循环神经网络

import numpy as np
import tensorflow as tf
from sklearn.datasets import load_iris

BATCH_SIZE = 8
COST = 9
PROFIT = 1
iris = load_iris()
iris_data = iris.data
iris_label = iris.target
iris_label = np.reshape(iris_label,(150,1))

## 在tensorflow中数据类型不同不能进行运算
iris_label = tf.to_double(iris_label,name = "ToDouble")
## training data and label
x = iris_data
length = len(x)
x = tf.to_double(iris_label,name="ToDouble")
#y_ = tf.placeholder(tf.float32,shape=(None,1))

## Variables
state = tf.Variable(tf.random_normal([4,4],stddev=1,dtype=tf.float64,seed=1))
#print(state.dtype)
w_cell_input = tf.Variable(tf.random_normal([1,4],stddev=1,dtype=tf.float64,seed=1))
w_cell_state = tf.Variable(tf.random_normal([4,4],stddev=1,dtype=tf.float64,seed=1))
w_output = tf.Variable(tf.random_normal([4,4],stddev=1,dtype=tf.float64,seed=1))
b_output = tf.Variable(tf.random_normal([4,4],stddev=1,dtype=tf.float64,seed=1))
b_cell = tf.Variable(tf.random_normal([4,4],stddev=1,dtype=tf.float64,seed=1))
w1_reshape = tf.Variable(tf.random_normal([4,1],stddev=1,dtype=tf.float64,seed=1))
w2_reshape = tf.Variable(tf.random_normal([1,4],stddev=1,dtype=tf.float64,seed=1))

for i in range(length):
    #x[i] = np.reshape(x[i],(4,1))
    print(x[i].dtype)
    before_activation = np.dot(state,w_cell_state) + x[i]*w_cell_input + b_cell

    state = tf.nn.tanh(before_activation)

    final_output = np.dot(state,w_output) + b_output

    temp = tf.matmul(final_output,w1_reshape)
    output = tf.matmul(w2_reshape,temp)

    loss = tf.reduce_sum(tf.where(tf.greater(iris_label[i],output),(iris_label[i]-output)*COST,
                                  (output-iris_label[i]*PROFIT)))
    train_step = tf.train.GradientDescentOptimizer(0.001).minimize(loss)


    with tf.Session() as sess:
        init_op = tf.global_variables_initializer()
        sess.run(init_op)
        steps = 5000
        for k in range(steps):
            start = (i * BATCH_SIZE) % 32
            end = (i * BATCH_SIZE) % 32 + BATCH_SIZE
            sess.run(train_step)
            if k%1000 == 0:
                print(sess.run(loss))

#########################################################################
字符串 ==> 数字
tf.string_to_number(string , out_type = None , name = None)
#########################################################################
==>float64
tf.to_double(x , name = "ToDouble")
#########################################################################
==>float32
tf.to_float(x , name = "Tofloat")
#########################################################################
==>int32
tf.to_int32(x , name = "ToInt32")
#########################################################################
==>int64
tf.to_int64(x , name = "ToInt64")
#########################################################################
==>将 x 转换为 dtype
tf.case( a ,tf.int32)