Merge branch 'main' of github.com:CITCOM-project/CausalTestingFramework into poisson-process-example

Author: jmafoster1

examples/covasim_/README.md

@@ -0,0 +1,20 @@
+# Covasim Case Study: Doubling Beta (Infectiousness)
+In this case study, we demonstrate how to use the causal testing framework with observational
+data collected Covasim to conduct Statistical Metamorphic Testing (SMT) a posteriori. Here, we focus on a set of simple
+modelling scenarios that investigate how the infectiousness of the virus (encoded as the parameter beta) affects the
+cumulative number of infections over a fixed duration. We also run several causal tests that focus on increasingly
+specific causal questions pertaining to more refined metamorphic properties and enabling us to learn more about the
+model's behaviour without further executions.
+
+More information about the case study can be found in Section 5.3
+(Doubling Beta) of the paper.
+
+## How to run
+To run this case study:
+1. Ensure all project dependencies are installed by running `pip install .` from the top
+level of this directory (instructions are provided in the project README).
+2. Change directory to `causal_testing/examples/covasim_/doubling_beta`.
+3. Run the command `python causal_test_beta.py`.
+
+This will print out a series of test results covering a range of different causal questions that correspond to those
+in Table 3 of the paper.

examples/covasim_/doubling_beta/README.md

@@ -13,18 +13,19 @@
 from causal_testing.testing.estimators import LinearRegressionEstimator
 from matplotlib.pyplot import rcParams
 
-# Make all graphs publication quality
-plt.rcParams["figure.figsize"] = (8, 8)
-rc_fonts = {
-    "font.size": 8,
-    "figure.figsize": (10, 6),
-    "text.usetex": True,
-    "font.family": "serif",
-    "text.latex.preamble": r"\usepackage{libertine}",
-}
-rcParams.update(rc_fonts)
+# Uncommenting the code below will make all graphs publication quality but requires a suitable latex installation
 
-OBSERVATIONAL_DATA_PATH = "./data/10k_observational_data.csv"
+# plt.rcParams["figure.figsize"] = (8, 8)
+# rc_fonts = {
+#     "font.size": 8,
+#     "figure.figsize": (10, 6),
+#     "text.usetex": True,
+#     "font.family": "serif",
+#     "text.latex.preamble": r"\usepackage{libertine}",
+# }
+# rcParams.update(rc_fonts)
+
+OBSERVATIONAL_DATA_PATH = "data/10k_observational_data.csv"
 
 
 def doubling_beta_CATE_on_csv(observational_data_path: str, simulate_counterfactuals: bool = False,
@@ -121,13 +122,13 @@ def doubling_beta_CATEs(observational_data_path: str, simulate_counterfactual: b
 
     # Split df into two age ranges
     younger_population_df = past_execution_df.loc[past_execution_df['avg_age'] <= mid_age]
-    younger_population_df.to_csv("./data/bessemer/younger_population.csv")
+    younger_population_df.to_csv("./data/younger_population.csv")
     older_population_df = past_execution_df.loc[past_execution_df['avg_age'] > mid_age]
-    older_population_df.to_csv("./data/bessemer/older_population.csv")
+    older_population_df.to_csv("./data/older_population.csv")
 
     # Repeat analysis on age-specific strata
-    separated_observational_data_paths = ["./data/bessemer/younger_population.csv",
-                                          "./data/bessemer/older_population.csv"]
+    separated_observational_data_paths = ["./data/younger_population.csv",
+                                          "./data/older_population.csv"]
 
     for col, separated_observational_data_path in enumerate(separated_observational_data_paths):
         age_data_results_dict = doubling_beta_CATE_on_csv(separated_observational_data_path, simulate_counterfactual,
@@ -146,13 +147,13 @@ def doubling_beta_CATEs(observational_data_path: str, simulate_counterfactual: b
 
         # Save dfs to csv
         low_contacts_df = age_stratified_df.loc[age_stratified_df['contacts'] <= mid_contacts]
-        low_contacts_df.to_csv(f"./data/bessemer/low_contacts_avg_age_{age_stratified_df_avg_age}.csv")
+        low_contacts_df.to_csv(f"./data/low_contacts_avg_age_{age_stratified_df_avg_age}.csv")
         high_contacts_df = age_stratified_df.loc[age_stratified_df['contacts'] > mid_contacts]
-        high_contacts_df.to_csv(f"./data/bessemer/high_contacts_avg_age_{age_stratified_df_avg_age}.csv")
+        high_contacts_df.to_csv(f"./data/high_contacts_avg_age_{age_stratified_df_avg_age}.csv")
 
-        contact_observational_data_paths = [f"./data/bessemer/low_contacts_avg_age_"
+        contact_observational_data_paths = [f"./data/low_contacts_avg_age_"
                                             f"{age_stratified_df_avg_age}.csv",
-                                            f"./data/bessemer/high_contacts_avg_age_"
+                                            f"./data/high_contacts_avg_age_"
                                             f"{age_stratified_df_avg_age}.csv"]
 
         # Compute the CATE for each age-contact group
@@ -181,7 +182,7 @@ def doubling_beta_CATEs(observational_data_path: str, simulate_counterfactual: b
 
 def identification(observational_data_path):
     # 1. Read in the Causal DAG
-    causal_dag = CausalDAG('./dag.dot')
+    causal_dag = CausalDAG('dag.dot')
 
     # 2. Create variables
     pop_size = Input('pop_size', int)

examples/covasim_/causal_test_beta.py renamed to examples/covasim_/doubling_beta/causal_test_beta.py

@@ -0,0 +1,25 @@
+# Covasim Case Study: Prioritising Vaccination for the Elderly
+In this case study, we demonstrate how to use causal testing to determine whether prioritising the elderly
+has the expected effect on a series of vaccine-related outcomes. Specifically, cumulative infections, vaccines (number
+of doses administered), vaccinated (number of agents vaccinated), and maximum doses per agent. As explained in Section
+5.3 (Prioritising the elderly for vaccination), we expect that changing the Pfizer vaccine to additionally prioritise
+the elderly should cause vaccines and vaccinated to decrease (more restrictive policy), infections to increase (less
+are vaccinated), and no change to the maximum doses since this should always be 2.
+
+This case study directly executes Covasim under two input configurations that differ only in their vaccine input: one
+is the default Pfizer vaccine, the other is the same vaccine but additionally sub-targeting the elderly using a method 
+provided in the Covasim vaccine tutorial. We execute each of these input configurations 30 times and repeat this for
+four test cases: one focusing on each of the four previously mentioned outputs.
+
+Further details are provided in Section 5.3 (Prioritising the elderly for vaccination) of the paper.
+
+## How to run
+To run this case study:
+1. Ensure all project dependencies are installed by running `pip install .` from the top
+level of this directory (instructions are provided in the project README).
+2. Additionally, in order to run Covasim, install version 3.07 by running `pip install covasim==3.0.7`.
+3. Change directory to `causal_testing/examples/covasim_/vaccinating_elderly`.
+4. Run the command `python causal_test_vaccine.py`.
+
+This will run Covasim as described above and print out the causal test results for the effect of each input on each
+output.

examples/covasim_/dag.dot renamed to examples/covasim_/doubling_beta/dag.dot

@@ -25,7 +25,7 @@ def experimental_causal_test_vaccinate_elderly(runs_per_test_per_config: int = 3
     """
 
     # 1. Read in the Causal DAG
-    causal_dag = CausalDAG('./vaccine_dag.dot')
+    causal_dag = CausalDAG('dag.dot')
 
     # 2. Create variables
     pop_size = Input('pop_size', int)