-
Notifications
You must be signed in to change notification settings - Fork 13
/
data_feeder.py
173 lines (132 loc) · 7.21 KB
/
data_feeder.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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
import numpy as np
import pandas as pd
# batch_size: the number of rows fed into the network at once.
# crop: the number of rows in the data set to be used in total.
# chunk_size: the number of lines to read from the file at once.
class TrainSlidingWindowGenerator():
"""Yields features and targets for training a ConvNet.
Parameters:
__file_name (string): The path where the training dataset is located.
__batch_size (int): The size of each batch from the dataset to be processed.
__chunk_size (int): The size of each chunk of data to be processed.
__shuffle (bool): Whether the dataset should be shuffled before being returned.
__offset (int):
__crop (int): The number of rows of the dataset to return.
__skip_rows (int): The number of rows of a dataset to skip before reading data.
__ram_threshold (int): The maximum amount of RAM to utilise at a time.
total_size (int): The number of rows read from the dataset.
"""
def __init__(self,
file_name,
chunk_size,
shuffle,
offset,
batch_size=1000,
crop=100000,
skip_rows=0,
ram_threshold=5 * 10 ** 5):
self.__file_name = file_name
self.__batch_size = batch_size
self.__chunk_size = 10 ** 8
self.__shuffle = shuffle
self.__offset = offset
self.__crop = crop
self.__skip_rows = skip_rows
self.__ram_threshold = ram_threshold
self.total_size = 0
def check_if_chunking(self):
"""Count the number of rows in the dataset and determine whether this is larger than the chunking
threshold or not. """
# Loads the file and counts the number of rows it contains.
print("Importing training file...")
chunks = pd.read_csv(self.__file_name,
header=0,
nrows=self.__crop,
skiprows=self.__skip_rows)
print("Counting number of rows...")
self.total_size = len(chunks)
del chunks
print("Done.")
print("The dataset contains ", self.total_size, " rows")
# Display a warning if there are too many rows to fit in the designated amount RAM.
if (self.total_size > self.__ram_threshold):
print("There is too much data to load into memory, so it will be loaded in chunks. Please note that this may result in decreased training times.")
def load_dataset(self):
"""Yields pairs of features and targets that will be used directly by a neural network for training.
Yields:
input_data (numpy.array): A 1D array of size batch_size containing features of a single input.
output_data (numpy.array): A 1D array of size batch_size containing the target values corresponding to
each feature set.
"""
if self.total_size == 0:
self.check_if_chunking()
# If the data can be loaded in one go, don't skip any rows.
if (self.total_size <= self.__ram_threshold):
# Returns an array of the content from the CSV file.
data_array = np.array(pd.read_csv(self.__file_name, nrows=self.__crop, skiprows=self.__skip_rows, header=0))
inputs = data_array[:, 0]
outputs = data_array[:, 1]
maximum_batch_size = inputs.size - 2 * self.__offset
if self.__batch_size < 0:
self.__batch_size = maximum_batch_size
indicies = np.arange(maximum_batch_size)
if self.__shuffle:
np.random.shuffle(indicies)
while True:
for start_index in range(0, maximum_batch_size, self.__batch_size):
splice = indicies[start_index : start_index + self.__batch_size]
input_data = np.array([inputs[index : index + 2 * self.__offset + 1] for index in splice])
output_data = outputs[splice + self.__offset].reshape(-1, 1)
yield input_data, output_data
# Skip rows where needed to allow data to be loaded properly when there is not enough memory.
if (self.total_size >= self.__ram_threshold):
number_of_chunks = np.arange(self.total_size / self.__chunk_size)
if self.__shuffle:
np.random.shuffle(number_of_chunks)
# Yield the data in sections.
for index in number_of_chunks:
data_array = np.array(pd.read_csv(self.__file_name, skiprows=int(index) * self.__chunk_size, header=0, nrows=self.__crop))
inputs = data_array[:, 0]
outputs = data_array[:, 1]
maximum_batch_size = inputs.size - 2 * self.__offset
if self.__batch_size < 0:
self.__batch_size = maximum_batch_size
indicies = np.arange(maximum_batch_size)
if self.__shuffle:
np.random.shuffle(indicies)
while True:
for start_index in range(0, maximum_batch_size, self.__batch_size):
splice = indicies[start_index : start_index + self.__batch_size]
input_data = np.array([inputs[index : index + 2 * self.__offset + 1] for index in splice])
output_data = outputs[splice + self.__offset].reshape(-1, 1)
yield input_data, output_data
class TestSlidingWindowGenerator(object):
"""Yields features and targets for testing and validating a ConvNet.
Parameters:
__number_of_windows (int): The number of sliding windows to produce.
__offset (int): The offset of the infered value from the sliding window.
__inputs (numpy.ndarray): The available testing / validation features.
__targets (numpy.ndarray): The target values corresponding to __inputs.
__total_size (int): The total number of inputs.
"""
def __init__(self, number_of_windows, inputs, targets, offset):
self.__number_of_windows = number_of_windows
self.__offset = offset
self.__inputs = inputs
self.__targets = targets
self.total_size = len(inputs)
def load_dataset(self):
"""Yields features and targets for testing and validating a ConvNet.
Yields:
input_data (numpy.array): An array of features to test / validate the network with.
"""
self.__inputs = self.__inputs.flatten()
max_number_of_windows = self.__inputs.size - 2 * self.__offset
if self.__number_of_windows < 0:
self.__number_of_windows = max_number_of_windows
indicies = np.arange(max_number_of_windows, dtype=int)
for start_index in range(0, max_number_of_windows, self.__number_of_windows):
splice = indicies[start_index : start_index + self.__number_of_windows]
input_data = np.array([self.__inputs[index : index + 2 * self.__offset + 1] for index in splice])
target_data = self.__targets[splice + self.__offset].reshape(-1, 1)
yield input_data, target_data