根据Tensorflow官网叙述，keras中的MNIST服装分类相当于tensorflow中的【hello world】，是一个良好的测试Tensorflow环境的工具。
新装的最新版1.13.1当然是要先测试一下在撸起袖子干啦！
以下摘自Tensorflow官网：
https://www.tensorflow.org/tutorials/keras/basic_classification
Fashion MNIST 的作用是成为经典 MNIST 数据集的简易替换，后者通常用作计算机视觉机器学习程序的“Hello, World”入门数据集。MNIST 数据集包含手写数字（0、1、2 等）的图像，这些图像的格式与我们在本教程中使用的服饰图像的格式相同。
本指南使用 Fashion MNIST 实现多样化，并且它比常规 MNIST 更具挑战性。这两个数据集都相对较小，用于验证某个算法能否如期正常运行。它们都是测试和调试代码的良好起点。
我们将使用 60000 张图像训练网络，并使用 10000 张图像评估经过学习的网络分类图像的准确率。您可以从 TensorFlow 直接访问 Fashion MNIST。
代码附上
由于博主留学，跟教研室交流方便，懒得翻译，注释就用英文啦。
# TensorFlow and tf.keras
import tensorflow as tf
from tensorflow import keras
# Helper libraries
import numpy as np
import matplotlib.pyplot as plt
# Show single picture
def mnist_show_pic(num):
plt.figure()
plt.imshow(train_images[num])
plt.colorbar()
plt.grid(False)
plt.show()
# show pictures with labels in an array
# num_img : number of first pictures to show
# num_x : number of pictures in x-dir of the array
# num_y : number of pictures in y-dir of the array
def mnist_show_label(num_img, num_x, num_y):
plt.figure(figsize=(10,10))
for i in range(num_img):
plt.subplot(num_x, num_y, i+1)
plt.xticks([])
plt.yticks([])
plt.grid(False)
plt.imshow(train_images[i], cmap=plt.cm.binary)
plt.xlabel(class_names[train_labels[i]])
plt.show()
# plot single image with predictions
def plot_image(i, predictions_array, true_label, img):
predictions_array, true_label, img = predictions_array[i], true_label[i], img[i]
plt.grid(False)
plt.xticks([])
plt.yticks([])
plt.imshow(img, cmap=plt.cm.binary)
predicted_label = np.argmax(predictions_array)
if predicted_label == true_label:
color = 'blue'
else:
color = 'red'
plt.xlabel("{} {:2.0f}% ({})".format(class_names[predicted_label],
100*np.max(predictions_array),
class_names[true_label]),
color=color)
# plot images with predictions in array
def plot_value_array(i, predictions_array, true_label, num_classes, class_names):
predictions_array, true_label = predictions_array[i], true_label[i]
plt.grid(False)
plt.xticks(np.arange(num_classes), [str(j) for j in np.arange(num_classes)])
plt.yticks()
thisplot = plt.bar(range(num_classes), predictions_array, color="#777777")
plt.ylim([0, 1])
predicted_label = np.argmax(predictions_array)
thisplot[predicted_label].set_color('red')
thisplot[true_label].set_color('blue')
# Output results for one image
def plot_result(num):
plt.figure(figsize=(6,3))
# Subplot 1: image with prediction and ground truth:
# <prediction> <probability> (<ground_truth>)
plt.subplot(1,2,1)
plot_image(num, predictions, test_labels, test_images)
# Subplot 2: image with prediction and ground truth:
# <prediction> <probability> (<ground_truth>)
plt.subplot(1,2,2)
plot_value_array(num, predictions, test_labels, num_classes, class_names)
plt.show()
# Output num_rows*num_cols results for image array
def plot_result_array(num_rows, num_cols):
num_rows = 5
num_cols = 3
num_images = num_rows*num_cols
plt.figure(figsize=(2*2*num_cols, 2*num_rows))
for i in range(num_images):
plt.subplot(num_rows, 2*num_cols, 2*i+1)
plot_image(i, predictions, test_labels, test_images)
plt.subplot(num_rows, 2*num_cols, 2*i+2)
plot_value_array(i, predictions, test_labels, num_classes, class_names)
plt.show()
"""
==============================================================================
Main
==============================================================================
"""
if __name__ == '__main__':
# Load fashion_mnist database
# 60000 training images and labels
# 10000 test images and labels
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
# Class names
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
num_classes = len(class_names)
# Scale pixel values to 0~1
train_images = train_images / 255.0
test_images = test_images / 255.0
# Show pictures (test)
# mnist_show_pic(8)
# mnist_show_label(9, 3, 3)
# Setup layers
# Flatten-fc128relu-fc10softmax
model = keras.Sequential([
# Stretch 28*28 image to 748*1 flattened vector
keras.layers.Flatten(input_shape=(28,28)),
# Fully-connected layer with 128 neurons, relu
keras.layers.Dense(128, activation=tf.nn.relu),
# Fully-connected layer with 10 neurons, softmax
keras.layers.Dense(10, activation=tf.nn.softmax)
])
# Compile model
# Loss function: to steer the model in the right direction
# Optimizer: the way to update model based on losses
# Metrics: monitor the training and testing steps
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
# Train model with 5 epochs
model.fit(train_images, train_labels, epochs=5)
# Evaluate model
test_loss, test_acc = model.evaluate(test_images, test_labels)
print("\nFinally:\ntest_loss=%.4f, test_acc=%.4f" % (test_loss, test_acc))
# Make predictions
predictions = model.predict(test_images)
# Show prediction confidence (probability) array of first picture
# predictions[0]
# Return the predicted class
# class_names[np.argmax(predictions[0])]
# Plot result for single image
plot_result(3) # result of image no.3
plot_result_array(3,4) # result of first 12 images in array 3*4
运行结果：
输出每次epoch的准确度：
Epoch 1/5
2019-03-06 18:41:25.739511: I tensorflow/stream_executor/dso_loader.cc:152] successfully opened CUDA library libcublas.so.10.0 locally
60000/60000 [==============================] - 4s 67us/sample - loss: 0.5017 - acc: 0.8220
Epoch 2/5
60000/60000 [==============================] - 2s 40us/sample - loss: 0.3773 - acc: 0.8635
Epoch 3/5
60000/60000 [==============================] - 2s 39us/sample - loss: 0.3373 - acc: 0.8771
Epoch 4/5
60000/60000 [==============================] - 2s 39us/sample - loss: 0.3140 - acc: 0.8845
Epoch 5/5
60000/60000 [==============================] - 2s 39us/sample - loss: 0.2943 - acc: 0.8924
10000/10000 [==============================] - 0s 22us/sample - loss: 0.3398 - acc: 0.8793
Finally:
test_loss=0.3398, test_acc=0.8793
输出了图片、标签、以及softmax输出（可以理解为“概率”）
图片3：长裤，分类正确，“confidence level”为100%
前12张图的结果汇总：其中Sandal的分类错误
扫一扫
1274
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
