手写数字识别记录

独热编码

是一种稀疏向量

其中一个元素为1，其余元素均为0

常用于表示有有限个可能值的字符串或标识符

mnist = input_data.read_data_sets("MNST_data/", one_hot = True)
mnist.train.labels[1] # 值为3
 #则输出为
array([0., 0., 0., 1., 0., 0., 0., 0., 0., 0.])

 

 

如果直接用阿拉伯数字，则1的距离更近于3，然而其实8更相似与3

 

  argmax

#argmax返回最大值的下标
A = tf.constant([3, 20, 60, 7, 6])
print(tf.argmax(A).numpy())

#输出为3

#numpy中也提供了argmax，可对numpy.array 数据使用

 

 argmax第二个参数为0时，取每列中最大，为1时取每行中最大

 

 

Softmax

 

 

 

 

 

 

 

交叉熵损失

 

 

 

 

 

 

 

code：

%matplotlib inline
from matplotlib import pyplot as plt
import tensorflow as tf
import numpy as np
from sklearn.preprocessing import scale
mnist = tf.keras.datasets.mnist
(train_image, train_label), (test_image, test_label) = mnist.load_data()

train_image = train_image.reshape(-1, 784)
test_image = test_image.reshape(-1, 784)

train_image = tf.cast(train_image / 255.0, dtype = tf.float32)
test_image = tf.cast(test_image / 255.0, dtype = tf.float32)

train_label = tf.one_hot(train_label, depth = 10)
#test_label = tf.one_hot(test_label, depth = 10)


w = tf.Variable(tf.random.normal([784, 10], mean = 0.0, stddev = 1.0, dtype = tf.float32))
b = tf.Variable(tf.random.normal([10], dtype = tf.float32))

def model(x, w, b):
    ret = tf.matmul(x, w) + b
    return tf.nn.softmax(ret)
def loss_fun(x, y, w, b):
    return tf.reduce_mean(tf.keras.losses.categorical_crossentropy(y_true = y, y_pred = model(x, w, b)))

def grad(x, y, w, b):
    with tf.GradientTape() as tap:
        loss_ = loss_fun(x, y, w, b)
    return tap.gradient(loss_, [w, b])

optimizer = tf.keras.optimizers.Adam(learning_rate = learn_rate)

learn_rate = 0.001
batch = 30
total = train_image.shape[0]
n = int(total / batch)
for i in range(20):
    for j in range(n):
        xs = train_image[j * batch : (j + 1) * batch]
        ys = train_label[j * batch : (j + 1) * batch]
        grads = grad(xs, ys, w, b)
        optimizer.apply_gradients(zip(grads, [w, b]))
        
    loss_ = loss_fun(train_image, train_label, w, b)
    print("Train: ", i + 1, "loss: ", loss_)



plt.imshow(tf.reshape(test_image[1], (28, 28)), cmap = "binary")
plt.show()
print("pred:", tf.argmax(model(tf.reshape(test_image[1], (-1, 784)), w, b), 1).numpy(), "True:", test_label[1])

 

 

cifar100：

import tensorflow as tf
import numpy
from tensorflow.python.keras.datasets import cifar100
import matplotlib.pyplot as plt


if __name__ == "__main__" :
    (train_x, train_y), (test_x, test_y) = cifar100.load_data()
    model = tf.keras.models.Sequential()
    model.add(tf.keras.layers.Conv2D(32, (3, 3), input_shape = (32, 32, 3), padding = 'same', activation = 'relu'))
    model.add(tf.keras.layers.Dropout(0.3))
    model.add(tf.keras.layers.MaxPooling2D((2, 2)))
    model.add(tf.keras.layers.Conv2D(64, (3, 3), padding = 'same', activation = 'relu'))
    model.add(tf.keras.layers.MaxPooling2D((2, 2)))
    model.add(tf.keras.layers.Flatten())
    model.add(tf.keras.layers.Dense(1024, activation = 'relu'))
    model.add(tf.keras.layers.Dense(100, activation = 'softmax'))
    model.compile(optimizer = 'adam', loss = 'sparse_categorical_crossentropy', metrics = 'accuracy')

    model.h = model.fit(train_x, train_y, epochs = 5, validation_split = 0.2, batch_size = 100, verbose = 2)
    test_pred = model.predict_classes(test_x)
    print(test_pred[0])
    plt.imshow(test_x[0])
    plt.show()