def objective_func(batch_size,
lookback,
lr,
lstm_units,
lstm_act='relu',
cnn_act='relu',
filters=64):
n_epochs = 5000
data_config, nn_config, args, intervals, verbosity = make_model(
int(batch_size), int(lookback), n_epochs, lr, lstm_act, cnn_act,
lstm_units, filters)
model = Model(data_config=data_config,
nn_config=nn_config,
args=args,
intervals=intervals,
verbosity=verbosity)
model.build_nn()
model.train_nn()
mse = np.min(model.losses['val_losses']['mse'])
reset_graph()
return mse
def objective_fn(**kwargs):
data_config, nn_config, total_intervals = make_model(**kwargs)
df = pd.read_csv('data/nk_data.csv')
model = Model(
data_config=data_config,
nn_config=nn_config,
data=df,
# intervals=total_intervals
)
model.build_nn()
idx = np.arange(720)
tr_idx, test_idx = train_test_split(idx, test_size=0.5, random_state=313)
history = model.train_nn(indices=list(tr_idx))
return np.min(history['val_loss'])
