def test_load_from_dt_tau_params(tf_graph):
reset_seeds(19846)
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
rnn_model = Basic(params)
weights = rnn_model.get_weights()
x, _, _, _ = mtc.get_trial_batch()
sim_model = BasicSimulator(params={
'dt': params['dt'],
'tau': params['tau']
},
weights=weights)
tf_output, tf_state = rnn_model.test(x)
sim_output, sim_state = sim_model.run_trials(x)
assert (tf_output.shape == sim_output.shape)
assert (tf_state.shape == sim_state.shape)
assert (np.allclose(tf_state, sim_state, atol=1e-06))
assert (np.allclose(tf_output, sim_output, atol=1e-06))
rnn_model.destruct()
def test_lstm_simulator_load_from_rnn_model(tf_graph):
reset_seeds(19846)
tf.compat.v1.keras.backend.set_floatx('float64')
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
rnn_model = LSTM(params)
x, _, _, _ = mtc.get_trial_batch()
sim_model = LSTMSimulator(rnn_model=rnn_model)
tf_output, tf_state = rnn_model.test(x)
sim_output, sim_state = sim_model.run_trials(x)
assert (tf_output.shape == sim_output.shape)
assert (tf_state.shape == sim_state.shape)
assert (np.allclose(tf_state, sim_state, atol=1e-01, rtol=1e-01))
assert (np.allclose(tf_output, sim_output, atol=1e-01, rtol=1e-01))
rnn_model.destruct()
rnn_model = LSTM(params)
tf_output, tf_state = rnn_model.test(x)
assert (not np.allclose(tf_state, sim_state, atol=1e-01, rtol=1e-01))
assert (not np.allclose(tf_output, sim_output, atol=1e-01, rtol=1e-01))
def test_load_from_file(tf_graph, tmpdir, capfd):
reset_seeds(19846)
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
rnn_model = Basic(params)
rnn_model.save(str(tmpdir.dirpath("save_weights.npz")))
x, _, _, _ = mtc.get_trial_batch()
sim_model = BasicSimulator(params=params,
weights_path=str(
tmpdir.dirpath("save_weights.npz")))
tmpdir.dirpath("save_weights.npz").remove()
tf_output, tf_state = rnn_model.test(x)
sim_output, sim_state = sim_model.run_trials(x)
assert (tf_output.shape == sim_output.shape)
assert (tf_state.shape == sim_state.shape)
assert (np.allclose(tf_state, sim_state, atol=1e-06))
assert (np.allclose(tf_output, sim_output, atol=1e-06))
rnn_model.destruct()
def test_rec_noise_params(tf_graph):
reset_seeds(19846)
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
params['rec_noise'] = .1
rnn_model = Basic(params)
weights = rnn_model.get_weights()
x, _, _, _ = mtc.get_trial_batch()
sim_model = BasicSimulator(params=params, weights=weights)
assert (sim_model.rec_noise == params['rec_noise'])
rnn_model.destruct()
def test_warnings(tf_graph, tmpdir, capfd):
reset_seeds(19846)
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
params['rec_noise'] = .1
rnn_model = Basic(params)
weights = rnn_model.get_weights()
rnn_model.save(str(tmpdir.dirpath("save_weights.npz")))
x, _, _, _ = mtc.get_trial_batch()
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(params=params, rnn_model=rnn_model)
assert 'rnn_model takes precedence' in str(excinfo.value)
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(weights=weights, rnn_model=rnn_model)
assert 'Weights from rnn_model and weights_path will be ignored' in str(
excinfo.value)
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(weights_path=str(
tmpdir.dirpath("save_weights.npz")),
weights=weights,
params=params)
assert 'Weights from rnn_model and weights_path will be ignored' in str(
excinfo.value)
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(weights_path=str(
tmpdir.dirpath("save_weights.npz")),
rnn_model=rnn_model)
assert 'Weights from rnn_model will be ignored.' in str(excinfo.value)
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(params=params)
assert 'Either weights, rnn_model, or weights_path must be passed in.' in str(
excinfo.value)
rnn_model.destruct()
tmpdir.dirpath("save_weights.npz").remove()
def test_transfer_function(tf_graph):
def my_relu(X):
return np.maximum(X, 0)
reset_seeds(19846)
mtc = MatchToCategory(dt=10, tau=100, T=2000, N_batch=50)
params = mtc.get_task_params()
params['name'] = 'test'
params['N_rec'] = 49
rnn_model = Basic(params)
x, _, _, _ = mtc.get_trial_batch()
with pytest.raises(UserWarning) as excinfo:
sim_model = BasicSimulator(rnn_model=rnn_model,
transfer_function=my_relu)
assert 'my_relu' in str(excinfo.value)
rnn_model.destruct()