hyperparam_search.py

import itertools

epochs = 3  # number of training epochs
test_batch_size = 32  # batch size for testing
arrays = [
    embedding_size,
    dropout,
    filters,
    kernel_size,
    pool_size,
    lstm_output_size,
    batch_size,
]  # all hyper-params

for ed, d, flt, ks, ps, ls, bs in itertools.product(*arrays):
    model = make_model(
        embedding_dim=ed,
        dropout=d,
        filters=flt,
        kernel_size=ks,
        pool_size=ps,
        lstm_output_size=ls,
        metrics=METRICS,
        vocab_size=max_features,
        maxlen=maxlen,
    )
    h = model.fit(x_train, y_train, batch_size=bs, epochs=epochs, verbose=2)
    train_loss = h.history["loss"][-1]
    test_metrics = model.evaluate(x=x_test, y=y_test, batch_size=test_batch_size)
    test_loss, test_acc, test_prec, test_rec = test_metrics

    # write everything to external JSON file