env fixes and updates

rl_core/environments/base.py

@@ -42,7 +42,10 @@ def pipeline_fitting_and_evaluating(self) -> float:
         y_pred = pred.predict
         y_true = self.val_data.target
 
-        self.metric_value = self.metric(y_score=y_pred, y_true=y_true)
+        try:
+            self.metric_value = self.metric(y_score=y_pred, y_true=y_true)
+        except:
+            self.metric_value = -0.999
 
         return self.metric_value
 

rl_core/environments/embedding.py

@@ -68,4 +68,4 @@ def _environment_response(self, reward: float, done: bool) -> (int, bool, dict):
             'metric_value': self.metric_value,
         }
 
-        return reward, done, info
+        return reward, done, info

rl_core/environments/linear.py

@@ -1,4 +1,5 @@
 from abc import ABC
+from copy import deepcopy
 
 import numpy as np
 import torch
@@ -33,12 +34,11 @@ def reset(self, **kwargs):
         self.pipeline = PipelineBuilder()
         self.time_step = 0
         self.metric_value = 0
-
         self.position = 0
 
         self.init_state()
 
-        return self.state
+        return deepcopy(self.state)
 
     def _train_step(self, action):
         self.last_action = action
@@ -69,7 +69,7 @@ def _train_step(self, action):
 
         reward, done, info = self._environment_response(reward, done)
 
-        return self.state, reward, done, info
+        return deepcopy(self.state), reward, done, info
 
     def _inference_step(self, action):
         raise NotImplementedError()
@@ -81,4 +81,4 @@ def _environment_response(self, reward: float, done: bool) -> (int, bool, dict):
             'metric_value': self.metric_value,
         }
 
-        return reward, done, info
+        return reward, done, info

test/unit/rl_test/test_environment.py

@@ -1,6 +1,7 @@
 import os.path
 
 import numpy as np
+import torch
 from fedot.core.pipelines.pipeline import Pipeline
 from fedot.core.repository.operation_types_repository import OperationTypesRepository
 from fedot.core.repository.tasks import TaskTypesEnum
@@ -136,4 +137,65 @@ def test_ensemble_pipeline_environment_pipeline_build():
     assert isinstance(info['pipeline'], Pipeline)
     assert isinstance(info['time_step'], int)
     assert isinstance(info['metric_value'], float)
-    assert info['metric_value'] > 0.5
+    assert info['metric_value'] > 0.5
+
+
+def test_linear_pipeline_environment_pipeline_reset():
+    primitives = OperationTypesRepository('all').suitable_operation(
+        task_type=TaskTypesEnum.classification)
+
+    state_dim = 2
+    env = LinearPipelineGenerationEnvironment(state_dim=state_dim, primitives=primitives)
+
+    path = os.path.join(str(project_root()), 'MetaFEDOT/rl_core/data/scoring_train.csv')
+
+    datasets = {
+        'scoring': path,
+    }
+
+    dataloader = DataLoader(datasets)
+
+    train_data, val_data = dataloader.get_data()
+    env.load_data(train_data, val_data)
+
+    actions = [env.primitives.index(p) for p in ['rf', 'dt', 'eop']]
+
+    for action in actions:
+        new_state, r, done, info = env.step(action)
+
+    env.reset()
+
+    assert torch.all(env.state == torch.tensor(np.array([0, 0]), dtype=torch.float64))
+
+    for action in actions:
+        new_state, r, done, info = env.step(action)
+
+    assert torch.all(new_state == torch.tensor(np.array([8, 2]), dtype=torch.float64))
+
+
+def test_linear_pipeline_environment_state():
+    primitives = OperationTypesRepository('all').suitable_operation(
+        task_type=TaskTypesEnum.classification)
+
+    state_dim = 4
+    env = LinearPipelineGenerationEnvironment(state_dim=state_dim, primitives=primitives)
+
+    path = os.path.join(str(project_root()), 'MetaFEDOT/rl_core/data/scoring_train.csv')
+
+    datasets = {
+        'scoring': path,
+    }
+
+    dataloader = DataLoader(datasets)
+
+    train_data, val_data = dataloader.get_data()
+    env.load_data(train_data, val_data)
+
+    actions = [env.primitives.index(p) for p in ['knn', 'dt', 'scaling', 'rf']]
+    states = []
+
+    for action in actions:
+        new_state, r, done, info = env.step(action)
+        states.append(new_state.numpy())
+
+    assert np.unique(states, axis=1).shape[1] == len(states)