numtest

▶ 미션 :

• 데이터획득: .mat 에서 x, y 를 가져옵니다
• 전처리 : x, y 를 학습할 수 있게 만듭니다
• x : (h,w,c,m) > (m,h,w,c) 로 수정
• y : (m,1) > (m) 로 수정
• y 의 라벨을 확인하세요. 숫자 0 의 라벨이 '10'입니다. 이것을 0으로 수정하세요
• 모델 생성: CNN, RNN, CRNN 등의 모델을 만듭니다
• 학습 및 평가: 데이터를 학습시키고 성능을 측정합니다

%tensorflow_version 2.0x
import tensorflow as tf
from tensorflow import keras
from google.colab import files

# 파일 업로드 방식
uploaded = files.upload()

from scipy import io
mat_file = io.loadmat('test_32x32.mat')

mat_file 

{'X': array([[[[ 38, 129, 150, ..., 115, 96, 101],

[103, 142, 160, ..., 132, 65, 75],

[ 60, 153, 169, ..., 142, 47, 60]],

---

[[ 42, 123, 152, ..., 144, 104, 102],

[113, 130, 168, ..., 157, 85, 71],

[ 71, 156, 183, ..., 165, 68, 50]]]], dtype=uint8),

 

'y': array([[5], [2], [1], ..., [7], [6], [7]], dtype=uint8)}

x = mat_file['X']
y = mat_file['y']

# (width, height, channels, size) -> (size, width, height, channels)
x, y = x.transpose((3,0,1,2)), y[:,0]

print("Test Set", x.shape)
print('')

num_images = x.shape[0]

print("Total Number of Images", num_images)

→ Test Set (26032, 32, 32, 3)

→ Total Number of Images 26032

y[:100]

→ array([ 5, 2, 1, 10, 6, 1, 9, 1, 1, 8, 3, 6, 5, 1, 4, 4, 1, 6, 3, 4, 2, 10, 1, 3, 2, 5, 4, 1, 4, 2, 8, 3, 8, 6, 10, 1, 5, 1, 1, 2, 9, 1, 6, 9, 2, 6, 1, 2, 10, 6, 9, 1, 5, 1, 9, 8, 1, 5, 1, 5, 2, 9, 2, 4, 1, 10, 6, 9, 3, 2, 3, 5, 1, 1, 4, 6, 2, 8, 3, 1, 4, 1, 7, 1, 2, 1, 7, 1, 10, 1, 4, 4, 6, 4, 4, 2, 8, 1, 7, 7], dtype=uint8)

y[y==10] = 0

x.shape, y.shape

→ ((26032, 32, 32, 3), (26032,))

import numpy as np
from sklearn.datasets import load_digits
import tensorflow as tf
import scipy.io as sio
import time
import matplotlib.pyplot as plt

plt.imshow(x[1])

class MyModel(keras.Model):
  def __init__(self):
    super(MyModel, self).__init__()#상속한 클래스의 생성자 호출 
    self.opt = tf.optimizers.RMSprop(learning_rate=0.01)#Stochatic Gradient Descent 확률적 경사 하강
    # RMSprop 최적화 프로그램
    self.conv0 = keras.layers.Conv2D(3, [3,3], padding='same', activation=keras.activations.relu)
    self.conv1 = keras.layers.Conv2D(6, [3,3], padding='same', activation=keras.activations.relu)
    self.pool0 = keras.layers.MaxPool2D([2,2], padding='same')
    self.pool1 = keras.layers.MaxPool2D([2,2], padding='same')
    self.flatten = keras.layers.Flatten()
    self.dense = keras.layers.Dense(units=10, activation=keras.activations.softmax)
  
  def call(self, x):
    x_4d = tf.reshape(x, [-1,32,32,3]) / 255
    x_4d = tf.cast(x_4d, tf.float32)
    net = self.conv0(x_4d)
    net = self.pool0(net)
    net = self.conv1(net)
    net = self.pool1(net)
    net = self.flatten(net)    
    h = self.dense(net)
    return h

  def get_loss(self, y, h):
    #학습할때 nan이 발생하는 경우 값을 clip(자르다) (최소값, 최대값) 
    h = tf.clip_by_value(h, 1e-6, 1 - 1e-6) # h 가 0이나 1이 되지 않도록 하는 안전장치 
    cross_entropy = - (y * tf.math.log(h) + (1 - y) * tf.math.log(1 - h)) 
    loss = tf.reduce_mean(cross_entropy)
    return loss

  def get_accuracy(self, y, h):    
    predict = tf.argmax(h, -1)
    self.acc = tf.reduce_mean(tf.cast(tf.equal(y, predict), tf.float32)) # True > 1, False > 0 로 cast

  def fit(self, x, y, epoch=1):
    # x : (m, 4), y: (m)    
    y_hot = tf.one_hot(y, depth=10, axis=-1)#(m, 3)  
    for i in range(epoch):
      with tf.GradientTape() as tape: #경사 기록 장치
        h = self.call(x)
        loss = self.get_loss(y_hot, h)        
      grads = tape.gradient(loss, self.trainable_variables) #경사 계산
      self.opt.apply_gradients(zip(grads, self.trainable_variables)) # 가중치에서 경사를 빼기
      self.get_accuracy(y, h)
      if i%10==0:
        print('%d/%d loss:%.3f acc:%.3f'%(i, epoch, loss, self.acc))
model = MyModel()

model.fit(x[:500], y[:500], epoch=500) #학습

0/500 loss:0.331 acc:0.056

10/500 loss:0.312 acc:0.196

20/500 loss:0.301 acc:0.224

30/500 loss:0.280 acc:0.342

---

490/500 loss:0.402 acc:0.596

 

 

---

## 모델 생성
epoch = 100
batch_m = 1000 # 메모리가 부족한 경우 데이터를 배치단위로 나누어서 학습시킵니다.
for i in range(epoch):
  iteration = np.ceil(len(y)/batch_m)
  for j in range(iteration):
    start = i*batch_m
    end = (i+1)*batch_m
    x_batch = x[start:end]
    y_batch = y[start:end]