Generic dataset (#43)

* Refactor Dataset class to enhance generalisation

* New config and simulator supporting a wide range of data generating scnearios.

* Config and simulate tests

* Delete weight files and classes

* Add inflate method for Dataset class and adapt StaticEffect, RandomEffect for the new treatment assignment mechanism.

* Testing inflate and effect

* Adapting transforms to new dataset class

* Test noise transformation

* Testing periodic transform

* Test trend transform

* Adjusting validation components according to new dataset class

* Plotter modification

* Remove models.py and associated tests

* test_models to test_az_causal

* Format changes

* Linting and documentation fix

Author: semihakbayrak

docs/examples/azcausal.ipynb

@@ -17,15 +17,16 @@ trend varies across each of the 10 control units.
 from causal_validation import Config, simulate
 from causal_validation.effects import StaticEffect
 from causal_validation.plotters import plot
-from causal_validation.transforms import Trend, Periodic
+from causal_validation.transforms import Trend
 from causal_validation.transforms.parameter import UnitVaryingParameter
+import numpy as np
 from scipy.stats import norm
 
-cfg = Config(
-    n_control_units=10,
-    n_pre_intervention_timepoints=60,
-    n_post_intervention_timepoints=30,
-)
+# Treatment assignment matrix
+D = np.zeros((90, 11))  # 90 time points, 11 units
+D[60:, -1] = 1  # Last unit treated after 60 time points
+
+cfg = Config(treatment_assignments=D)
 
 # Simulate the base observation
 base_data = simulate(cfg)
@@ -38,6 +39,8 @@ trended_data = trend_component(base_data)
 # Simulate a 5% lift in the treated unit's post-intervention data
 effect = StaticEffect(0.05)
 inflated_data = effect(trended_data)
+
+plot(inflated_data)
 ```
 
 ![Gaussian process posterior.](static/imgs/readme_fig.png)
@@ -50,6 +53,7 @@ combination with AZCausal by the following.
 
 ```python
 from azcausal.estimators.panel.sdid import SDID
+from causal_validation.estimator.utils import AZCausalWrapper
 from causal_validation.validation.placebo import PlaceboTest
 
 model = AZCausalWrapper(model=SDID())

docs/examples/basic.ipynb

@@ -1,6 +1,5 @@
 from dataclasses import (
     dataclass,
-    field,
 )
 import datetime as dt
 import typing as tp
@@ -11,84 +10,99 @@
 
 from causal_validation.types import (
     Number,
-    WeightTypes,
+    TreatedSimulationTypes,
 )
-from causal_validation.weights import UniformWeights
-
-
-@dataclass(kw_only=True, frozen=True)
-class WeightConfig:
-    weight_type: "WeightTypes" = field(default_factory=UniformWeights)
 
 
 @dataclass(kw_only=True)
 class Config:
     """Configuration for causal data generation.
 
     Args:
-        n_control_units (int): Number of control units in the synthetic dataset.
-        n_pre_intervention_timepoints (int): Number of time points before intervention.
-        n_post_intervention_timepoints (int): Number of time points after intervention.
+        treatment_assignments (Float[np.ndarray, "T N"]): Treatment assignments for T
+            time steps and N units. Only supported with binary assignments.
+        treated_simulation_type ("TreatedSimulationTypes"): Treated units can be
+            simulated either "independent" of control units or "control-weighted",
+            where waiting scheme is controlled by Dirichlet concentration parameter.
+            Set to "control-weighted" by default.
+        dirichlet_concentration (Number): Dirichlet parameters are set to a vector of
+            dirichlet_concentration with length number of control units. This parameter
+            controls how dense and sparse the generated weights are. Set to 1 by default
+            and in effect only if treated_simulation_type is "control-weighted".
         n_covariates (Optional[int]): Number of covariates. Defaults to None.
-        covariate_means (Optional[Float[np.ndarray, "D K"]]): Mean values for covariates
-            D is n_control_units and K is n_covariates. Defaults to None. If it is set
-            to None while n_covariates is provided, covariate_means will be generated
+        covariate_means (Optional[np.ndarray]): Normal dist. mean values for covariates.
+            The lenght must be n_covariates. Defaults to None. If it is set to
+            None while n_covariates is provided, covariate_means will be generated
             randomly from Normal distribution.
-        covariate_stds (Optional[Float[np.ndarray, "D K"]]): Standard deviations for
-            covariates. D is n_control_units and K is n_covariates. Defaults to None.
-            If it is set to None while n_covariates is provided, covariate_stds
-            will be generated randomly from Half-Cauchy distribution.
+        covariate_stds (Optional[np.ndarray]): Normal dist. std values for covariates.
+            The lenght must be n_covariates. Defaults to None. If it is set to
+            None while n_covariates is provided, covariate_stds will be generated
+            randomly from Half-Cauchy distribution.
         covariate_coeffs (Optional[np.ndarray]): Linear regression
             coefficients to map covariates to output observations. K is n_covariates.
             Defaults to None.
         global_mean (Number): Global mean for data generation. Defaults to 20.0.
         global_scale (Number): Global scale for data generation. Defaults to 0.2.
         start_date (dt.date): Start date for time series. Defaults to 2023-01-01.
         seed (int): Random seed for reproducibility. Defaults to 123.
-        weights_cfg (WeightConfig): Configuration for unit weights. Defaults to
-            UniformWeights.
+        weights (Optional[list[np.ndarray]]): Length num of treateds list of weights.
+            Each element is length num of control, indicating how to weigh control
+            units to generate treated.
     """
 
-    n_control_units: int
-    n_pre_intervention_timepoints: int
-    n_post_intervention_timepoints: int
+    treatment_assignments: Float[np.ndarray, "T N"]
+    treated_simulation_type: "TreatedSimulationTypes" = "control-weighted"
+    dirichlet_concentration: Number = 1.0
     n_covariates: tp.Optional[int] = None
-    covariate_means: tp.Optional[Float[np.ndarray, "D K"]] = None
-    covariate_stds: tp.Optional[Float[np.ndarray, "D K"]] = None
+    covariate_means: tp.Optional[np.ndarray] = None
+    covariate_stds: tp.Optional[np.ndarray] = None
     covariate_coeffs: tp.Optional[np.ndarray] = None
     global_mean: Number = 20.0
     global_scale: Number = 0.2
     start_date: dt.date = dt.date(year=2023, month=1, day=1)
     seed: int = 123
-    weights_cfg: WeightConfig = field(default_factory=WeightConfig)
+    weights: tp.Optional[list[np.ndarray]] = None
 
     def __post_init__(self):
         self.rng = np.random.RandomState(self.seed)
         if self.covariate_means is not None:
-            assert self.covariate_means.shape == (
-                self.n_control_units,
-                self.n_covariates,
-            )
+            assert self.covariate_means.shape == (self.n_covariates,)
 
         if self.covariate_stds is not None:
-            assert self.covariate_stds.shape == (
-                self.n_control_units,
-                self.n_covariates,
-            )
+            assert self.covariate_stds.shape == (self.n_covariates,)
 
         if (self.n_covariates is not None) & (self.covariate_means is None):
             self.covariate_means = self.rng.normal(
-                loc=0.0, scale=5.0, size=(self.n_control_units, self.n_covariates)
+                loc=0.0, scale=5.0, size=(self.n_covariates)
             )
 
         if (self.n_covariates is not None) & (self.covariate_stds is None):
             self.covariate_stds = halfcauchy.rvs(
                 scale=0.5,
-                size=(self.n_control_units, self.n_covariates),
+                size=(self.n_covariates),
                 random_state=self.rng,
             )
 
         if (self.n_covariates is not None) & (self.covariate_coeffs is None):
             self.covariate_coeffs = self.rng.normal(
                 loc=0.0, scale=5.0, size=self.n_covariates
             )
+
+        n_units = self.treatment_assignments.shape[1]
+        treated_units = [
+            i for i in range(n_units) if any(self.treatment_assignments[:, i] != 0)
+        ]
+        n_treated_units = len(treated_units)
+        n_control_units = n_units - n_treated_units
+
+        if self.treated_simulation_type == "control-weighted":
+            if self.weights is None:
+                self.weights = [
+                    self.rng.dirichlet(
+                        self.dirichlet_concentration * np.ones(n_control_units)
+                    )
+                    for _ in range(n_treated_units)
+                ]
+            else:
+                assert len(self.weights) == n_treated_units
+                assert all([len(w) == n_control_units for w in self.weights])