06 TensorFlow 2.0：CNN CIFAR-100實戰

白衣縱馬風流年少
佳人傾城回眸淺笑
玉笛聲聲月色皎皎
起舞翩翩清影窈窕
姻緣樹下共求月老
執手暮暮朝朝
                                                                                                                                《慕夏》

import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'

import tensorflow as tf
from tensorflow.keras import  layers, optimizers, datasets, Sequential 

print(tf.__version__)
gpu_ok = tf.test.is_gpu_available()
print("\nuse GPU",gpu_ok)

tf.random.set_seed(1234)
lr = 1e-4

conv_layer = [
    layers.Conv2D(filters=64, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.Conv2D(filters=64, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(filters=128, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.Conv2D(filters=128, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(filters=256, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.Conv2D(filters=256, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),
    
    layers.Conv2D(filters=512, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.Conv2D(filters=512, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same'),

    layers.Conv2D(filters=512, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.Conv2D(filters=512, kernel_size=[3, 3], strides=(1, 1), padding='same', activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding='same')
]

fc_layer = [
    layers.Dense(units=256, activation=tf.nn.relu),
    layers.Dense(units=128, activation=tf.nn.relu),
    layers.Dense(units=100, activation=None)
]


conv_net = Sequential(conv_layer)
conv_net.build(input_shape=[None, 32, 32, 3])
fc_net = Sequential(fc_layer)
fc_net.build(input_shape=[None, 512])
optimizer = optimizers.Adam(learning_rate=lr)

print(conv_net.summary())
print(fc_net.summary())

def normlize_data(x, y):
    x = tf.cast(x, dtype=tf.float32)/255.
    y = tf.cast(y, dtype=tf.int32)
    return x, y
    
(x, y),(x_test, y_test) = datasets.cifar100.load_data() 
    
y = tf.squeeze(y, axis=1)
y_test = tf.squeeze(y_test, axis=1) 
    
train_db = tf.data.Dataset.from_tensor_slices((x, y))
train_db = train_db.shuffle(2000).map(normlize_data).batch(128)
    
test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test))
test_db = test_db.map(normlize_data).batch(128)

type(train_db)
sample = next(iter(train_db))
print(sample[0].shape)
print(sample[1].shape)
print(tf.reduce_max(sample[0]), tf.reduce_min(sample[0]))
print(tf.reduce_max(sample[1]), tf.reduce_min(sample[1]))

variables = conv_net.trainable_variables+fc_net.trainable_variables
def execut():
    for epoch in range(3):
        for step, (x, y) in enumerate(train_db):
            with tf.GradientTape() as tape:
                conv_net_out = conv_net(x) # [b, 32, 32, 3] -> [b, 1, 1, 512]
                
                conv_net_out = tf.reshape(conv_net_out, shape=[-1, 512])
                
                out = fc_net(conv_net_out) # [b, 512] -> [b, 100]
                y_one_hot = tf.one_hot(y, depth=100)
                
                loss = tf.losses.categorical_crossentropy(y_one_hot, out, from_logits=True)
                loss = tf.reduce_mean(loss)
                
            grads = tape.gradient(loss, variables)
            optimizer.apply_gradients(zip(grads, variables))
            
            if step%100==0:
                print(epoch, step, 'loss:', float(loss))
                
        # every epoch compute acc
        total_num = 0
        total_correct = 0
        for test_x, test_y in test_db:
            conv_net_test_out = conv_net(test_x)
            conv_net_test_out = tf.reshape(conv_net_test_out, shape=[-1, 512])
            test_out = fc_net(conv_net_test_out)
            prob = tf.nn.softmax(test_out, axis=1)
            pred = tf.argmax(prob, axis=1)
            pred = tf.cast(pred, dtype=tf.int32)
            
            correct = tf.reduce_sum(tf.cast(tf.equal(pred, y)), dtype=tf.int32)
            
            total_num += x.shape[0]
            total_correct += correct
            
        acc = total_correct/total_num
        print(epoch, 'acc:', acc)           

execut()