Merge pull request #61 from CDCgov/45-improve-the-prior-interface-for…

…-make_epi_model_inference

Improve the prior interface for make epi model inference

EpiAware/README.md

@@ -9,8 +9,9 @@
 - Solid lines indicate implemented features/analysis.
 - Dashed lines indicate planned features/analysis.
 
+## Proposed `EpiAware` model diagram
 ```mermaid
-flowchart TD
+flowchart LR
 
     A["Underlying dists.
 and specify length of sims
@@ -29,24 +30,40 @@ C["Observational Data
 Obs. cases y_t"]
 D["Latent processes
 ---------------------
-Random Walk"]
-E[Turing model constructor]
-F["Latent Process priors"]
+random_walk"]
+E["Turing model constructor
+---------------------
+make_epi_inference_model"]
+F["Latent Process priors
+---------------------
+default_rw_priors"]
 G[Posterior draws]
 H[Posterior checking]
 I[Post-processing]
 DataW[Data wrangling and QC]
-J["Observation Model
+J["Observation models
 ---------------------
 delay_observations"]
+K["Observation model priors
+---------------------
+default_delay_obs_priors"]
+ObservationModel["ObservationModel
+---------------------
+delay_observations_model"]
+LatentProcess["LatentProcess
+---------------------
+random_walk_process"]
 
 A --> EpiModel
 B --> EpiModel
 EpiModel -->E
 C-->E
-D-->|random_walk| E
-J-->E
-F-->|default_rw_priors|E
+D-->LatentProcess
+F-->LatentProcess
+J-->ObservationModel
+K-->ObservationModel
+LatentProcess-->E
+ObservationModel-->E
 E-->|sample...NUTS...| G
 G-.->H
 H-.->I

EpiAware/src/EpiAware.jl

@@ -38,12 +38,12 @@ export create_discrete_pmf, default_rw_priors, default_delay_obs_priors, spread_
 export EpiData, Renewal, ExpGrowthRate, DirectInfections
 
 # Exported Turing model constructors
-export make_epi_inference_model, random_walk, delay_observations
+export make_epi_inference_model, delay_observations_model, random_walk_process
 
 include("epimodel.jl")
 include("utilities.jl")
-include("models.jl")
 include("latent-processes.jl")
 include("observation-processes.jl")
+include("models.jl")
 
 end

EpiAware/src/latent-processes.jl

@@ -18,3 +18,34 @@ end
     end
     return rw, init, (; σ_RW,)
 end
+
+"""
+    struct LatentProcess{F<:Function}
+
+A struct representing a latent process with its priors.
+
+# Fields
+- `latent_process`: The latent process function for a `Turing` model.
+- `latent_process_priors`: NamedTuple containing the priors for the latent process.
+
+"""
+struct LatentProcess{F <: Function}
+    latent_process::F
+    latent_process_priors::NamedTuple
+end
+
+"""
+    random_walk_process(; latent_process_priors = default_rw_priors())
+
+Create a `LatentProcess` struct reflecting a random walk process with optional priors.
+
+# Arguments
+- `latent_process_priors`: Optional priors for the random walk process.
+
+# Returns
+- `LatentProcess`: A random walk process.
+
+"""
+function random_walk_process(; latent_process_priors = default_rw_priors())
+    LatentProcess(random_walk, latent_process_priors)
+end

EpiAware/src/models.jl

@@ -1,25 +1,26 @@
 @model function make_epi_inference_model(
         y_t,
         epimodel::AbstractEpiModel,
-        latent_process,
-        observation_process;
-        process_priors,
+        latent_process_obj::LatentProcess,
+        observation_process_obj::ObservationModel;
         pos_shift = 1e-6
 )
     #Latent process
     time_steps = epimodel.data.time_horizon
-    @submodel latent_process, init, latent_process_aux = latent_process(
-        time_steps; latent_process_priors = process_priors)
+    @submodel latent_process, init, latent_process_aux = latent_process_obj.latent_process(
+        time_steps;
+        latent_process_priors = latent_process_obj.latent_process_priors
+    )
 
     #Transform into infections
     I_t = epimodel(latent_process, init)
 
     #Predictive distribution of ascerted cases
-    @submodel generated_y_t, generated_y_t_aux = observation_process(
+    @submodel generated_y_t, generated_y_t_aux = observation_process_obj.observation_model(
         y_t,
         I_t,
         epimodel::AbstractEpiModel;
-        observation_process_priors = process_priors,
+        observation_process_priors = observation_process_obj.observation_model_priors,
         pos_shift = pos_shift
     )
 

EpiAware/src/observation-processes.jl

@@ -21,3 +21,34 @@ end
 
     return y_t, (; neg_bin_cluster_factor,)
 end
+
+"""
+    struct ObservationModel{F<:Function}
+
+A struct representing an observation model with its priors.
+
+# Fields
+- `observation_model`: The observation model function for a `Turing` model.
+- `observation_model_priors`: NamedTuple containing the priors for the observation model.
+
+"""
+struct ObservationModel{F <: Function}
+    observation_model::F
+    observation_model_priors::NamedTuple
+end
+
+"""
+    delay_observations_model(; latent_process_priors = default_rw_priors())
+
+Create an `ObservationModel` struct reflecting a delayed observation process with optional priors.
+
+# Arguments
+- `latent_process_priors`: Optional priors for the delayed observation process.
+
+# Returns
+- `ObservationModel`: An observation model with delayed observations.
+
+"""
+function delay_observations_model(; observation_model_priors = default_delay_obs_priors())
+    ObservationModel(delay_observations, observation_model_priors)
+end

EpiAware/test/predictive_checking/toy_model_log_infs_RW.jl

@@ -101,20 +101,16 @@ In this case we use the `DirectInfections` model.
 =#
 
 toy_log_infs = DirectInfections(model_data)
+rwp = random_walk_process()
+obs_mdl = delay_observations_model()
 
 #=
 ## Generate a `Turing` `Model`
 We don't have observed data, so we use `missing` value for `y_t`.
 =#
 
 log_infs_model = make_epi_inference_model(
-    missing,
-    toy_log_infs,
-    random_walk,
-    delay_observations;
-    process_priors = merge(default_rw_priors(), default_delay_obs_priors()),
-    pos_shift = 1e-6
-)
+    missing, toy_log_infs, rwp, obs_mdl; pos_shift = 1e-6)
 
 #=
 ## Sample from the model
@@ -147,14 +143,7 @@ We treat the generated data as observed data and attempt to infer underlying inf
 
 truth_data = random_epidemic.y_t
 
-model = make_epi_inference_model(
-    truth_data,
-    toy_log_infs,
-    random_walk,
-    delay_observations;
-    process_priors = merge(default_rw_priors(), default_delay_obs_priors()),
-    pos_shift = 1e-6
-)
+model = make_epi_inference_model(truth_data, toy_log_infs, rwp, obs_mdl; pos_shift = 1e-6)
 
 @time chn = sample(
     model,

EpiAware/test/test_latent-processes.jl

@@ -2,7 +2,7 @@
 @testitem "Testing random_walk against theoretical properties" begin
     using DynamicPPL, Turing
     n = 5
-    model = random_walk(n)
+    model = EpiAware.random_walk(n)
     fixed_model = fix(model, (σ²_RW = 1.0, init_rw_value = 0.0)) #Fixing the standard deviation of the random walk process
     n_samples = 1000
     samples_day_5 = sample(fixed_model, Prior(), n_samples) |>

EpiAware/test/test_models.jl

@@ -4,20 +4,13 @@
     # Define test inputs
     y_t = missing # Data will be generated from the model
     data = EpiData([0.2, 0.3, 0.5], [0.1, 0.4, 0.5], 0.8, 10, exp)
-    process_priors = merge(default_rw_priors(), default_delay_obs_priors())
     pos_shift = 1e-6
 
     epimodel = DirectInfections(data)
-
+    rwp = random_walk_process()
+    obs_mdl = delay_observations_model()
     # Call the function
-    test_mdl = make_epi_inference_model(
-        y_t,
-        epimodel,
-        random_walk,
-        delay_observations;
-        process_priors,
-        pos_shift
-    )
+    test_mdl = make_epi_inference_model(y_t, epimodel, rwp, obs_mdl; pos_shift)
 
     # Define expected outputs for a conditional model
     # Underlying log-infections are const value 1 for all time steps and
@@ -38,20 +31,14 @@ end
     # Define test inputs
     y_t = missing # Data will be generated from the model
     data = EpiData([0.2, 0.3, 0.5], [0.1, 0.4, 0.5], 0.8, 10, exp)
-    process_priors = merge(default_rw_priors(), default_delay_obs_priors())
     pos_shift = 1e-6
 
     epimodel = ExpGrowthRate(data)
+    rwp = random_walk_process()
+    obs_mdl = delay_observations_model()
 
     # Call the function
-    test_mdl = make_epi_inference_model(
-        y_t,
-        epimodel,
-        random_walk,
-        delay_observations;
-        process_priors,
-        pos_shift
-    )
+    test_mdl = make_epi_inference_model(y_t, epimodel, rwp, obs_mdl; pos_shift)
 
     # Define expected outputs for a conditional model
     # Underlying log-infections are const value 1 for all time steps and
@@ -76,16 +63,10 @@ end
     pos_shift = 1e-6
 
     epimodel = Renewal(data)
-
+    rwp = random_walk_process()
+    obs_mdl = delay_observations_model()
     # Call the function
-    test_mdl = make_epi_inference_model(
-        y_t,
-        epimodel,
-        random_walk,
-        delay_observations;
-        process_priors,
-        pos_shift
-    )
+    test_mdl = make_epi_inference_model(y_t, epimodel, rwp, obs_mdl; pos_shift)
 
     # Define expected outputs for a conditional model
     # Underlying log-infections are const value 1 for all time steps and

EpiAware/test/test_observation-processes.jl

@@ -11,7 +11,7 @@
     observation_process_priors = default_delay_obs_priors()
 
     # Call the function
-    mdl = delay_observations(
+    mdl = EpiAware.delay_observations(
         missing,
         I_t,
         epimodel;