fix: deterministic expenv, feature: uniform distribution

Author: iraedeus

experimental_env/analysis/analyze_summarizers/error_summarizer.py

@@ -38,10 +38,11 @@ def calculate(self, results: list[ExperimentDescription]) -> tuple:
 
             errors.append(error)
 
+        if not errors:
+            return 0, 0, 0
+
         mean = np.sum(errors) / len(errors)
         standart_deviation = np.sqrt(np.sum([(x - mean) ** 2 for x in errors]) / len(errors))
-
-        errors.sort()
         median = errors[len(errors) // 2]
 
         return float(mean), float(standart_deviation), float(median)

experimental_env/experiment/estimators.py

@@ -50,7 +50,7 @@ def __init__(self, brkpointer: EM.ABreakpointer, dst_checker: EM.ADistributionCh
 
     @property
     def name(self):
-        return "MLE-EM"
+        return "EM"
 
     def _helper(self, problem: OrderedProblem):
         """
@@ -90,7 +90,7 @@ def __init__(self, brkpointer, dst_checker):
 
     @property
     def name(self):
-        return "LM-EM"
+        return "ELM"
 
     def _helper(self, problem: OrderedProblem):
         """

experimental_env/experiment/experiment_executors/abstract_executor.py

@@ -1,5 +1,6 @@
 """A module that provides an abstract class for performing the 2nd stage of the experiment"""
 
+import random
 import warnings
 from abc import ABC, abstractmethod
 from pathlib import Path
@@ -20,7 +21,7 @@ class AExecutor(ABC):
     as well as the implementation of the execute method, to implement the 2nd stage of the experiment.
     """
 
-    def __init__(self, path: Path, cpu_count: int, seed):
+    def __init__(self, path: Path, cpu_count: int, seed: int):
         """
         Class constructor
 
@@ -31,6 +32,8 @@ def __init__(self, path: Path, cpu_count: int, seed):
         self._out_dir = path
         self._cpu_count = cpu_count
         self._seed = seed
+
+        random.seed(self._seed)
         np.random.seed(self._seed)
 
     @abstractmethod

experimental_env/experiment/experiment_executors/random_executor.py

@@ -19,7 +19,7 @@ def init_problems(self, ds_descriptions, models):
         return [
             Problem(
                 descr.samples,
-                RandomMixtureGenerator(self._seed).create_mixture(models),
+                RandomMixtureGenerator().create_mixture(models),
             )
             for i, descr in enumerate(ds_descriptions)
         ]

experimental_env/experiment/experiment_executors/standart_executor.py

@@ -19,7 +19,7 @@ def init_problems(self, ds_descriptions, models):
         return [
             Problem(
                 descr.samples,
-                StandartMixtureGenerator(self._seed).create_mixture(models),
+                StandartMixtureGenerator().create_mixture(models),
             )
             for i, descr in enumerate(ds_descriptions)
         ]

experimental_env/mixture_generators/abstract_generator.py

@@ -1,6 +1,5 @@
 """A module that provides an abstract class for generating a mixture."""
 
-import random
 from abc import ABC, abstractmethod
 
 from mpest import Distribution, MixtureDistribution
@@ -12,9 +11,6 @@ class AMixtureGenerator(ABC):
     An abstract class for generating mixtures.
     """
 
-    def __init__(self, seed: int = 42):
-        random.seed(seed)
-
     @abstractmethod
     def generate_priors(self, models: list[type[AModel]]) -> list[float | None]:
         """

experimental_env/mixture_generators/utils.py

@@ -3,7 +3,8 @@
 from random import uniform
 
 from mpest import Distribution
-from mpest.models import AModel, ExponentialModel, GaussianModel
+from mpest.models import AModel, Beta, Cauchy, ExponentialModel, GaussianModel, Pareto
+from mpest.models.uniform import Uniform
 
 
 def generate_standart_params(models: list[type[AModel]]) -> list[Distribution]:
@@ -14,10 +15,14 @@ def generate_standart_params(models: list[type[AModel]]) -> list[Distribution]:
     for m in models:
         if m == ExponentialModel:
             params = [1.0]
-        elif m == GaussianModel:
+        elif m in (GaussianModel, Uniform, Cauchy):
             params = [0.0, 1.0]
-        else:
-            params = [1.0, 1.5]
+        elif m == Beta:
+            params = [1.0, 1.0]
+        elif m == Pareto:
+            params = [1.0, 2.0]
+        else:  # Weibull
+            params = [1.0, 1.0]
 
         dists.append(Distribution.from_params(m, params))
 
@@ -34,7 +39,15 @@ def generate_uniform_params(models: list[type[AModel]]) -> list[Distribution]:
             params = [uniform(0.1, 5.0)]
         elif m == GaussianModel:
             params = [uniform(-5.0, 5.0), uniform(0.1, 5.0)]
-        else:
+        elif m == Uniform:
+            params = list(sorted([uniform(-5.0, 5.0), uniform(-5.0, 5.0)]))
+        elif m == Cauchy:
+            params = [uniform(-5.0, 5.0), uniform(0.1, 5.0)]
+        elif m == Beta:
+            params = [uniform(0.1, 5.0), uniform(0.1, 5.0)]
+        elif m == Pareto:
+            params = [uniform(0.1, 5.0), uniform(1.0, 5.0)]
+        else:  # Weibull
             params = [uniform(0.1, 5.0), uniform(0.1, 5.0)]
 
         dists.append(Distribution.from_params(m, params))

experimental_env/preparation/dataset_generator.py

@@ -26,9 +26,11 @@ def __init__(self, seed: int = 42):
         """
         Setting seed for determined result.
         """
-        random.seed(seed)
         self._seed = seed
 
+        random.seed(self._seed)
+        np.random.seed(self._seed)
+
     def generate(
         self,
         samples_size: int,
@@ -59,7 +61,6 @@ class ConcreteDatasetGenerator:
     """
 
     def __init__(self, seed: int = 42):
-        np.random.seed(seed)
         self._dists: list[Distribution] = []
         self._priors: list[float | None] = []
 

mpest/em/methods/likelihood_method.py

@@ -8,8 +8,7 @@
 from mpest.core.mixture_distribution import MixtureDistribution
 from mpest.core.problem import Problem, Result
 from mpest.em.methods.abstract_steps import AExpectation, AMaximization
-from mpest.exceptions import SampleError
-from mpest.models import AModel, AModelDifferentiable
+from mpest.models import AModel, AModelDifferentiable, Uniform
 from mpest.optimizers import AOptimizerJacobian, TOptimizer
 from mpest.utils import ResultWithError
 
@@ -31,16 +30,8 @@ def step(self, problem: Problem) -> EResult:
         samples = problem.samples
         mixture = problem.distributions
         p_xij = []
-        active_samples = []
         for x in samples:
-            p = np.array([d.model.pdf(x, d.params) for d in mixture])
-            if np.any(p):
-                p_xij.append(p)
-                active_samples.append(x)
-
-        if not active_samples:
-            error = SampleError("None of the elements in the sample is correct for this mixture")
-            return ResultWithError(mixture, error)
+            p_xij.append(np.array([d.model.pdf(x, d.params) for d in mixture]))
 
         # h[j, i] contains probability of X_i to be a part of distribution j
         m = len(p_xij)
@@ -56,7 +47,7 @@ def step(self, problem: Problem) -> EResult:
 
             h[:, i] = wp / swp
 
-        return active_samples, h, problem
+        return samples, h, problem
 
 
 # class ML(AExpectation[EResult]):
@@ -109,8 +100,30 @@ def step(self, e_result: EResult) -> Result:
         for j, ch in enumerate(h[:]):
             d = mixture[j]
 
+            if isinstance(d.model, Uniform):
+                threshold = 1e-2
+                curr_a, curr_b = d.params
+                relevant_indices = np.where(ch > threshold)[0]
+                if len(relevant_indices) == 0:
+                    new_params = d.params
+                else:
+                    relevant_samples = np.array(samples)[relevant_indices]
+
+                    new_a = np.min(relevant_samples)
+                    new_b = np.max(relevant_samples)
+
+                    new_params = np.array([new_a, new_b])
+
+                new_distributions.append(Distribution(d.model, new_params))
+                continue
+
             def log_likelihood(params, ch, model: AModel):
-                return -np.sum(ch * [model.lpdf(x, params) for x in samples])
+                Y = np.array([model.lpdf(x, params) for x in samples])
+                penalty = 1e20
+                weighted_neg_lpdf = [-c * y if y != -np.inf else penalty * c for y, c in zip(Y, ch)]
+                output = np.sum(weighted_neg_lpdf)
+
+                return output
 
             def jacobian(params, ch, model: AModelDifferentiable):
                 return -np.sum(

mpest/models/__init__.py

@@ -7,6 +7,7 @@
 )
 from mpest.models.exponential import ExponentialModel
 from mpest.models.gaussian import GaussianModel
+from mpest.models.uniform import Uniform
 from mpest.models.weibull import WeibullModelExp
 from mpest.models.cauchy import Cauchy
 from mpest.models.pareto import Pareto
@@ -19,4 +20,5 @@
     Cauchy().name: Cauchy,
     Pareto().name: Pareto,
     Beta().name: Beta,
+    Uniform().name: Uniform,
 }