def test_givenSeq2SeqModelRetrained_whenLoadRetrainedWeights_thenLoadProperly(
self, torch_nn_mock, torch_mock):
all_layers_params_mock = MagicMock()
all_layers_params_mock.__getitem__(
).__len__.return_value = self.decoder_output_size
torch_mock.load.return_value = all_layers_params_mock
seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
verbose=True,
)
seq2seq_model._load_weights(self.a_fake_retrain_path)
torch_mock.assert_has_calls([
call.load(self.a_fake_retrain_path, map_location=self.a_cpu_device)
])
torch_nn_mock.assert_called()
torch_nn_mock.asser_has_calls([call(all_layers_params_mock)])
def test_whenInstantiateASeq2SeqModelGPU_thenParametersAreOk(self):
seq2seq_model = Seq2SeqModel(
self.a_torch_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
)
self.assertEqual(self.a_torch_device, seq2seq_model.device)
self.assertEqual(self.encoder_input_size_dim,
seq2seq_model.encoder.lstm.input_size)
self.assertEqual(self.encoder_hidden_size,
seq2seq_model.encoder.lstm.hidden_size)
self.assertEqual(self.encoder_num_layers,
seq2seq_model.encoder.lstm.num_layers)
self.assertEqual(self.a_torch_device,
seq2seq_model.encoder.lstm.all_weights[0][0].device)
self.assertEqual(self.decoder_input_size_dim,
seq2seq_model.decoder.lstm.input_size)
self.assertEqual(self.decoder_hidden_size,
seq2seq_model.decoder.lstm.hidden_size)
self.assertEqual(self.decoder_num_layers,
seq2seq_model.decoder.lstm.num_layers)
self.assertEqual(self.decoder_output_size,
seq2seq_model.decoder.linear.out_features)
self.assertEqual(self.a_torch_device,
seq2seq_model.decoder.lstm.all_weights[0][0].device)
def setUp(self) -> None:
super().setUp()
self.pre_trained_seq2seq_model = Seq2SeqModel(self.a_torch_device)
self.encoder_input_setUp(
"fasttext"
) # fasttext since the simplest case (bpemb use a embedding layer)
self.none_target = None # No target (for teacher forcing)
self.a_value_greater_than_threshold = 0.1
def test_whenSameOutput_thenReturnTrue(self):
seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
)
self.assertTrue(seq2seq_model.same_output_dim(
self.decoder_output_size))
def test_givenSeq2seqModel_whenLoadPreTrainedWeightsVerboseCPU_thenWarningsRaised(
self, torch_nn_mock, torch_mock, isfile_mock, last_version_mock):
seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
verbose=True,
)
isfile_mock.return_value = True
last_version_mock.return_value = False
with patch("deepparse.network.seq2seq.download_weights"):
with self.assertWarns(UserWarning):
seq2seq_model._load_pre_trained_weights("a_model_type")
def setUp(self) -> None:
super().setUp()
self.pre_trained_seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.input_size,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.num_layers,
output_size=self.output_size,
)
self.encoder_input_setUp(
"fasttext", self.a_cpu_device
) # fasttext since the simplest case (bpemb use a embedding layer)
self.none_target = None # No target (for teacher forcing)
self.a_value_greater_than_threshold = 0.1
self.a_target_vector = torch.tensor(
[[0, 1, 1, 4, 5, 8], [1, 0, 3, 8, 0, 0]], device=self.a_cpu_device)
def test_givenSeq2SeqModelRetrained_whenLoadRetrainedWeightsNewTagModel_thenLoadProperDict(
self, torch_nn_mock, torch_mock):
all_layers_params_mock = MagicMock(spec=dict)
all_layers_params_mock.__getitem__(
).__len__.return_value = self.decoder_output_size
torch_mock.load.return_value = all_layers_params_mock
seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
verbose=True,
)
seq2seq_model._load_weights(self.a_fake_retrain_path)
all_layers_params_mock.get.assert_called()
def test_whenHandleNewOutputDim_thenProperlyHandleNewDim(self):
seq2seq_model = Seq2SeqModel(
self.a_cpu_device,
input_size=self.encoder_input_size_dim,
encoder_hidden_size=self.encoder_hidden_size,
encoder_num_layers=self.encoder_num_layers,
decoder_hidden_size=self.decoder_hidden_size,
decoder_num_layers=self.decoder_num_layers,
output_size=self.decoder_output_size,
)
a_new_dim = 1
seq2seq_model.handle_new_output_dim(a_new_dim)
expected = a_new_dim
actual = seq2seq_model.output_size
self.assertEqual(expected, actual)
actual = seq2seq_model.decoder.linear.out_features
self.assertEqual(expected, actual)