Finish 3-2.

module3/exo2/exercice_python_en.org

@@ -61,8 +61,8 @@ Alright, looks good 👍
 
 ** Sanitise the data
 
-But there may be problems with the data. Let's check for the obvious
-case of missing/empty data:
+But the data may not be without its issues. Let's check for the
+obvious case of missing/empty data:
 
 #+begin_src python :results output :exports both
 valid_table = []
@@ -97,29 +97,29 @@ separators in the resulting table (which contains the first five and
 last five weeks' data):
 
 #+begin_src python :results value :exports both
-[('week', 'inc'), None] + data[:5] + [None] + data[-5:]
+[('week', 'incidence'), None] + data[:5] + [None] + data[-5:]
 #+end_src
 
 #+RESULTS:
-|   week |   inc |
-|--------+-------|
-| 202416 | 19330 |
-| 202415 | 24807 |
-| 202414 | 16181 |
-| 202413 | 18322 |
-| 202412 | 12818 |
-|--------+-------|
-| 199101 | 15565 |
-| 199052 | 19375 |
-| 199051 | 19080 |
-| 199050 | 11079 |
-| 199049 |  1143 |
+|   week | incidence |
+|--------+-----------|
+| 202416 |     19330 |
+| 202415 |     24807 |
+| 202414 |     16181 |
+| 202413 |     18322 |
+| 202412 |     12818 |
+|--------+-----------|
+| 199101 |     15565 |
+| 199052 |     19375 |
+| 199051 |     19080 |
+| 199050 |     11079 |
+| 199049 |      1143 |
 
 ** Convert dates
 
 Dates are represented in ISO 8601 format (YYYYWW) so let's parse
-those. It should already be sorted chronologically, but let's make
-sure of that too.
+those. Entries are sorted chronologically, but in reverse, so we'll
+fix that here too.
 
 #+begin_src python :results silent :exports both
 import datetime
@@ -133,40 +133,42 @@ Let's check again:
 
 #+begin_src python :results value :exports both
 data_as_str = [(str(date), str(inc)) for date, inc in converted_data]
-[('date', 'inc'), None] + data_as_str[:5] + [None] + data_as_str[-5:]
+[('date', 'incidence'), None] + data_as_str[:5] + [None] + data_as_str[-5:]
 #+end_src
 
 #+RESULTS:
-|       date |   inc |
-|------------+-------|
-| 1990-12-03 |  1143 |
-| 1990-12-10 | 11079 |
-| 1990-12-17 | 19080 |
-| 1990-12-24 | 19375 |
-| 1990-12-31 | 15565 |
-|------------+-------|
-| 2024-03-18 | 12818 |
-| 2024-03-25 | 18322 |
-| 2024-04-01 | 16181 |
-| 2024-04-08 | 24807 |
-| 2024-04-15 | 19330 |
+|       date | incidence |
+|------------+-----------|
+| 1990-12-03 |      1143 |
+| 1990-12-10 |     11079 |
+| 1990-12-17 |     19080 |
+| 1990-12-24 |     19375 |
+| 1990-12-31 |     15565 |
+|------------+-----------|
+| 2024-03-18 |     12818 |
+| 2024-03-25 |     18322 |
+| 2024-04-01 |     16181 |
+| 2024-04-08 |     24807 |
+| 2024-04-15 |     19330 |
 
 ** Visual inspection
 
 So, now we can take a look at incidence over time. (The 'flu notebook
 switches to R here, but we're going to stick with python.)
 
-#+begin_src python :results output file :var filename="./incidence.png" :exports both
+#+begin_src python :results value file :var filename="./incidence.png" :exports both
 import matplotlib.pyplot as plt
 
 plt.clf()
 
 date,incidence = zip(*converted_data)
 
+plt.figure(figsize=(7.5,5))
+
 plt.plot(date,incidence)
-plt.tight_layout()
+# plt.tight_layout()
 plt.savefig(filename)
-print(filename)
+filename
 #+end_src
 
 #+RESULTS:
@@ -174,17 +176,17 @@ print(filename)
 
 And we can zoom in on a period of, say, five years:
 
-#+begin_src python :results output file :var filename="./incidence-zoom.png" :exports both
+#+begin_src python :results value file :var filename="./incidence-zoom.png" :exports both
 plt.clf()
 
-start = 10
+start = 15
 years = 5
 date,incidence = zip(*converted_data[52*start:52*(start+years)])
 
 plt.plot(date,incidence)
-plt.tight_layout()
+# plt.tight_layout()
 plt.savefig(filename)
-print(filename)
+filename
 #+end_src
 
 #+RESULTS:
@@ -192,3 +194,78 @@ print(filename)
 
 It looks like incidence peaks in the spring, with lowest numbers
 around September.
+
+* Study of annual incidence
+
+** Compute annual incidence
+
+So, let's calculate the incidence for each year. We'll define this is
+as the sum of the incidence reports from the beginning of September in
+year /N-1/ to the end of August in year /N/.
+
+#+begin_src python :results silent :exports both
+def annual_incidence(year):
+    start = datetime.datetime.strptime(f"{year-1}-09-01", '%Y-%m-%d').date()
+    end = datetime.datetime.strptime(f"{year}-09-01", '%Y-%m-%d').date()
+    weeks = [d for d in converted_data if d[0] > start and d[0] <= end]
+    return sum([w[1] for w in weeks])
+#+end_src
+
+That was quick and dirty and not entirely a nice time.
+
+Now we can define the years we're interested in. These are the years
+for which we have a full year's worth of data:
+
+#+begin_src python :results silent :exports both
+years = range(1991, 2024)
+#+end_src
+
+NB. The second argument to ~range~ is non-inclusive.
+
+Now we can perform a list comprehension to get the incidence for each
+year:
+
+#+begin_src python :results value :exports both
+incidence_per_year = [(y,annual_incidence(y)) for y in years]
+head, *tail = incidence_per_year
+head
+#+end_src
+
+#+RESULTS:
+| 1991 | 553895 |
+
+** Visual Inspection
+
+Now we can plot incidence against year.
+
+#+begin_src python :results value file :var filename="./annual_incidence.png" :exports both
+plt.clf()
+
+years,incidences = zip(*incidence_per_year)
+plt.bar(years,incidences)
+plt.ylabel("Annual incidence")
+
+plt.savefig(filename)
+filename
+#+end_src
+
+#+RESULTS:
+[[file:./annual_incidence.png]]
+
+Eyeballing the plots, it looks like although 2003-04 had the greatest
+spikes in terms of weekly cases, 2009-10 featured longer spells of
+consistently high numbers of cases.
+
+Anyway, let's get the strongest and weakest epidemics:
+
+#+begin_src python :results value :exports both
+key = lambda y: y[1]
+strongest = max(incidence_per_year, key=key)
+weakest = min(incidence_per_year, key=key)
+
+(strongest,weakest)
+#+end_src
+
+#+RESULTS:
+| 2009 | 841233 |
+| 2020 | 221183 |