文章目录

• ​​tensorflow数据类型​​
• ​​创建Tensor​​
• ​​索引操作​​
• ​​维度变换​​
• ​​Broadcasting​​
• ​​数学运算​​
• ​​前向传播​​
• ​​合并与分割​​
• ​​数据统计​​
• ​​张量排序​​
• ​​填充与复制​​
• ​​张量限幅​​
• ​​高阶op​​
• ​​数据加载​​
• ​​测试 (张量) 实战​​
• ​​全连接层​​
• ​​输出方式​​
• ​​误差计算​​
• ​​梯度计算​​
• ​​函数优化实战​​

tensorflow数据类型

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test1.ipynb​​

创建Tensor

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test2.ipynb​​

索引操作

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/bj/blob/master/tensor_test3.ipynb​​

维度变换

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/bj/blob/master/tensor_test4.ipynb​​

Broadcasting

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test5.ipynb​​

数学运算

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test6.ipynb​​

前向传播

import tensorflow as tf
from tensorflow.keras import datasets
import os

os.environ["TF_CPP_MIN_LOG_LEVEL"] = '2'
# x: [60k,28,28],
# y: [60k]
# 加载数据集
(x, y), _ = datasets.mnist.load_data()

# 转换成tensor形式
# x:[0~255] => [0~1.] # 便于deep learning 优化
x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.
y = tf.convert_to_tensor(y, dtype=tf.int32)

# print(tf.reduce_min(x), tf.reduce_max(x))
# print(tf.reduce_min(y), tf.reduce_max(y))

train_db = tf.data.Dataset.from_tensor_slices((x, y)).batch(128)
train_iter = iter(train_db)
sample = next(train_iter)
# (128, 28, 28) (128,)
print("batch:", sample[0].shape, sample[1].shape)

# [b, 784] => [b, 256] => [b, 128] => [b, 10]
# [dim_in, dim_out], [dim_out]
# 截断正态分布 方差0.1
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
b1 = tf.Variable(tf.zeros([256]))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 = tf.Variable(tf.zeros([128]))
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros([10]))
lr = 1e-3

# iterate db for 10
for epoch in range(10):
    for step, (x, y) in enumerate(train_db): # for every batch
        # x: [128, 28, 28]
        # y: [128]

        # x: [b, 28, 28] => [b, 28*28]
        x = tf.reshape(x, [-1, 28*28])

        # 只能跟踪 tf.Variable 类型
        with tf.GradientTape() as tape:

            # x: [b, 28*28]
            # h1 = x@w1 + b1
            # [b, 784]@[784, 256] + [256] => [b, 256] + [256]
            h1 = x@w1 + b1
            h1 = tf.nn.relu(h1)
            # [b, 256] => [b, 128]
            h2 = h1@w2 + b2
            h2 = tf.nn.relu(h2)
            # [b, 128] => [b, 10]
            out = h2@w3 + b3

            # compute loss
            # out: [b, 10]
            # y: [b] => [b, 10]
            y_onehot = tf.one_hot(y, depth=10)

            # mse = mean(sum(y-out)^2)
            # square平方
            # [b, 10]
            loss = tf.square(y_onehot - out)
            # mean: scalar
            loss = tf.reduce_mean(loss)

        # compute gradients
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])

        # w1 = w1 - lr * grads[0]
        # 原地更新
        w1.assign_sub(lr * grads[0])
        b1.assign_sub(lr * grads[1])
        w2.assign_sub(lr * grads[2])
        b2.assign_sub(lr * grads[3])
        w3.assign_sub(lr * grads[4])
        b3.assign_sub(lr * grads[5])

        if step % 100 == 0:
            print("epoch",epoch,"step",step, 'loss:', float(loss))

合并与分割

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test7.ipynb​​

数据统计

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test8.ipynb​​

张量排序

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test9.ipynb​​

填充与复制

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test10.ipynb​​

张量限幅

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test11.ipynb​​

高阶op

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test12.ipynb​​

数据加载

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test13.ipynb​​

测试 (张量) 实战

import tensorflow as tf
from tensorflow.keras import datasets
import os

os.environ["TF_CPP_MIN_LOG_LEVEL"] = '2'
# x: [60k,28,28], [10,28,28]
# y: [60k]
# 加载数据集
(x, y), (x_test, y_test) = datasets.mnist.load_data()

# 转换成tensor形式
# x:[0~255] => [0~1.] # 便于deep learning 优化
x = tf.convert_to_tensor(x, dtype=tf.float32) / 255.
y = tf.convert_to_tensor(y, dtype=tf.int32)

x_test = tf.convert_to_tensor(x_test, dtype=tf.float32) / 255.
y_test = tf.convert_to_tensor(y_test, dtype=tf.int32)

# print(tf.reduce_min(x), tf.reduce_max(x))
# print(tf.reduce_min(y), tf.reduce_max(y))

train_db = tf.data.Dataset.from_tensor_slices((x, y)).batch(128)
test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(128)

train_iter = iter(train_db)
sample = next(train_iter)
# (128, 28, 28) (128,)
print("batch:", sample[0].shape, sample[1].shape)

# [b, 784] => [b, 256] => [b, 128] => [b, 10]
# [dim_in, dim_out], [dim_out]
# 截断正态分布 方差0.1
w1 = tf.Variable(tf.random.truncated_normal([784, 256], stddev=0.1))
b1 = tf.Variable(tf.zeros([256]))
w2 = tf.Variable(tf.random.truncated_normal([256, 128], stddev=0.1))
b2 = tf.Variable(tf.zeros([128]))
w3 = tf.Variable(tf.random.truncated_normal([128, 10], stddev=0.1))
b3 = tf.Variable(tf.zeros([10]))
lr = 1e-3

# iterate db for 10
for epoch in range(100):
    for step, (x, y) in enumerate(train_db): # for every batch
        # x: [128, 28, 28]
        # y: [128]

        # x: [b, 28, 28] => [b, 28*28]
        x = tf.reshape(x, [-1, 28*28])

        # 只能跟踪 tf.Variable 类型
        with tf.GradientTape() as tape:

            # x: [b, 28*28]
            # h1 = x@w1 + b1
            # [b, 784]@[784, 256] + [256] => [b, 256] + [256]
            h1 = x@w1 + b1
            h1 = tf.nn.relu(h1)
            # [b, 256] => [b, 128]
            h2 = h1@w2 + b2
            h2 = tf.nn.relu(h2)
            # [b, 128] => [b, 10]
            out = h2@w3 + b3

            # compute loss
            # out: [b, 10]
            # y: [b] => [b, 10]
            y_onehot = tf.one_hot(y, depth=10)

            # mse = mean(sum(y-out)^2)
            # square平方
            # [b, 10]
            loss = tf.square(y_onehot - out)
            # mean: scalar
            loss = tf.reduce_mean(loss)

        # compute gradients
        grads = tape.gradient(loss, [w1, b1, w2, b2, w3, b3])

        # w1 = w1 - lr * grads[0]
        # 原地更新
        w1.assign_sub(lr * grads[0])
        b1.assign_sub(lr * grads[1])
        w2.assign_sub(lr * grads[2])
        b2.assign_sub(lr * grads[3])
        w3.assign_sub(lr * grads[4])
        b3.assign_sub(lr * grads[5])

        if step % 100 == 0:
            print("epoch", epoch, "step", step, 'loss:', float(loss))

    # test/evluation
    # [w1, b1, w2, b2, w3, b3]
    total_correct, total_num = 0, 0
    for step, (x, y) in enumerate(test_db):

        # [b, 28, 28] => [b, 28*28]
        x = tf.reshape(x, [-1, 28*28])

        # [b, 784] => [b, 256] => [b, 128] => [b, 10]
        h1 = tf.nn.relu(x@w1 + b1)
        h2 = tf.nn.relu(h1@w2 + b2)
        out = h2@w3 + b3

        # out: [b, 10] ~ R
        # prob [b, 10] ~ [0, 1]

        # 映射到0~1的范围内
        prob = tf.nn.softmax(out, axis=1)

        # [b, 10] => [b]
        # int64!!!!
        # 返回axis=1维度上最大值对应的索引
        pred = tf.argmax(prob, axis=1)
        pred = tf.cast(pred, dtype=tf.int32)

        # y: [b]
        # [b], int32
        correct = tf.cast(tf.equal(pred, y), dtype=tf.int32)
        correct = tf.reduce_sum(correct)

        total_correct += int(correct)
        total_num += x.shape[0]

    # 计算准确率
    acc = total_correct / total_num
    print("accu:",acc)

全连接层

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test14.ipynb​​

import tensorflow as tf
from tensorflow import keras

# [batch, data]
x = tf.random.normal([2, 5])

# 指定每一层的神经元个数和激活函数
model = keras.Sequential([
    keras.layers.Dense(2, activation='relu'),
    keras.layers.Dense(2, activation='relu'),
    keras.layers.Dense(2)
])
# 指定模型的输入的维度
model.build(input_shape=[None, 5])
# 输出每一层的网络结构
model.summary()

# 显示每一层训练的参数和参数的shape
for p in model.trainable_variables:
    print(p.name, p.shape)

# dense/kernel:0 (5, 2)
# dense/bias:0 (2,)
# dense_1/kernel:0 (2, 2)
# dense_1/bias:0 (2,)
# dense_2/kernel:0 (2, 2)
# dense_2/bias:0 (2,)

输出方式

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test15.ipynb​​

误差计算

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test16.ipynb​​

梯度计算

​​https://nbviewer.jupyter.org/github/sweetheart-7-7/whb/blob/master/tensor_test17.ipynb​​

函数优化实战

import numpy as np
from mpl_toolkits.mplot3d import Axes3D
from matplotlib import pyplot as plt
import tensorflow as tf

def fn(x):
    return (x[0] ** 2 + x[1] - 11) ** 2 + (x[0] + x[1] ** 2 - 7) ** 2


x = np.arange(-6, 6, 0.1)
y = np.arange(-6, 6, 0.1)
print("x,y range:", x.shape, y.shape)
# 生成网格数据
X, Y = np.meshgrid(x, y)
print("X,Y maps:", X.shape, Y.shape)

Z = fn([X, Y])
# 获取到当前figure对象
fig = plt.figure("himmelblau")
# 获取图中的当前极轴。如果不存在，或者不是极轴，则将创建相应的轴，然后返回。
# 此时得到的ax对象的类型是Axes3D的子类，这个对象将是绘制3D图形的入口
ax = fig.gca(projection="3d")
# 为了绘制 3D 图形，需要调用 Axes3D 对象的 plot_surface()方法来完成。
ax.plot_surface(X, Y, Z)
# 转换视角进行观察
ax.view_init(60, -30)
ax.set_xlabel("x")
ax.set_ylabel("y")
plt.show()

x = tf.constant([-4., 0.])


for step in range(200):

    with tf.GradientTape() as tape:
        tape.watch([x])
        y = fn(x)

    grads = tape.gradient(y, [x])[0]
    x -= 0.01 * grads

    if step % 20 == 0:
        print("step {}: x = {}, f(x) = {}".format(step, x.numpy(), y.numpy()))