomebiology.biomedcentral.com/articles/10.1186/s13059-016-1044-7][Gene name errors are widespread in the scientific literature]]
- [[https://genomebiology.biomedcentral.com/articles/10.1186/s13059-015-0850-7][Five selfish reasons to work reproducibly]]: style sympa
- Jupyter extension with Reprozip: https://www.youtube.com/watch?v=Y8YmGVYHhS8
- [[https://github.com/pantsbuild/pex][Pex]]: un environnement permettant de transformer un script python en
un exécutable autonome (packing interne des libs).
- [[http://o2r.info/][O2R]]: hyper préliminaire, aucun intérêt.
Entered on [2017-12-07 jeu. 11:09]
[[file:~/org/journal.org::*Autotuning%20context:][Autotuning context:]]
**** [[file:~/Archives/Cours/maths/R/Verzani-SimpleR.pdf][simpleR – Using R for Introductory StatisticsR]] :Stats:R:
Un document que je n'avais pas pris le temps de lire mais qui est pas
mal fichu du tout. Points intéressants sur R:
- Trouver des outliers en interagissant avec les plots:
=identify(BUSH,BUCHANAN,n=2)=.
- Using rlm or lqs for resistant regression. rlm minimise la somme
d'une fraction des résidus et pas de l'intégralité.
- pas mal d'exemples/exos sur le test (paired, avec même variance ou
pas, Cox-Wilkinson, chi2, etc.)
Entered on [2017-12-07 jeu. 12:28]
[[file:~/org/journal.org::*Refs%20Recherche%20reproductible%20int%C3%A9ressantes][Refs Recherche reproductible intéressantes]]
* 2018
** 2018-10 octobre
*** 2018-10-02 mardi
**** Learning emacs lisp
https://www.gnu.org/software/emacs/manual/html_mono/elisp.html
file:///home/alegrand/tmp/Programming%20in%20Emacs%20Lisp%20-%20https:_www.gnu.org_software_emacs_manual_html_mono_eintr.html
Also eval some emacs lisp using M-:
***** Evaluation
#+begin_src emacs-lisp
(+ 2 2)
#+end_src
#+RESULTS:
: 4
#+begin_src emacs-lisp
'(this is a quoted list) ;; a list
#+end_src
#+begin_src emacs-lisp :
(this is a quoted list) ;; won't work as it will call "this" with args "is" "a" "quoted" "list"
#+end_src
***** Setting variables
#+begin_src emacs-lisp
(setq toto 2)
toto
#+end_src
#+RESULTS:
: 2
#+begin_src emacs-lisp
(setq toto 2)
(set 'toto' 2)
toto
#+end_src
#+RESULTS:
: 2
#+begin_src emacs-lisp
(setq counter 0) ; Let's call this the initializer.
(setq counter (+ counter 1)) ; This is the incrementer.
counter ; This is the counter.
#+end_src
#+RESULTS:
: 1
#+begin_src emacs-lisp
(let ((var1 2)
(var2 3))
(+ var1 var2))
#+end_src
#+RESULTS:
: 5
#+begin_src emacs-lisp
(let ((zebra "stripes")
(tiger "fierce"))
(message "One kind of animal has %s and another is %s."
zebra tiger))
#+end_src
#+RESULTS:
: One kind of animal has stripes and another is fierce.
***** Defining and calling function
#+begin_src emacs-lisp
(defun tutu() 2)
(defun tutu() '(2))
(tutu)
#+end_src
#+RESULTS:
| 2 |
#+begin_src emacs-lisp
(functionp 'tutu)
#+end_src
#+RESULTS:
: t
***** Testing
#+begin_src emacs-lisp
(if (functionp 'tutu) (message "this is a function"))
#+end_src
#+RESULTS:
: this is a function
#+begin_src emacs-lisp
(defun type-of-animal (characteristic)
"Print message in echo area depending on CHARACTERISTIC.
If the CHARACTERISTIC is the string \"fierce\",
then warn of a tiger."
(if (equal characteristic "fierce")
(message "It is a tiger!")))
(type-of-animal "fierce")
;; (type-of-animal "striped")
#+end_src
#+RESULTS:
: It is a tiger!
#+begin_src emacs-lisp
(if (> 4 5) ; if-part
(message "4 falsely greater than 5!") ; then-part
(message "4 is not greater than 5!")) ; else-part
#+end_src
#+RESULTS:
: 4 is not greater than 5!
***** Useful functions
#+begin_src emacs-lisp
;; describe-function
;; describe-key
;; list-matching-lines
;; delete-window
;; point-to-register
;; eval-expression
;; car, cdr, cons
#+end_src
***** Playing with babel templates (1)
#+begin_src emacs-lisp
;; (add-to-list 'org-structure-template-alist
;; '("Y" "#+begin_src R\n?\n#+end_src"))
;; (add-to-list 'org-structure-template-alist
;; '("Y" '(tutu)))
(setq tata "2")
(add-to-list 'org-structure-template-alist
'("Y" tata))
#+end_src
#+RESULTS:
| Y | tata |
| Y | toto |
| Y | (quote (tutu)) |
| Y | (tutu) |
| Y | tutu |
| Y | #+begin_src R |
#+begin_src emacs-lisp
(setq a (assoc "Y" org-structure-template-alist))
a
#+end_src
#+RESULTS:
| Y | tata |
Unfortunately, when expending, the code checks whether the right value
is a string (through th stringp function).
[[file:~/Work/org-mode/lisp/org.el::(defun%20org-try-structure-completion%20()][This is where org-structure-template-alist is used in org-mode 9.0.5's code]].
#+begin_src emacs-lisp
(defun org-try-structure-completion ()
"Try to complete a structure template before point.
This looks for strings like \"<e\" on an otherwise empty line and
expands them."
(let ((l (buffer-substring (point-at-bol) (point)))
a)
(when (and (looking-at "[ \t]*$")
(string-match "^[ \t]*<\\([a-zA-Z]+\\)$" l)
(setq a (assoc (match-string 1 l) org-structure-template-alist)))
(org-complete-expand-structure-template (+ -1 (point-at-bol)
(match-beginning 1)) a)
t)))
(defun org-complete-expand-structure-template (start cell)
"Expand a structure template."
(let ((rpl (nth 1 cell))
(ind ""))
(delete-region start (point))
(when (string-match "\\`[ \t]*#\\+" rpl)
(cond
((bolp))
((not (string-match "\\S-" (buffer-substring (point-at-bol) (point))))
(setq ind (buffer-substring (point-at-bol) (point))))
(t (newline))))
(setq start (point))
(when (string-match "%file" rpl)
(setq rpl (replace-match
(concat
"\""
(save-match-data
(abbreviate-file-name (read-file-name "Include file: ")))
"\"")
t t rpl)))
(setq rpl (mapconcat 'identity (split-string rpl "\n")
(concat "\n" ind)))
(insert rpl)
(when (re-search-backward "\\?" start t) (delete-char 1))))
#+end_src
#+begin_src emacs-lisp
(defun org-try-structure-completion ()
"Try to complete a structure template before point.
This looks for strings like \"<e\" on an otherwise empty line and
expands them."
(let ((l (buffer-substring (point-at-bol) (point)))
a)
(when (and (looking-at "[ \t]*$")
(string-match "^[ \t]*<\\([a-zA-Z]+\\)$" l)
(setq a (assoc (match-string 1 l) org-structure-template-alist)))
(org-complete-expand-structure-template (+ -1 (point-at-bol)
(match-beginning 1)) a)
t)))
(defun org-complete-expand-structure-template (start cell)
"Expand a structure template."
(let ((rpl (nth 1 cell))
(ind ""))
;;; TODO: Here I could check whether rpl is a function and if so, evaluate it
(delete-region start (point))
(when (string-match "\\`[ \t]*#\\+" rpl)
(cond
((bolp))
((not (string-match "\\S-" (buffer-substring (point-at-bol) (point))))
(setq ind (buffer-substring (point-at-bol) (point))))
(t (newline))))
(setq start (point))
(when (string-match "%file" rpl)
(setq rpl (replace-match
(concat
"\""
(save-match-data
(abbreviate-file-name (read-file-name "Include file: ")))
"\"")
t t rpl)))
(setq rpl (mapconcat 'identity (split-string rpl "\n")
(concat "\n" ind)))
(insert rpl)
(when (re-search-backward "\\?" start t) (delete-char 1))))
#+end_src
***** Playing with babel templates (2)
#+begin_src emacs-lisp
(defun myfun() (concat "figure-" (buffer-name) "-line" (what-line) ".png"))
(add-to-list 'org-structure-template-alist
'("Y" "#+begin_src R :results output graphics :file (myfun) :exports both :width 600 :height 400 :session *R*\n?\n#+end_src"))
#+end_src
#+RESULTS:
| Y | #+begin_src R :results output graphics :file (myfun) :exports both :width 600 :height 400 :session *R* |
#+begin_src R :results output graphics :file (myfun) :exports both :width 600 :height 400 :session *R*
plot(cars)
#+end_src
#+RESULTS:
[[file:figure-journal.org-lineLine 76657.png]]
Entered on [2018-10-02 mar. 06:17]
*** 2018-10-24 mercredi
**** Logistic regression and confidence intervals :Stats:R:
https://stackoverflow.com/questions/14423325/confidence-intervals-for-predictions-from-logistic-regression
https://stackoverflow.com/questions/32464641/confidence-intervals-for-logistic-fit-in-seaborn
Entered on [2018-10-24 mer. 10:55]
[[file:~/org/info.org.gpg::*Contact][Contact]]
** 2018-11 novembre
*** 2018-11-06 mardi
**** The importance of stupidity in scientific research :twitter:
[[http://jcs.biologists.org/content/121/11/1771][The importance of stupidity in scientific research | Journal of Cell Science]]
For almost all of us, one of the reasons that we liked science in high
school and college is that we were good at it. That can't be the only
reason – fascination with understanding the physical world and an
emotional need to discover new things has to enter into it too. But
high-school and college science means taking courses, and doing well
in courses means getting the right answers on tests. If you know those
answers, you do well and get to feel smart.
A Ph.D., in which you have to do a research project, is a whole
different thing. For me, it was a daunting task. How could I possibly
frame the questions that would lead to significant discoveries; design
and interpret an experiment so that the conclusions were absolutely
convincing; foresee difficulties and see ways around them, or, failing
that, solve them when they occurred?
[..]
I was a third-year graduate student and I figured that Taube knew
about 1000 times more than I did (conservative estimate). If he didn't
have the answer, nobody did.
That's when it hit me: nobody did. That's why it was a research
problem. And being my research problem, it was up to me to solve.
Entered on [2018-11-06 mar. 13:21]
**** Blog de David Monniaux : malhonnêteté scientifique en informatique :WP8:Emistemology:twitter:
[[http://david.monniaux.free.fr/dotclear/index.php/post/2018/11/01/De-la-malhonn%25C3%25AAtet%25C3%25A9-scientifique-en-informatique][De la malhonnêteté scientifique en informatique - La vie est mal
configurée]]
Entered on [2018-11-06 mar. 13:33]
**** The Inspection Paradox is Everywhere :twitter:Stats:
[[http://allendowney.blogspot.com/2015/08/the-inspection-paradox-is-everywhere.html][Probably Overthinking It: The Inspection Paradox is Everywhere]]
- My favorite example (true in many European countries): most families have a single child, but most kids have siblings.
- Another very obvious one is the speed of cars you pass on the
highway. You always feel people are driving either much faster or
much slower than you and you can't tell how many cars drive at the
same speed as you.
Now moved here: https://www.allendowney.com/blog/
Entered on [2018-11-06 mar. 13:35]
**** Efficient tuning of online systems using Bayesian optimization (Facebook) :WP8:twitter:Stats:
[[https://research.fb.com/efficient-tuning-of-online-systems-using-bayesian-optimization/][Efficient tuning of online systems using Bayesian optimization – Facebook Research]]
- Bayesian optimization for A/B testing, mix of bandit, and
gaussian process/kriging/DoE.
Entered on [2018-11-06 mar. 13:46]
**** Re-Thinking Reproducibility as a Criterion for Research Quality :WP8:Emistemology:twitter:
[[http://philsci-archive.pitt.edu/14352/][Re-Thinking Reproducibility as a Criterion for Research Quality -
Philsci-Archive]]
Discussion sur la notion de reproducibilité, sa faisabilité, sa
portée, son intérêt, selon les différents domaines. La transparence
est plus importante que la reproductibilité "pure" qui ne doit pas
être normative.
Rien de particulièrement nouveau mais c'est bien écrit et
pas dogmatique. Ça change... http://philsci-archive.pitt.edu/14352/
Entered on [2018-11-06 mar. 13:49]
**** How should novelty be valued in science? :WP8:Emistemology:twitter:
[[https://elifesciences.org/articles/28699][Point of View: How should novelty be valued in science? | eLife]]
Pladoyer après une claque d'un reviewer qui trouvait que les travaux
n'étaient pas assez novateurs.
Scientists are under increasing pressure to do "novel" research. Here
I explore whether there are risks to overemphasizing novelty when
deciding what constitutes good science. I review studies from the
philosophy of science to help understand how important an explicit
emphasis on novelty might be for scientific progress. I also review
studies from the sociology of science to anticipate how emphasizing
novelty might impact the structure and function of the scientific
community. I conclude that placing too much value on novelty could
have counterproductive effects on both the rate of progress in science
and the organization of the scientific community. I finish by
recommending that our current emphasis on novelty be replaced by a
renewed emphasis on predictive power as a characteristic of good
science.
Entered on [2018-11-06 mar. 14:04]
**** How statistics are twisted to obscure public understanding :twitter:Stats:Teaching:
[[https://aeon.co/ideas/how-statistics-are-twisted-to-obscure-public-understanding][How statistics are twisted to obscure public understanding | Aeon Ideas]]
Statistics are often used to support points that aren’t true, but
we tend to attack only the data that conflict with some preexisting
notion of our own. The numbers themselves – unless purposefully
falsified – cannot lie, but they can be used to misrepresent the
public statements and ranking systems we take seriously. Statistical
data do not allow for lies so much as semantic manipulation: numbers
drive the misuse of words. When you are told a fact, you must question
how the terms within the fact are defined, and how the data have been
generated. When you read a statistic, of any kind, be sure to ask how
– and more importantly, why – the statistic was generated, whom it
benefits, and whether it can be trusted.
Entered on [2018-11-06 mar. 14:08]
**** What it was like to be peer reviewed in the 1860s :Emistemology:twitter:
[[https://physicstoday.scitation.org/do/10.1063/PT.5.9098/full/][What it was like to be peer reviewed in the 1860s]]
Rather than relying on anonymous referee reports to improve their
papers, authors engaged in extensive personal exchanges with their
reviewers. Such a collegial approach gradually lost favor but recently
has undergone something of a resurgence.
Those recent developments demonstrate that although peer review is a
core component of scientific publishing, its form can change and has
changed to adapt to the ever-evolving needs of the scientific
community.
For a complete history of peer review in the sciences, see “In
referees we trust?” (Physics Today, February 2017, page 44).
Entered on [2018-11-06 mar. 14:12]
**** Statistique & Machine Learning: de Statisticien à Data Scientist :twitter:Stats:Teaching:
[[http://wikistat.fr/][wikistat.fr]]
Statistique niveau L, M1, M2:
- description, inférence, ACP, régression linéaire, logistique,
loglinéaire, ...
Un peu sec mais à explorer.
Entered on [2018-11-06 mar. 14:17]
**** American Scientist: The Science of Scientific Writing :twitter:
Article sur la réécriture/clarification de paragraphes.
[[https://www.americanscientist.org/blog/the-long-view/the-science-of-scientific-writing][The Science of Scientific Writing | American Scientist]]
Entered on [2018-11-06 mar. 14:22]
**** Open peer review: A randomised controlled trial :WP8:twitter:
https://doi.org/10.1192/bjp.176.1.47
Published online: 02 January 2018
***** Background
Most scientific journals practise anonymous peer review. There is no
evidence, however, that this is any better than an open system.
***** Aims
To evaluate the feasibility of an open peer review system.
***** Method
Reviewers for the British Journal of Psychiatry were asked whether
they would agree to have their name revealed to the authors whose
papers they review; 408 manuscripts assigned to reviewers who agreed
were randomised to signed or unsigned groups. We measured review
quality, tone, recommendation for publication and time taken to
complete each review. Results
A total of 245 reviewers (76%) agreed to sign. Signed reviews were of
higher quality, were more courteous and took longer to complete than
unsigned reviews. Reviewers who signed were more likely to recommend
publication.
***** Conclusions
This study supports the feasibility of an open peer review system and
identifies such a system's potential drawbacks.
Entered on [2018-11-06 mar. 14:25]
[[https://www.cambridge.org/core/journals/the-british-journal-of-psychiatry/article/open-peer-review-a-randomised-controlled-trial/1F81447FC67B3BAFDCCCCE82B6C7A187][Open peer review: A randomised controlled trial | The British Journal of Psychiatry | Cambridge Core]]
**** 30 best practices for software development and testing :twitter:
[[https://opensource.com/article/17/5/30-best-practices-software-development-and-testing][30 best practices for software development and testing |
Opensource.com]]
Entered on [2018-11-06 mar. 14:30]
module2/ressources/video_examples/labbook_several.org deleted 100755 → 0
View file @ fcc23c8c
This source diff could not be displayed because it is too large. You can view the blob instead.
module2/ressources/video_examples/labbook_single.org deleted 100755 → 0
View file @ fcc23c8c
This source diff could not be displayed because it is too large. You can view the blob instead.
module2/ressources/video_examples/paper.org deleted 100755 → 0
View file @ fcc23c8c
This source diff could not be displayed because it is too large. You can view the blob instead.
module2/ressources/video_examples/technical_report.org deleted 100755 → 0
View file @ fcc23c8c
# -*- coding: utf-8 -*-
# -*- mode: org -*-
#+TITLE: A reproducible comparison between @@latex:\\@@ GNU MPFR and machine double-precision
#+AUTHOR: Paul Zimmermann (reproduction with org-mode by Arnaud Legrand)
#+STARTUP: overview indent inlineimages logdrawer
#+LANGUAGE: en
#+LATEX_CLASS: IEEEtran
#+LaTeX_CLASS_OPTIONS: [onecolumn]
# #+HTML_HEAD: <link rel="stylesheet" title="Standard" href="http://orgmode.org/worg/style/worg.css" type="text/css" />
#+HTML_HEAD: <link rel="stylesheet" type="text/css" href="http://www.pirilampo.org/styles/readtheorg/css/htmlize.css"/>
#+HTML_HEAD: <link rel="stylesheet" type="text/css" href="http://www.pirilampo.org/styles/readtheorg/css/readtheorg.css"/>
#+HTML_HEAD: <script src="https://ajax.googleapis.com/ajax/libs/jquery/2.1.3/jquery.min.js"></script>
#+HTML_HEAD: <script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.4/js/bootstrap.min.js"></script>
#+HTML_HEAD: <script type="text/javascript" src="http://www.pirilampo.org/styles/lib/js/jquery.stickytableheaders.js"></script>
#+HTML_HEAD: <script type="text/javascript" src="http://www.pirilampo.org/styles/readtheorg/js/readtheorg.js"></script>
#+PROPERTY: header-args :eval never-export
Several authors claim that GNU MPFR [1] is $x$ times slower than
double-precision floating-point numbers, for various values of $x$,
without any way for the reader to reproduce their claim. For example
in [2], Joris van der Hoeven writes “the MPFR library for arbitrary
precision and IEEE-style standardized floating-point arithmetic is
typically about a factor 100 slower than double precision machine
arithmetic”. Such a claim typically: (i) does not say which version of
MPFR was used (and which version of GMP, since MPFR being based on
GMP, its efficiency also depends on GMP); (ii) does not detail the
environment used (processor, compiler, operating system); (iii) does
not explain which application was used for the comparison. Therefore
it cannot be reproduced by the reader, which could thus have no
confidence in the claimed factor of 100. In this short note we provide
reproducible figures that can be checked by the reader.
** Reproducible Experimental Setup
We use the programs in appendix to multiply two $1000 × 1000$
matrices. The matrix $A$ has coefficients $1/(i + j + 1)$ for $0 ≤ i,
j < 1000$, and matrix $b$ has coefficients $1/(ij + 1)$. Both programs
print the time for the matrix product (not counting the time to
initialize the matrix), and the sum of coefficients of the product
matrix (used as a simple checksum between both programs).
We used MFPR version 3.1.5, configured with GMP 6.1.2 (both are the
latest releases as of the date of this document).
We used as test processor =gcc12.fsffrance.org=, which is a machine from
the GCC Compile Farm, a set of machines available for developers of
free software. The compiler used was GCC 4.5.1, which is installed in
~/opt/cfarm/release/4.5.1~ on this machine, with optimization level
~-O3~. Both GMP and MPFR were also compiled with this compiler, and the
GMP and MPFR libraries were linked statically with the application
programs (given in appendix).
** Experimental Results From Arnaud Legrand
*** Code
The program (=a.c=) using the C double-precision type is the
following. It takes as command-line argument the matrix dimension.
#+BEGIN_SRC C :tangle /tmp/a.c
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/resource.h>
static int cputime()
{
struct rusage rus;
getrusage(0, &rus);
return rus.ru_utime.tv_sec * 1000 + rus.ru_utime.tv_usec / 1000;
}
int main(int argc, char *argv[])
{
double **a;
double **b;
double **c;
double t = 0.0;
int i, j, k, st;
int N = atoi(argv[1]);
st = cputime();
a = malloc(N * sizeof(double *));
b = malloc(N * sizeof(double *));
c = malloc(N * sizeof(double *));
for (i = 0; i < N; i++) {
a[i] = malloc(N * sizeof(double));
b[i] = malloc(N * sizeof(double));
c[i] = malloc(N * sizeof(double));
for (j = 0; j < N; j++) {
a[i][j] = 1.0 / (1.0 + i + j);
b[i][j] = 1.0 / (1.0 + i * j);
}
}
st = cputime();
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
c[i][j] = 0.0;
for (i = 0; i < N; i++)
for (k = 0; k < N; k++)
for (j = 0; j < N; j++)
c[i][j] += a[i][k] * b[k][j];
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
t += c[i][j];
printf("matrix product took %dms\n", cputime() - st);
printf("t=%f\n", t);
for (i = 0; i < N; i++) {
free(a[i]);
free(b[i]);
free(c[i]);
}
free(a);
free(b);
free(c);
return 0;
}
#+END_SRC
The program (=d.c=) using GNU MPFR is the following. It takes as
command-line argument the matrix dimension and the MPFR precision (in
bits).
#+BEGIN_SRC C :tangle /tmp/d.c
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/resource.h>
#include <mpfr.h>
static int cputime()
{
struct rusage rus;
getrusage(0, &rus);
return rus.ru_utime.tv_sec * 1000 + rus.ru_utime.tv_usec / 1000;
}
int main(int argc, char *argv[])
{
mpfr_t **a;
mpfr_t **b;
mpfr_t **c;
mpfr_t s;
double t = 0.0;
int i, j, k, st;
int N = atoi(argv[1]);
int prec = atoi(argv[2]);
printf("MPFR library: %-12s\nMPFR header: %s (based on %d.%d.%d)\n",
mpfr_get_version(), MPFR_VERSION_STRING, MPFR_VERSION_MAJOR,
MPFR_VERSION_MINOR, MPFR_VERSION_PATCHLEVEL);
st = cputime();
a = malloc(N * sizeof(mpfr_t *));
b = malloc(N * sizeof(mpfr_t *));
c = malloc(N * sizeof(mpfr_t *));
mpfr_init2(s, prec);
for (i = 0; i < N; i++) {
a[i] = malloc(N * sizeof(mpfr_t));
b[i] = malloc(N * sizeof(mpfr_t));
c[i] = malloc(N * sizeof(mpfr_t));
for (j = 0; j < N; j++) {
mpfr_init2(a[i][j], prec);
mpfr_init2(b[i][j], prec);
mpfr_init2(c[i][j], prec);
mpfr_set_ui(a[i][j], 1, MPFR_RNDN);
mpfr_div_ui(a[i][j], a[i][j], i + j + 1, MPFR_RNDN);
mpfr_set_ui(b[i][j], 1, MPFR_RNDN);
mpfr_div_ui(b[i][j], b[i][j], i * j + 1, MPFR_RNDN);
}
}
st = cputime();
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
mpfr_set_ui(c[i][j], 0, MPFR_RNDN);
for (i = 0; i < N; i++)
for (k = 0; k < N; k++)
for (j = 0; j < N; j++) {
mpfr_mul(s, a[i][k], b[k][j], MPFR_RNDN);
mpfr_add(c[i][j], c[i][j], s, MPFR_RNDN);
}
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
t += mpfr_get_d(c[i][j], MPFR_RNDN);
printf("matrix product took %dms\n", cputime() - st);
printf("t=%f\n", t);
for (i = 0; i < N; i++) {
for (j = 0; j < N; j++) {
mpfr_clear(a[i][j]);
mpfr_clear(b[i][j]);
mpfr_clear(c[i][j]);
}
free(a[i]);
free(b[i]);
free(c[i]);
}
mpfr_clear(s);
free(a);
free(b);
free(c);
return 0;
}
#+END_SRC
*** Setup
- Name of the machine and OS version:
#+begin_src shell :results output :exports results :tangle get_info.sh
uname -a
#+end_src
#+RESULTS:
: Linux sama 4.2.0-1-amd64 #1 SMP Debian 4.2.6-1 (2015-11-10) x86_64 GNU/Linux
- CPU/architecture information:
#+begin_src shell :results output :exports both :tangle get_info.sh
cat /proc/cpuinfo
#+end_src
#+RESULTS:
#+begin_example
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3687U CPU @ 2.10GHz
stepping : 9
microcode : 0x15
cpu MHz : 2165.617
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 2
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveopt
bugs :
bogomips : 5182.68
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
processor : 1
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3687U CPU @ 2.10GHz
stepping : 9
microcode : 0x15
cpu MHz : 3140.515
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 1
cpu cores : 2
apicid : 2
initial apicid : 2
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveopt
bugs :
bogomips : 5182.68
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3687U CPU @ 2.10GHz
stepping : 9
microcode : 0x15
cpu MHz : 2860.000
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 2
apicid : 1
initial apicid : 1
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveopt
bugs :
bogomips : 5182.68
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
processor : 3
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3687U CPU @ 2.10GHz
stepping : 9
microcode : 0x15
cpu MHz : 2813.585
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 1
cpu cores : 2
apicid : 3
initial apicid : 3
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveopt
bugs :
bogomips : 5182.68
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
#+end_example
- Compiler version
#+begin_src shell :results output :exports both :tangle get_info.sh
gcc --version
#+end_src
#+RESULTS:
: gcc (Debian 5.3.1-6) 5.3.1 20160114
: Copyright (C) 2015 Free Software Foundation, Inc.
: This is free software; see the source for copying conditions. There is NO
: warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
:
- Libpmfr version:
#+begin_src shell :results output :exports both :tangle get_info.sh
apt-cache show libmpfr-dev
#+end_src
#+RESULTS:
#+begin_example
Package: libmpfr-dev
Source: mpfr4
Version: 3.1.5-1
Installed-Size: 1029
Maintainer: Debian GCC Maintainers <debian-gcc@lists.debian.org>
Architecture: amd64
Replaces: libgmp3-dev (<< 4.1.4-3)
Depends: libgmp-dev, libmpfr4 (= 3.1.5-1)
Suggests: libmpfr-doc
Breaks: libgmp3-dev (<< 4.1.4-3)
Description-en: multiple precision floating-point computation developers tools
This development package provides the header files and the symbolic
links to allow compilation and linking of programs that use the libraries
provided in the libmpfr4 package.
.
MPFR provides a library for multiple-precision floating-point computation
with correct rounding. The computation is both efficient and has a
well-defined semantics. It copies the good ideas from the
ANSI/IEEE-754 standard for double-precision floating-point arithmetic
(53-bit mantissa).
Description-md5: a2580b68a7c6f1fcadeefc6b17102b32
Multi-Arch: same
Homepage: http://www.mpfr.org/
Tag: devel::lang:c, devel::library, implemented-in::c, role::devel-lib,
suite::gnu
Section: libdevel
Priority: optional
Filename: pool/main/m/mpfr4/libmpfr-dev_3.1.5-1_amd64.deb
Size: 207200
MD5sum: e5c7872461f263e27312c9ef4f4218b9
SHA256: 279970e210c7db4e2550f5a3b7abb2674d01e9f0afd2a4857f1589a6947e0cbd
#+end_example
*** A first measurement
#+begin_src shell :results output :exports both :tangle measure.sh
cd /tmp/
gcc -O3 a.c -o a
./a 1000
#+end_src
#+RESULTS:
: matrix product took 680ms
: t=9062.368470
#+begin_src shell :results output :exports both :tangle measure.sh
cd /tmp/
gcc -O3 d.c -o d -lmpfr
./d 1000 53
#+end_src
#+RESULTS:
: MPFR library: 3.1.5
: MPFR header: 3.1.5 (based on 3.1.5)
: matrix product took 74460ms
: t=9062.368470
Et donc, chez moi, le ratio est plutôt de
#+begin_src R :results output :session *R* :exports both
74460/844
#+end_src
#+RESULTS:
: [1] 88.22275
*** A second measurement
Ceci étant dit, si je reexécute ces deux codes:
#+begin_src shell :results output :exports both
cd /tmp/
gcc -O3 a.c -o a
./a 1000
#+end_src
#+RESULTS:
: matrix product took 676ms
: t=9062.368470
#+begin_src shell :results output :exports both
cd /tmp/
gcc -O3 d.c -o d -lmpfr
./d 1000 53
#+end_src
#+RESULTS:
: MPFR library: 3.1.5
: MPFR header: 3.1.5 (based on 3.1.5)
: matrix product took 68732ms
: t=9062.368470
J'obtiens une valeur assez différente qui me donnerait cette fois ci
un ratio de
#+begin_src R :results output :session *R* :exports both
68732/676
#+end_src
#+RESULTS:
: [1] 101.6746
c'est à dire "plus proche" de ce qui est annoncé dans [2] mais c'est
un coup de chance, j'aurais tout aussi bien pu obtenir 120 ! Bref,
c'est pas le même setup que vous mais statistiquement parlant, il doit
aussi y avoir quelque chose à faire là, non ?
** References
[1] Fousse, L., Hanrot, G., Lefèvre, V., Pélissier, P., and
Zimmermann, P. MPFR: A multiple-precision binary floating- point
library with correct rounding. ACM Trans. Math. Softw. 33, 2 (2007),
article 13.
[2] van der Hoeven, J. Multiple precision floating-point arithmetic on
SIMD processors. In Proceedings of Arith’24 (2017), IEEE, pp. 2–9.
Entered on [2017-09-01 ven. 17:12]
* Emacs Setup :noexport:
This document has local variables in its postembule, which should
allow Org-mode (9) to work seamlessly without any setup. If you're
uncomfortable using such variables, you can safely ignore them at
startup. Exporting may require that you copy them in your .emacs.
# Local Variables:
# eval: (require 'org-install)
# eval: (org-babel-do-load-languages 'org-babel-load-languages '((sh . t) (R . t) (perl . t) (python .t) ))
# eval: (setq org-confirm-babel-evaluate nil)
# eval: (unless (boundp 'org-latex-classes) (setq org-latex-classes nil))
# eval: (add-to-list 'org-latex-classes '("IEEEtran"
# "\\documentclass[conference, 10pt, compsocconf]{IEEEtran}\n \[NO-DEFAULT-PACKAGES]\n \[EXTRA]\n \\usepackage{graphicx}\n \\usepackage{hyperref}" ("\\section{%s}" . "\\section*{%s}") ("\\subsection{%s}" . "\\subsection*{%s}") ("\\subsubsection{%s}" . "\\subsubsection*{%s}") ("\\paragraph{%s}" . "\\paragraph*{%s}") ("\\subparagraph{%s}" . "\\subparagraph*{%s}")))
# eval: (setq org-alphabetical-lists t)
# eval: (setq org-src-fontify-natively t)
# eval: (add-to-list 'load-path ".")
# eval: (add-to-list 'org-latex-packages-alist '("" "minted"))
# eval: (setq org-latex-listings 'minted)
# eval: (setq org-latex-pdf-process '("pdflatex -shell-escape -interaction nonstopmode -output-directory %o %f" "pdflatex -shell-escape -interaction nonstopmode -output-directory %o %f" "pdflatex -shell-escape -interaction nonstopmode -output-directory %o %f"))
# End:
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment