스터디파이에서 ‘딥러닝으로 주식/코인 가격 예측하기’를 참여하며 공부한 내용을 정리한 것입니다..
학습 자료 :
1. Introduction to Deep Learning - Deep Learning basics with Python, TensorFlow and Keras p.1
2. [Colab] 텐서플로우 백엔드를 사용해서 케라스 상에서 무료 GPU를 사용하는 방법을 알아봅니다.
Google Colab 사용하기
Google Colab을 추가하고, 구글 드라이브에서 노트북을 연다.
Runtime -> Change runtime Type 에서 Python 3, GPU를 선택.
위의 코드로 성능을 확인하면 아래와 같다.
MemTotal: 13335204 kB
model name : Intel(R) Xeon(R) CPU @ 2.30GHz
Google Colab에서 GPU를 사용한 MNIST
1
2
3
4
5
| import tensorflow as tf
import matplotlib.pyplot as plt
mnist = tf.keras.datasets.mnist
%matplotlib inline
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
| (x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train = tf.keras.utils.normalize(x_train, axis = 1)
x_test = tf.keras.utils.normalize(x_test, axis = 1)
### GPU 사용
with tf.device('/gpu:0'):
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Flatten())
# Flatten : Flattens the input. Does not affect the batch size.
model.add(tf.keras.layers.Dense(128, activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(128, activation=tf.nn.relu))
model.add(tf.keras.layers.Dense(10, activation=tf.nn.softmax))
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics = ['accuracy']
)
model.fit(x_train, y_train, epochs=3)
|
1
2
| val_loss, val_acc = model.evaluate(x_test, y_test)
print(val_loss, val_acc)
|
Epoch 1/3
60000/60000 [==============================] - 10s 173us/step - loss: 0.2626 - acc: 0.9228
Epoch 2/3
60000/60000 [==============================] - 10s 170us/step - loss: 0.1075 - acc: 0.9667
Epoch 3/3
60000/60000 [==============================] - 10s 169us/step - loss: 0.0742 - acc: 0.9766
```python
10000/10000 [==============================] - 1s 76us/step
0.09344221491254866 0.9716
1
2
3
4
| predictions = model.predict(x_test)
import numpy as np
print(np.argmax(predictions[0]))
|
7
1
2
| plt.imshow(x_test[0],cmap=plt.cm.binary)
plt.show()
|
array([1.7975790e-10, 2.3375832e-09, 3.0333268e-08, 3.2107891e-07,
2.4037560e-11, 8.0090983e-09, 4.9225248e-15, 9.9999917e-01,
1.3270974e-08, 4.5924011e-07], dtype=float32)
prediction 값을 직접 살펴보면 7이 제일 확률이 높다.(softmax 값)
categorical_crossentropy와 sparse_categorical_crossentropy 의 차이
1) categorical_crossentropy
-> 원-핫-인코딩된 레이블로 cross entropy를 계산 (target과 output의 shape가 같아야 함)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
| def categorical_crossentropy(target, output, from_logits=False, axis=-1):
"""Categorical crossentropy between an output tensor and a target tensor.
# Arguments
target: A tensor of the same shape as `output`.
output: A tensor resulting from a softmax
(unless `from_logits` is True, in which
case `output` is expected to be the logits).
from_logits: Boolean, whether `output` is the
result of a softmax, or is a tensor of logits.
axis: Int specifying the channels axis. `axis=-1`
corresponds to data format `channels_last`,
and `axis=1` corresponds to data format
`channels_first`.
# Returns
Output tensor.
# Raises
ValueError: if `axis` is neither -1 nor one of
the axes of `output`.
"""
|
2) sparse_categorical_crossentropy
-> 한개의 있는 정답레이블을 가지고 있음
-> MNIST의 경우 : 정답 레이블이 0 ~ 9로, 하나의 열에 정답 레이블을 정수로 표시되어 있으므로 sparse_categorical_crossentropy
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
| def sparse_categorical_crossentropy(target, output, from_logits=False, axis=-1):
"""Categorical crossentropy with integer targets.
# Arguments
target: An integer tensor.
output: A tensor resulting from a softmax
(unless `from_logits` is True, in which
case `output` is expected to be the logits).
from_logits: Boolean, whether `output` is the
result of a softmax, or is a tensor of logits.
axis: Int specifying the channels axis. `axis=-1`
corresponds to data format `channels_last`,
and `axis=1` corresponds to data format
`channels_first`.
# Returns
Output tensor.
# Raises
ValueError: if `axis` is neither -1 nor one of
the axes of `output`.
"""
|
Tensorflow 백엔드 소스 :
https://github.com/keras-team/keras/blob/master/keras/backend/tensorflow_backend.py
