CS236-Project/sample_unconditioned_model.py at master · schlagercollin/CS236-Project · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
"""
File to sample midi files from a provided unconditioned model.
(e.g. the bass model alone or an unconditioned bass/melody bassline)

Use the --logdir flag to point to a directory that houses
checkpoints for the model. If --logdir points to a bass-track model,
you can optionally include --logdir2 pointing to a unconditioned
melody model to generate a simultaneous, unconditioned sample
of the melody and bass track (which serves as a baseline.)

Note that --e_dim and --h_dim should
match the parameters used by the model that is being loaded. --bass_temp
and --melody_temp control the temperature of the sample. A temperature of
zero corresponds to greedy sampling. If --k is provided, only the top k
logits are made available for temperature-based sampling.
"""

import os
import time
import glob
import torch
import argparse
import numpy as np
import music21 as m21
import tqdm
from lstm import UnconditionalLSTM
from data_utils import decode, open_file

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--logdir', type=str, default='logs/example_trained_bass')
    parser.add_argument('--logdir2', type=str, default='logs/example_trained_unconditional_melody')
    parser.add_argument('--condition', type=int, nargs='+', required=False, default=[60, 8, 8])
    parser.add_argument('--condition2', type=int, nargs='+', default=[36, 8, 8])
    parser.add_argument('--ckp', type=int, required=False)
    parser.add_argument('--e_dim', type=int, default=200)
    parser.add_argument('--h_dim', type=int, default=400)
    parser.add_argument('--sample_len', type=int, default=240)
    parser.add_argument('--sample_len2', type=int, default=240)
    parser.add_argument('--k', type=int, default=40)
    parser.add_argument('--temp', type=float, default=0.8)
    parser.add_argument('--num_samples', type=int, default=1)

    # NOTE: if --temp == 0, then we perform greedy generation

    args = parser.parse_args()

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')


    # GENERATE SAMPLE FROM FIRST LSTM
    lstm = UnconditionalLSTM(embed_dim=args.e_dim, hidden_dim=args.h_dim, log_level=0)
    lstm = lstm.to(device)

    # if specified, get specific checkpoint
    checkpoint_dir = os.path.join(args.logdir, 'checkpoints')
    if args.ckp:
        full_path = os.path.join(checkpoint_dir, 'model_checkpoint_step_{}.pt'.format(args.ckp))
        num_steps = args.ckp

    # otherwise, get the last checkpoint (alphanumerically sorted)
    else:
        checkpoints = glob.glob(os.path.join(checkpoint_dir, "*.pt"))

        # model_checkpoint_step_<step_number>.pt --> <step_number>
        step_numbers = np.array(list(map(lambda x: int(x.split(".")[0].split("_")[-1]), checkpoints)))

        sort_order = np.argsort(step_numbers)
        num_steps = step_numbers[sort_order[-1]]

        # gets the checkpoint path with the greatest number of steps
        last_checkpoint_path = checkpoints[sort_order[-1]]
        full_path = last_checkpoint_path

    print("Loading model weights from {}...".format(full_path))
    lstm.load_state_dict(torch.load(full_path, map_location=device))

    # IF SPECIFIED, SAMPLE FROM SECOND LSTM
    if args.logdir2 is not '':
        lstm2 = UnconditionalLSTM(embed_dim=args.e_dim, hidden_dim=args.h_dim, log_level=0)
        lstm2 = lstm2.to(device)

        # if specified, get specific checkpoint
        checkpoint_dir = os.path.join(args.logdir2, 'checkpoints')
        if args.ckp:
            full_path = os.path.join(checkpoint_dir, 'model_checkpoint_step_{}.pt'.format(args.ckp))
            num_steps = args.ckp

        # otherwise, get the last checkpoint (alphanumerically sorted)
        else:
            checkpoints = glob.glob(os.path.join(checkpoint_dir, "*.pt"))

            # model_checkpoint_step_<step_number>.pt --> <step_number>
            step_numbers = np.array(list(map(lambda x: int(x.split(".")[0].split("_")[-1]), checkpoints)))

            sort_order = np.argsort(step_numbers)
            num_steps = step_numbers[sort_order[-1]]

            # gets the checkpoint path with the greatest number of steps
            last_checkpoint_path = checkpoints[sort_order[-1]]
            full_path = last_checkpoint_path

        print("Loading model weights from {}...".format(full_path))
        lstm2.load_state_dict(torch.load(full_path, map_location=device))

    for i in tqdm.tqdm(range(args.num_samples)):
        generation = lstm.generate(condition=args.condition, k=None, length=args.sample_len, temperature=args.temp)
        stream = decode(generation)

        if args.logdir2 is not '':

            generation2 = lstm2.generate(condition=args.condition, k=None, length=args.sample_len2,
                                         temperature=args.temp)
            stream2 = decode(generation2)

            # COMBINE THE SAMPLES
            combined_stream = m21.stream.Stream()
            first_part = m21.stream.Part(id='1')
            first_part.append(stream)
            second_part = m21.stream.Part(id='2')
            second_part.append(stream2)

            combined_stream.insert(0, first_part)
            combined_stream.insert(0, second_part)

        else:
            combined_stream = stream

        # print("Finished generating!")
        # stream.show('midi')

        # write_dir = os.path.join(args.logdir, 'eval_samples')
        # if not os.path.exists(write_dir):
        #     os.mkdir(write_dir)
        #
        # num_eval_samples = len(glob.glob(os.path.join(write_dir, 'eval_sample*')))
        # output_name = os.path.join(write_dir,
        #                            'eval_sample_checkpoint_{}_temp_{}_num_{}.mid'.format(str(num_steps),
        #                                                                              str(args.temp).replace('.', '-'),
        #                                                                              num_eval_samples))

        sample_dir = './generated_samples/unconditional'
        bass_sample_dir = "{}_{}_bass.mid".format(sample_dir, len(glob.glob(sample_dir + "*")))
        melody_sample_dir = "{}_{}_melody.mid".format(sample_dir, len(glob.glob(sample_dir + "*")))
        combined_sample_dir = "{}_{}.mid".format(sample_dir, len(glob.glob(sample_dir+"*")))

        # print("Writing sample to {}...".format(sample_dir))
        combined_stream.write('midi', fp=combined_sample_dir)
        stream.write('midi', fp=bass_sample_dir)
        stream2.write('midi', fp=melody_sample_dir)

    open_file(combined_sample_dir)