Tensorflow : MNIST 손글씨 숫자예측 ANN으로 모델링 하기

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from PIL import Image
%matplotlib inline

import tensorflow as tf
from tensorflow.keras.datasets import mnist
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Activation
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.utils import to_categorical

(X_train, y_train), (X_test, y_test) = mnist.load_data()

데이터 모양 확인

 

28행 28열

---

데이터 일렬로 만들기

X_train = X_train.reshape(60000,784)
X_test = X_test.reshape(10000,784)

---

float으로 바꾸고 정규화 시켜주기

X_train = X_train.astype(float)
X_test = X_test.astype(float)

 

 

X_test = X_test / 255.0
X_train = X_train / 255.0

---

인코딩

https://seonggongstory.tistory.com/154

tf.keras.utils.to_categorical 원핫 인코딩으로 바꾸기

y_train = tf.keras.utils.to_categorical(y_train,num_classes=10)
y_test = tf.keras.utils.to_categorical(y_test,num_classes=10)

---

모델링

def build_model() :
  model = Sequential()
  model.add( Dense(512  , 'relu'  , input_shape=(784,)  )  )
  model.add( Dropout(0.4))
  model.add( Dense(10,'softmax')    )
  model.compile('adam',loss='categorical_crossentropy',  metrics=['accuracy'])
  return model

https://seonggongstory.tistory.com/133

딥러닝 하이퍼 파라미터 용어: optimizer 와 손실함수(loss function)

loss펑션 다른점 확인

 

---

학습

model = build_model()
epoch_history = model.fit(X_train,y_train,epochs=5,validation_data=(X_test,y_test) )

---

평가

model.evaluate(X_test,y_test)

---

컨퓨전 매트릭스와 히트맵

 

 

https://seonggongstory.tistory.com/150

레이블인코딩으로 변환( argmax ): Classification metrics can't handle a mix of multilabel-indicator and continuous-mult

from sklearn.metrics import confusion_matrix
y_pred = model.predict(X_test)
y_pred =  y_pred.argmax(axis=1)
y_test = y_test.argmax(axis=1)
cm = confusion_matrix(y_test,y_pred)

 

import seaborn as sb
sb.heatmap(cm,annot=True, fmt='.0f', cmap='RdPu')
plt.show()