CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutSign UpSign In
amanchadha

Real-time collaboration for Jupyter Notebooks, Linux Terminals, LaTeX, VS Code, R IDE, and more,
all in one place. Commercial Alternative to JupyterHub.

GitHub Repository: amanchadha/coursera-deep-learning-specialization
 Path: blob/master/C5 - Sequence Models/Week 1/Jazz improvisation with LSTM/music_utils.py
Views: 4819
1
from __future__ import print_function

2
import tensorflow as tf

3
#import tensorflow.keras.backend as K

4
from tensorflow.keras.layers import RepeatVector

5
import sys

6
from music21 import *

7
import numpy as np

8
from grammar import *

9
from preprocess import *

10
from qa import *

11


12


13
def data_processing(corpus, values_indices, m = 60, Tx = 30):

14
    # cut the corpus into semi-redundant sequences of Tx values

15
    Tx = Tx 

16
    N_values = len(set(corpus))

17
    np.random.seed(0)

18
    X = np.zeros((m, Tx, N_values), dtype=np.bool)

19
    Y = np.zeros((m, Tx, N_values), dtype=np.bool)

20
    for i in range(m):

21
#         for t in range(1, Tx):

22
        random_idx = np.random.choice(len(corpus) - Tx)

23
        corp_data = corpus[random_idx:(random_idx + Tx)]

24
        for j in range(Tx):

25
            idx = values_indices[corp_data[j]]

26
            if j != 0:

27
                X[i, j, idx] = 1

28
                Y[i, j-1, idx] = 1

29
    

30
    Y = np.swapaxes(Y,0,1)

31
    Y = Y.tolist()

32
    return np.asarray(X), np.asarray(Y), N_values 

33


34
def next_value_processing(model, next_value, x, predict_and_sample, indices_values, abstract_grammars, duration, max_tries = 1000, temperature = 0.5):

35
    """

36
    Helper function to fix the first value.

37
    

38
    Arguments:

39
    next_value -- predicted and sampled value, index between 0 and 77

40
    x -- numpy-array, one-hot encoding of next_value

41
    predict_and_sample -- predict function

42
    indices_values -- a python dictionary mapping indices (0-77) into their corresponding unique value (ex: A,0.250,< m2,P-4 >)

43
    abstract_grammars -- list of grammars, on element can be: 'S,0.250,<m2,P-4> C,0.250,<P4,m-2> A,0.250,<P4,m-2>'

44
    duration -- scalar, index of the loop in the parent function

45
    max_tries -- Maximum numbers of time trying to fix the value

46
    

47
    Returns:

48
    next_value -- process predicted value

49
    """

50


51
    # fix first note: must not have < > and not be a rest

52
    if (duration < 0.00001):

53
        tries = 0

54
        while (next_value.split(',')[0] == 'R' or 

55
            len(next_value.split(',')) != 2):

56
            # give up after 1000 tries; random from input's first notes

57
            if tries >= max_tries:

58
                #print('Gave up on first note generation after', max_tries, 'tries')

59
                # np.random is exclusive to high

60
                rand = np.random.randint(0, len(abstract_grammars))

61
                next_value = abstract_grammars[rand].split(' ')[0]

62
            else:

63
                next_value = predict_and_sample(model, x, indices_values, temperature)

64


65
            tries += 1

66
            

67
    return next_value

68


69


70
def sequence_to_matrix(sequence, values_indices):

71
    """

72
    Convert a sequence (slice of the corpus) into a matrix (numpy) of one-hot vectors corresponding 

73
    to indices in values_indices

74
    

75
    Arguments:

76
    sequence -- python list

77
    

78
    Returns:

79
    x -- numpy-array of one-hot vectors 

80
    """

81
    sequence_len = len(sequence)

82
    x = np.zeros((1, sequence_len, len(values_indices)))

83
    for t, value in enumerate(sequence):

84
        if (not value in values_indices): print(value)

85
        x[0, t, values_indices[value]] = 1.

86
    return x

87


88