main.py

from sklearn.model_selection import train_test_split

from src.data_sequences import training_sequence
from src.dataset import read_datasets, notes_to_dataset, dataset_to_notes, combine_datasets
from src.generate import generate_notes
from src.midi_func import notes_to_midi
from src.model import get_model
from src.utils import get_midi_filenames

VOCAB_SIZE = 128
SEQ_LEN = 40
RESOLUTION = 480
TEMPO = 172

# When set to false, load scaler and model from files, otherwise fit everything again
# Make sure to set it to True if you have changed the training data.
INIT = True
MODEL_NAME = "h5_models/mozart_4.h5"
OUTPUT_FILE_NAME = "./output/mozart_4.mid"

if __name__ == '__main__':
    filenames = get_midi_filenames(main_dir='samples', subdirs=['Swing Beats'])

    all_notes = read_datasets(filepaths=filenames)
    dataset, unique_vals = notes_to_dataset(all_notes)
    col_indexes = dataset.columns

    print(f"unique vals: {unique_vals.__repr__()}")

    notes = dataset_to_notes(dataset)
    notes_to_midi(notes, instrument_name="Acoustic Grand Piano", resolution=RESOLUTION, tempo=TEMPO,
                  out_file="./output/export test.mid")

    input_len = 25

    xs, ys, scaler = training_sequence(dataset, input_len=input_len, output_len=1, init_scaler=INIT,
                                       scaler_path=f"{MODEL_NAME}_scaler.pkl")

    [n_rows_input, n_cols_input] = xs[0].shape
    [n_rows_output, n_cols_output] = ys[0].shape

    x_train, x_val, y_train, y_val = train_test_split(xs, ys, test_size=0.3)

    model = get_model(xs=x_train, ys=y_train, validation_data=(x_val, y_val), init=INIT, model_name=MODEL_NAME)

    gen_dataset = generate_notes(model=model, col_indexes=col_indexes, scaler=scaler, sample_notes=x_val[0],
                                 n_generate=1000, unique_vals=unique_vals, diff_fix_factor=0.8, vel_fix_factor=0.5,
                                 crop_training=True, staccato_factor=0.7)

    gen_notes = dataset_to_notes(gen_dataset)
    notes_to_midi(notes=gen_notes, instrument_name="Acoustic Grand Piano", out_file=OUTPUT_FILE_NAME,
                  resolution=RESOLUTION, tempo=TEMPO)

    pass