LeNet5訓練cifar10

import os

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

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
tf.random.set_seed(2345)

conv_layers = [
    # unit1
    layers.Conv2D(6, kernel_size=[5, 5], strides=1, padding='valid', activation=tf.nn.sigmoid),
    layers.MaxPool2D(pool_size=[2, 2], padding='same', strides=2),

    # unit2
    layers.Conv2D(16, kernel_size=[5, 5], strides=1, padding='valid', activation=tf.nn.sigmoid),
    layers.MaxPool2D(pool_size=[2, 2], padding='same', strides=2),
]


def preprocess(x, y):
    x = tf.cast(x, dtype=tf.float32) / 255
    # x = tf.reshape(x, [32, 32])
    y = tf.cast(y, dtype=tf.int32)
    # y = tf.one_hot(y, depth=10)
    return x, y


def main():
    (x_train, y_train), (x_test, y_test) = datasets.cifar10.load_data()
    print("x_train.shape:", x_train.shape, "y_train.shape:", y_train.shape)
    print("x_test.shape:", x_test.shape, 'y_test.shape:', y_test.shape)
    y_test = tf.squeeze(y_test, axis=1)
    y_train = tf.squeeze(y_train, axis=1)
    train_db = tf.data.Dataset.from_tensor_slices((x_train, y_train))
    train_db = train_db.shuffle(1000).map(preprocess).batch(256)

    test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test))
    test_db = test_db.map(preprocess).batch(256)

    sample_train = next(iter(train_db))
    sample_test = next(iter(test_db))
    print("sample_train[0].shape:", sample_train[0].shape, "sample_train[1].shape:", sample_train[1].shape)
    print("sample_test[0].shape:", sample_test[0].shape, "sample_test[1].shape:", sample_test[1].shape)

    conv_net = Sequential(conv_layers)
    fc_net = Sequential([
        layers.Dense(120, activation=tf.nn.tanh),
        layers.Dense(84, activation=tf.nn.tanh),
        layers.Dense(10, activation=None),
    ])
    conv_net.build(input_shape=[None, 32, 32, 3])
    fc_net.build(input_shape=[None, 400])

    optimizer = optimizers.Adam(lr=1e-4)
    variables = conv_net.trainable_variables + fc_net.trainable_variables
    for epoch in range(2000):
        for step, (x, y) in enumerate(train_db):
            with tf.GradientTape() as tape:
                # [b, 32, 32, 3]->[b, 5,5,16]
                out = conv_net(x)
                # [b,5,5,16]->[b,400]
                out = tf.reshape(out, [-1, 400])
                # [b, 16, 16, 5]->[b, 10]
                logits = fc_net(out)
                y_onehot = tf.one_hot(y, depth=10)
                loss = tf.losses.categorical_crossentropy(y_onehot, logits, 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))
        total_num = 0
        total_correct = 0
        for x, y in test_db:
            out = conv_net(x)
            out = tf.reshape(out, [-1, 400])
            logits = fc_net(out)
            prob = tf.nn.softmax(logits, axis=1)
            pred = tf.argmax(prob, axis=1)
            pred = tf.cast(pred, dtype=tf.int32)
            # y = tf.cast(y, dtype=tf.int32)
            correct = tf.cast(tf.equal(pred, y), dtype=tf.int32)
            correct = tf.reduce_sum(correct)

            total_num += x.shape[0]
            total_correct += int(correct)
        acc = total_correct / total_num
        print(epoch, 'acc:', acc)


if __name__ == '__main__':
    main()

 

 利用LeNet5訓練cifar10數據集，跑了2000個epoch，准確率只有0.63，不是很理性，主要是LeNet5網絡結構過於簡單