def test_s2s_predict_save_restore(self):
model = Seq2SeqPytorch()
config = {"batch_size": 128}
model.fit_eval(self.train_data[0], self.train_data[1], self.val_data,
**config)
pred = model.predict(self.test_data[0])
assert pred.shape == self.test_data[1].shape
with tempfile.TemporaryDirectory() as tmp_dir_name:
ckpt_name = os.path.join(tmp_dir_name, "ckpt")
model.save(ckpt_name)
model_1 = Seq2SeqPytorch()
model_1.restore(ckpt_name)
pred_1 = model_1.predict(self.test_data[0])
assert np.allclose(pred, pred_1)
def test_s2s_teacher_forcing_fit_evaluate(self):
model = Seq2SeqPytorch()
config = {"batch_size": 128, "teacher_forcing": True}
model.fit_eval(self.train_data[0], self.train_data[1], self.val_data,
**config)
mse, smape = model.evaluate(self.val_data[0],
self.val_data[1],
metrics=["mse", "smape"])
assert len(
mse) == self.val_data[1].shape[-1] * self.val_data[1].shape[-2]
assert len(
smape) == self.val_data[1].shape[-1] * self.val_data[1].shape[-2]
def __init__(
self,
input_feature_num,
future_seq_len,
output_feature_num,
lstm_hidden_dim=128,
lstm_layer_num=1,
teacher_forcing=False,
dropout=0.25,
lr=0.001,
loss="mse",
optimizer="Adam",
):
"""
Build a LSTM Sequence to Sequence Forecast Model.
:param future_seq_len: Specify the output time steps (i.e. horizon).
:param input_feature_num: Specify the feature dimension.
:param output_feature_num: Specify the output dimension.
:param lstm_hidden_dim: LSTM hidden channel for decoder and encoder.
:param lstm_layer_num: LSTM layer number for decoder and encoder.
:param teacher_forcing: If use teacher forcing in training.
:param dropout: Specify the dropout close possibility (i.e. the close
possibility to a neuron). This value defaults to 0.25.
:param optimizer: Specify the optimizer used for training. This value
defaults to "Adam".
:param loss: Specify the loss function used for training. This value
defaults to "mse". You can choose from "mse", "mae" and
"huber_loss".
:param lr: Specify the learning rate. This value defaults to 0.001.
"""
self.check_optional_config = False
self.model_config = {
"input_feature_num": input_feature_num,
"future_seq_len": future_seq_len,
"output_feature_num": output_feature_num,
"lstm_hidden_dim": lstm_hidden_dim,
"lstm_layer_num": lstm_layer_num,
"teacher_forcing": teacher_forcing,
"dropout": dropout,
"lr": lr,
"loss": loss,
"optimizer": optimizer,
}
self.internal = Seq2SeqPytorch(check_optional_config=False)