Model Output API

examples/emcee_vs_r_inference/.gitignore

@@ -0,0 +1,2 @@
+model_output/
+*.ipynb

examples/emcee_vs_r_inference/Makefile

@@ -0,0 +1,91 @@
+# Cross platform customizations
+ifeq ($(OS),Windows_NT)
+    RM_CMD = del /Q
+    RM_DIR = rmdir /S /Q
+else
+    RM_CMD = rm -f
+    RM_DIR = rm -rf
+endif
+
+
+# Default target to build all outputs
+.PHONY: all
+all: \
+	model_output/three_state_seir_stacked_hosp_stacked/emcee_inference \
+	model_output/three_state_seir_stacked_hosp_stacked/r_inference
+
+
+# Clean all generated files & targets
+.PHONY: clean
+clean:
+	$(RM_DIR) model_output
+	$(RM_CMD) model_input/ground_truth_hospitalizations.csv
+	$(RM_CMD) three_state_simulate.yml
+	$(RM_CMD) three_state_emcee_inference.yml
+	$(RM_CMD) three_state_r_inference.yml
+	$(RM_CMD) log*.txt
+	$(RM_CMD) *.h5
+	$(RM_CMD) *.pdf
+
+
+# Simulate ground truth data
+three_state_simulate.yml:
+	flepimop patch \
+		sim_base.yml \
+		--method stochastic \
+		--nslots 1 \
+		--in-id sim > three_state_simulate.yml
+
+model_output/three_state_state_varied_Ro_state_varied_incidH/sim: three_state_simulate.yml
+	flepimop simulate \
+		--seir_modifiers_scenarios state_varied_Ro \
+		--outcome_modifiers_scenarios state_varied_incidH \
+		three_state_simulate.yml
+
+model_input/ground_truth_hospitalizations.csv: model_output/three_state_state_varied_Ro_state_varied_incidH/sim
+	python ground_truth_from_simulation.py
+
+
+# Inference with EMCEE
+three_state_emcee_inference.yml:
+	flepimop patch \
+		inference_base.yml \
+		inference_outcome_modifiers.yml \
+		inference_seir_modifiers.yml \
+		emcee_inference.yml \
+		> three_state_emcee_inference.yml
+
+model_output/three_state_seir_stacked_hosp_stacked/emcee_inference: \
+	three_state_emcee_inference.yml model_input/ground_truth_hospitalizations.csv
+	flepimop-calibrate \
+		--config three_state_emcee_inference.yml \
+		--jobs 12 \
+		--nslots 12 \
+		--niterations 500 \
+		--nsamples 100 \
+		--id emcee_inference
+
+
+# Inference with R
+three_state_r_inference.yml:
+	flepimop patch \
+		inference_base.yml \
+		inference_outcome_modifiers.yml \
+		inference_seir_modifiers.yml \
+		r_inference.yml \
+		> three_state_r_inference.yml
+
+model_output/three_state_seir_stacked_hosp_stacked/r_inference: \
+	three_state_r_inference.yml model_input/ground_truth_hospitalizations.csv
+	flepimop-inference-main \
+		--config three_state_r_inference.yml \
+		--run_id r_inference \
+		--seir_modifiers_scenarios seir_stacked \
+		--outcome_modifiers_scenarios hosp_stacked \
+		--jobs 12 \
+		--iterations_per_slot 600 \
+		--slots 12 \
+		--python "$(CONDA_PREFIX)/bin/python" \
+		--rpath "$(CONDA_PREFIX)/bin/R" \
+		--save_hosp TRUE \
+		--save_seir TRUE

examples/emcee_vs_r_inference/README.md

@@ -0,0 +1,3 @@
+# Three State Inference Comparison
+
+This example highlights the differences between R and EMCEE inference with a small but semi-realistic three state model.

examples/emcee_vs_r_inference/emcee_inference.yml

@@ -0,0 +1,13 @@
+inference:
+  iterations_per_slot: 250
+  method: emcee
+  do_inference: true
+  gt_data_path: model_input/ground_truth_hospitalizations.csv
+  statistics:
+    incidH:
+      name: incidH
+      sim_var: hospitalizations
+      data_var: incidH
+      zero_to_one: True
+      likelihood: 
+        dist: pois

examples/emcee_vs_r_inference/ground_truth_from_simulation.py

@@ -0,0 +1,24 @@
+from pathlib import Path
+
+from gempyor.utils import read_directory
+
+
+project_path = Path(__file__).parent
+
+simulation_output = (
+    project_path
+    / "model_output"
+    / "three_state_state_varied_Ro_state_varied_incidH"
+    / "sim"
+)
+if not simulation_output.exists():
+    raise FileNotFoundError(
+        f"Simulation output directory {simulation_output} does not exist. "
+        "Please run the simulation first."
+    )
+
+hosp = read_directory(simulation_output, filters="hosp")
+hosp = hosp[["date", "subpop", "hospitalizations_curr"]]
+hosp = hosp.rename(columns={"hospitalizations_curr": "incidH"})
+hosp["incidH"] = hosp["incidH"].astype(int)
+hosp.to_csv(project_path / "model_input" / "ground_truth_hospitalizations.csv", index=False)

examples/emcee_vs_r_inference/inference_base.yml

@@ -0,0 +1,58 @@
+name: three_state
+start_date: 2024-01-01
+end_date: 2024-12-31
+
+subpop_setup:
+  geodata: model_input/geodata.csv
+  mobility: model_input/mobility.csv
+
+initial_conditions:
+  method: SetInitialConditions
+  initial_conditions_file: model_input/initial_conditions.csv
+  allow_missing_subpops: TRUE
+  allow_missing_compartments: TRUE
+
+compartments:
+  infection_stage: ["S", "E", "I", "R"]
+
+seir:
+  integration:
+    method: rk4
+    dt: 0.25
+  parameters:
+    sigma:
+      value: 1
+    gamma:
+      value: 1
+    Ro:
+      value: 1
+  transitions:
+    - source: ["S"]
+      destination: ["E"]
+      rate: ["Ro * gamma"]
+      proportional_to: [["S"], ["I"]]
+      proportion_exponent: ["1", "1"]
+    - source: ["E"]
+      destination: ["I"]
+      rate: ["sigma"]
+      proportional_to: ["E"]
+      proportion_exponent: ["1"]
+    - source: ["I"]
+      destination: ["R"]
+      rate: ["gamma"]
+      proportional_to: ["I"]
+      proportion_exponent: ["1"]
+
+outcomes:
+  method: delayframe
+  outcomes:
+    hospitalizations:
+      source:
+        incidence:
+          infection_stage: "I"
+      probability:
+        value: 1
+      delay:
+        value: 1
+      duration:
+        value: 1

examples/emcee_vs_r_inference/inference_outcome_modifiers.yml

@@ -0,0 +1,78 @@
+outcome_modifiers:
+  scenarios:
+    - hosp_stacked
+  modifiers:
+    hosp_probability:
+      method: SinglePeriodModifier
+      parameter: incidH::probability
+      subpop: "all"
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value: 
+        distribution: truncnorm
+        mean: 0.1
+        sd: 0.05
+        a: 0
+        b: 1
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    hosp_delay:
+      method: SinglePeriodModifier
+      parameter: incidH::delay
+      subpop: "all"
+      subpop_groups:
+        -
+          - NC
+          - SC
+        -
+          - GA
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value: 
+        distribution: truncnorm
+        mean: 7.0
+        sd: 2.0
+        a: 0
+        b: 21.0
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    hosp_duration:
+      method: SinglePeriodModifier
+      parameter: incidH::duration
+      subpop: "all"
+      subpop_groups:
+        -
+          - NC
+          - SC
+          - GA
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value: 
+        distribution: truncnorm
+        mean: 14.0
+        sd: 7.0
+        a: 3.0
+        b: 21.0
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    hosp_stacked:
+      method: StackedModifier
+      subpop: "all"
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      modifiers:
+        - hosp_probability
+        - hosp_delay
+        - hosp_duration

examples/emcee_vs_r_inference/inference_seir_modifiers.yml

@@ -0,0 +1,66 @@
+seir_modifiers:
+  scenarios:
+    - seir_stacked
+  modifiers:
+    sigma_fit:
+      method: SinglePeriodModifier
+      parameter: sigma
+      subpop: "all"
+      subpop_groups: ["NC", "SC", "GA"]
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value:
+        distribution: truncnorm
+        mean: 0.5
+        sd: 0.2
+        a: 0
+        b: 3
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    gamma_fit:
+      method: SinglePeriodModifier
+      parameter: gamma
+      subpop: "all"
+      subpop_groups: ["NC", "SC", "GA"]
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value:
+        distribution: truncnorm
+        mean: 0.5
+        sd: 0.2
+        a: 0
+        b: 3
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    Ro_fit:
+      method: SinglePeriodModifier
+      parameter: Ro
+      subpop: "all"
+      period_start_date: 2024-01-01
+      period_end_date: 2024-12-31
+      value:
+        distribution: truncnorm
+        mean: 2.5
+        sd: 0.1
+        a: 0
+        b: 5
+      perturbation:
+        distribution: truncnorm
+        mean: 0
+        sd: 0.025
+        a: -0.1
+        b: 0.1
+    seir_stacked:
+      method: StackedModifier
+      modifiers:
+        - sigma_fit
+        - gamma_fit
+        - Ro_fit

examples/emcee_vs_r_inference/model_input/geodata.csv

@@ -0,0 +1,4 @@
+"subpop","population"
+"NC","10264876"
+"SC","5020806"
+"GA","10403847"