Create testing.py

reinforcement_learning/testing.py

@@ -0,0 +1,83 @@
+"""Unit testing
+
+To run unit testing:
+
+python -m unittest test_dqn_helper_functions
+
+"""
+
+import unittest
+import numpy as np
+from dqn_helper_functions import ReplayBuffer
+
+
+class TestReplayBuffer(unittest.TestCase):
+    def test_type(self):
+        # Make sure type erors are raised when necessary
+        self.assertRaises(TypeError, ReplayBuffer, (1.0, 20, 3, True))
+        self.assertRaises(TypeError, ReplayBuffer, (1, 2.0, 3, True))
+        self.assertRaises(TypeError, ReplayBuffer, (1, 2, 3.0, True))
+        self.assertRaises(TypeError, ReplayBuffer, (1, 2, 3, 'value'))
+        
+    def test_collect_experience(self):
+        # test a sample of collect_experience
+        max_size = 2
+        input_shape = 20
+        n_actions = 3
+        discrete = False
+        
+        memory = ReplayBuffer(max_size, input_shape, n_actions, discrete)
+        
+        state = np.random.rand(input_shape)
+        action = np.random.rand(n_actions)
+        reward = np.random.randint(10)
+        state_= np.random.rand(input_shape)
+        done = False
+        
+        memory.collect_experience(state, action, reward, state_, done)
+        
+        self.assertEqual(memory.mem_cntr, 1)
+        np.testing.assert_array_equal(memory.terminal_memory, [1.,0.])
+        np.testing.assert_array_equal(memory.reward_memory, [reward, 0])
+        np.testing.assert_array_equal(memory.action_memory, np.array([action, np.zeros(n_actions)], 
+                                                                     dtype=np.float32))
+        np.testing.assert_array_equal(memory.new_observation_memory, np.array([state_, np.zeros(input_shape)]))
+        np.testing.assert_array_equal(memory.observation_memory, np.array([state, np.zeros(input_shape)]))
+        
+    def test_sample_buffer(self):
+        # test a sample of collect_experience
+        max_size = 2
+        input_shape = 20
+        n_actions = 3
+        discrete = False
+        batch_size = 2
+        # create memory
+        memory = ReplayBuffer(max_size, input_shape, n_actions, discrete)
+        # create dummy environment
+        state = np.random.rand(input_shape)
+        action = np.random.rand(n_actions)
+        reward = np.random.randint(10)
+        state_ = np.random.rand(input_shape)
+        done = False
+        # perform 2 collection
+        memory.collect_experience(state, action, reward, state_, done)
+        memory.collect_experience(state, action, reward, state_, done)
+        # get sample
+        observations, actions, rewards, new_observations, terminal = memory.sample_buffer(batch_size)
+        # test observations
+        np.testing.assert_array_equal(observations, np.array([state, state]))
+        # test observations
+        np.testing.assert_array_equal(new_observations, np.array([state_, state_]))
+        # test actions
+        np.testing.assert_array_equal(actions, np.array([action, action], dtype=np.float32))
+        # test rewards
+        np.testing.assert_array_equal(rewards, np.array([reward, reward]))
+        # test finished episode
+        np.testing.assert_array_equal(terminal, np.array([1, 1], dtype=np.float32))
+        
+        
+class TestDqnAgent(unittest.TestCase):
+    
+     def test_type(self):
+        # Make sure type erors are raised when necessary
+        pass