-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_loader.py
139 lines (112 loc) · 4.53 KB
/
data_loader.py
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
import numpy as np
import pandas as pd
import tensorflow as tf
class DataLoader():
"""
Load the CSV file as a dataframe and do some preprocessing before
return it as tf.Dataset.
Args:
input_path: str, path to the csv file.
training_ratio: float, ratio for training validation split.
Returns:
raw_train_ds, raw_val_ds: tf.Dataset typed of train and val set.
"""
def __init__(self, input_path, training_ratio=0.7):
self.input_path = input_path
self.training_ratio = training_ratio
self.grnas = np.array([])
def load_test_set(self):
df = pd.read_csv(self.input_path)
X = df[['rna', 'grna']].values
# Fix the label into 2 x unique grna (cut and non cut)
self.grnas = np.unique(X[:, 1])
if X.shape[0]:
self.rna_length = len(X[0][0])
self.grna_length = len(X[0][1])
# Tokenize every nucleotides
tokenize = lambda x: [np.array([c for c in x[0]]),
np.array([c for c in x[1]])]
self.test = np.apply_along_axis(tokenize, 1, X)
raw_test_ds = tf.data.Dataset.from_generator(
self.gen_test,
output_signature=(
tf.RaggedTensorSpec(shape=(2, None), dtype=tf.string)
)
)
return raw_test_ds
def load(self):
df = pd.read_csv(self.input_path)
dataset = df[['rna', 'grna', 'label']].values
np.random.seed(42)
np.random.shuffle(dataset)
# Fix the label into 2 x unique grna (cut and non cut)
self.grnas = np.unique(dataset[:, 1])
dataset[:, 2] = np.array([
(np.where(self.grnas == grna)[0][0] * 2 + label) \
for grna, label in zip(dataset[:, 1], dataset[:, -1])
])
X = dataset[:, :2]
y = dataset[:, -1]
if X.shape[0]:
self.rna_length = len(X[0][0])
self.grna_length = len(X[0][1])
# Tokenize every nucleotides
tokenize = lambda x: [np.array([c for c in x[0]]),
np.array([c for c in x[1]])]
X = np.apply_along_axis(tokenize, 1, X)
X_train = X[:int(self.training_ratio * X.shape[0])]
X_val = X[int(self.training_ratio * X.shape[0]):]
y_train = y[:int(self.training_ratio * y.shape[0])]
y_val = y[int(self.training_ratio * y.shape[0]):]
self.train = np.concatenate(
(X_train, np.expand_dims(y_train, axis=1)),
axis=1,
)
self.val = np.concatenate(
(X_val, np.expand_dims(y_val, axis=1)),
axis=1,
)
raw_train_ds = tf.data.Dataset.from_generator(
self.gen_train,
output_signature=(
tf.RaggedTensorSpec(shape=(2, None), dtype=tf.string),
tf.TensorSpec(shape=(), dtype=tf.float32),
)
)
raw_val_ds = tf.data.Dataset.from_generator(
self.gen_val,
output_signature=(
tf.RaggedTensorSpec(shape=(2, None), dtype=tf.string),
tf.TensorSpec(shape=(), dtype=tf.float32),
)
)
return raw_train_ds, raw_val_ds
def get_rna_length(self):
if self.rna_length == None:
print("Error: rna_length is missing. Run DataLoader.load() to generate it.")
return self.rna_length
def get_grna_length(self):
if self.grna_length == None:
print("Error: grna_length is missing. Run DataLoader.load() to generate it.")
return self.grna_length
def get_num_labels(self):
if not len(self.grnas):
print("Error: grnas is missing. Run DataLoader.load() to generate it.")
return len(self.grnas) * 2
# Output it as a ragged type, because we want to
# preprocess it through tf.Dataset map built-in.
def gen_train(self):
np.random.shuffle(self.train)
for idx, row in enumerate(self.train):
feature = tf.ragged.constant([row[0].tolist(), row[1].tolist()])
yield feature, row[2]
def gen_val(self):
np.random.shuffle(self.val)
for idx, row in enumerate(self.val):
feature = tf.ragged.constant([row[0].tolist(), row[1].tolist()])
yield feature, row[2]
def gen_test(self):
np.random.shuffle(self.test)
for idx, row in enumerate(self.test):
feature = tf.ragged.constant([row[0].tolist(), row[1].tolist()])
yield feature