def test_model_predict_wrong_shape():
model = dqn.ModelWrapper(state_axes=1,
action_size=2,
batch_size=3,
model=small_model)
with pytest.raises(ValueError):
model.predict(np.zeros((1, )))
def test_model_predict():
model = dqn.ModelWrapper(state_axes=1,
action_size=2,
batch_size=3,
model=small_model)
result = model.predict(np.zeros((1, 3)))
assert result.shape == (2, 3)
def test_model_predict_single_shape():
model = dqn.ModelWrapper(state_axes=1,
action_size=2,
batch_size=3,
model=small_model)
# missing axis is assumed to be a single length batch access
result = model.predict_single(np.zeros((1, )))
assert result.shape == (2, )
def test_model_predict_different():
model = dqn.ModelWrapper(state_axes=1,
action_size=2,
batch_size=3,
model=small_model)
result_before = model.predict(np.zeros((1, 3)))
result_before = np.copy(result_before)
result_after = model.predict(np.ones((1, 3)))
assert np.any(np.not_equal(result_before, result_after))
def test_model_train():
model = dqn.ModelWrapper(state_axes=1,
action_size=2,
batch_size=3,
model=small_model)
result_before = model.predict(np.ones((1, 3)))
result_before = np.copy(result_before)
for _ in range(100):
model.train(np.zeros((1, 1)), np.ones((2, 1)))
result_after = model.predict(np.ones((1, 3)))
assert np.any(np.not_equal(result_before, result_after))