# Load data
data_path = "data/count_by_hour_with_header.csv"
data_scaled = reader.get_scaled_mrt_data(data_path, [s],
datetime_features=True,
holidays=HOLIDAYS)
n = data_scaled.shape[0]
x_pretrain, y_pretrain = reader.get_pretrain_data(
data_scaled[0:round(0.6 * n)], 3, config.input_time_steps)
# model.pretrain_encoder(x_pretrain, y_pretrain, 0.0005, 2000)
# model = Seq2Seq(config, False)
train, val, test = reader.produce_seq2seq_data(
data_scaled,
config.train_batch_size,
config.input_time_steps,
config.output_time_steps,
time_major=True,
y_has_features=True)
x_train, y_train = train[0], train[1]
x_val, y_val, = val[0], val[1]
x_test, y_test = test[0], test[1]
targets_train = y_train[:, :, :, [0]]
features_train = y_train[:, :, :, 1:]
targets_val = y_val[:, :, :, [0]]
features_val = y_val[:, :, :, 1:]
targets_test = y_test[:, :, :, [0]]
features_test = y_test[:, :, :, 1:]
# stations = [0]
for s in stations:
config = LSTMConfig(s)
# Build model
lstm_model = Simple_LSTM(config)
# Load data
data_path = "data/count_by_hour_with_header.csv"
data_scaled = reader.get_scaled_mrt_data(data_path, [s],
datetime_features=True,
holidays=HOLIDAYS)
train, val, test = reader.produce_seq2seq_data(
data_scaled,
batch_size=config.train_batch_size,
input_seq_len=config.input_time_steps,
output_seq_len=config.output_time_steps)
x_train, y_train = train[0], train[1]
x_val, y_val, = val[0], val[1]
x_test, y_test = test[0], test[1]
x_val = np.squeeze(x_val, axis=1)
x_test = np.squeeze(x_test, axis=1)
y_train = np.squeeze(y_train, axis=3)
y_val = np.squeeze(y_val, axis=(1, 3))
y_test = np.squeeze(y_test, axis=(1, 3))
# test_time_features = np.squeeze(test_time_features, axis=(1, 2))
# Run training
for s in stations:
print("station %i" % s)
data_path = "../data/count_by_hour_with_header.csv"
checkpoint_file = "../checkpoints/simple_lstm_multiple_steps/lstm-station-%s.pt" % str(
s)
data_scaled = reader.get_scaled_mrt_data(data_path, [s],
datetime_features=True,
holidays=HOLIDAYS)
n = data_scaled.shape[0]
train_data, val_data, test_data = reader.produce_seq2seq_data(
data_scaled,
batch_size,
input_seq_len,
output_seq_len,
time_major=True,
y_has_features=True)
x_train, y_train = train_data[0], train_data[1]
x_val, y_val, = val_data[0], val_data[1]
x_test, y_test = test_data[0], test_data[1]
x_train = torch.from_numpy(x_train).contiguous()
y_train = torch.from_numpy(y_train).contiguous()
x_val = torch.from_numpy(x_val).contiguous()
y_val = torch.from_numpy(y_val).contiguous()
targets_train = y_train[:, :, :, [0]]
features_train = y_train[:, :, :, 1:]