tensorflow学习（一）

今天开始学习tensorflow框架，从极客学院下载了官方中文教程(15年翻译的），第一天开始学习第一章ng基本流程和原理，作为前奏。然后写了代码，验证一下，准确率确实非常高，非常好用。把代码上传，作为以后备用。

import tensorflow as tf
import numpy as np
import math

class Model:
    def __init__(self,w = np.empty(None),b = None):
        self.b = b
        self.w = w

    def predict(self, input):
        return np.dot(input,self.w) + self.b

data = np.float32(np.random.rand(1000,2))
label = np.dot(data,np.array([[0.100],[0.200]])) + 0.3
m,n = data.shape

num_train = int(m * 0.6)
num_validation = int(m * 0.2)
num_test = int(m * 0.2)

data_train = data[:num_train,:]
data_validation = data[num_train:(num_train+num_validation),:]
data_test = data[(num_train+num_validation):,:]

label_train = label[:num_train,:]
label_validation = label[num_train:(num_train+num_validation),:]
label_test = label[(num_train+num_validation):,:]

w = tf.Variable(tf.random_uniform([2,1],-1.0,1.0))
b = tf.Variable(tf.zeros([1]))

y_train = tf.matmul(data_train,w) + b
loss = tf.reduce_mean(tf.square(label_train - y_train))
'''
bestModel = Model()
minRMSE = (1 << 31) -1
alphas = [0.1,0.3,0.5];
iters = [100,150,200,250];
for iter in iters:
    for alpha in alphas:
        optimizer = tf.train.GradientDescentOptimizer(alpha)
        train = optimizer.minimize(loss)
        init_state = tf.global_variables_initializer()
        with tf.Session() as sess:
            sess.run(init_state)
            for step in range(0,iter):
                sess.run(train)
            model = Model(sess.run(w),sess.run(b))
            p = model.predict(data_validation)
            rmse = np.sqrt(np.mean(np.square(label_validation - p)))
            if rmse < minRMSE:
                minRMSE = rmse
                bestModel = model
np.save("E:\Python\models\weights.npy",bestModel.w)
np.save("E:\Python\models\b.npy",bestModel.b)
'''

weights = np.load("E:\Python\models\weights.npy")
b = np.load("E:\Python\models\b.npy")

model = Model(weights,b)
predicts = model.predict(data_test)
print(predicts)
print(label_test)
print(label_test - predicts)