Fashion MNIST dataset classification training

First of all, the previous article recorded MNIST Simple classification play . At present, there is no challenge , As its replacement Fashion MNIST . Of course we're going to play

It grows like this. , There are also 10 class . and MNIST The size and number of classified pictures are the same

Implementation code ： I use jupyter. The installation of the previous blog post tf2.6.0 Written in the environment

import tensorflow as tf
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
# First import the required package 
# Load data set , You can introduce yourself here , No need to download , For speed , You can download several files , Put corresponding jupyter File location , Let me see the picture below , There is sharing in the document comments 
(train_image, train_lable), (test_image, test_label) = tf.keras.datasets.fashion_mnist.load_data()

# The following is not necessary , This is to look at the contents of the data set 
train_image.shape,train_lable.shape,test_image.shape, test_label.shape
plt.imshow(train_image[1])

# Next 
train_image = train_image/255
test_image = test_image/255        # normalization 
# Creating models , This part can be complicated by increasing the number of layers 
model = tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28,28)))  
# 28*28  Flatten 
model.add(tf.keras.layers.Dense(128, activation='relu')) 
# 128 Number of hidden layer units , Too big is easy to fit , Too small to miss information 
model.add(tf.keras.layers.Dense(10, activation='softmax'))
# Output layer ,10 A probability 
model.summary()

'''
model = tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28,28)))  # 28*28
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(10, activation='softmax'))
'''

# Finally, choose the optimizer adam, Loss function 
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',  
              metrics=['acc']
)
# Start training , I use this epochs=5.50 You can achieve 96% many 
model.fit(train_image, train_lable, epochs=5)

  The dataset has 60000 Zhang 28*28 Training set of , and 10000 Zhang test set

Take out the second picture of the training set and see what it looks like

  The content of the model is like this :

  Add a hidden layer to the model content, so :

 5 individual epochs Time accurate removal rate acc=89.18%

  50 individual epochs Time accurate removal rate acc=96.76%

  additional ： 100% complete and clean 96.75 The simple code of

import tensorflow as tf
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

(train_image, train_lable), (test_image, test_label) = tf.keras.datasets.fashion_mnist.load_data()

train_image = train_image/255
test_image = test_image/255        # normalization 

model = tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28,28)))  # 28*28  Flatten 
model.add(tf.keras.layers.Dense(128, activation='relu')) # 128 Number of hidden layer units 
model.add(tf.keras.layers.Dense(10, activation='softmax'))# Output layer ,10 A probability 

model.summary()

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',   
              metrics=['acc']
)

model.fit(train_image, train_lable, epochs=50)