用ＣＮＮ与RNN(LSTM)提取DNA序列中的特征的定义函数
# -*- coding: utf-8 -*-
# read the data of npy,the data is input(label)
import os
import numpy as np
import tensorflow as tf
import time
import math
batch_size = 128
path =os.getcwd()
keep_prob = tf.placeholder(tf.float32)
# In order to get the suitable type of data and label
def initial_data(all_data,all_label): #data and label are all the name of .npy
Data_Examples = np.load(path+'/'+ all_data) # all_data = .npy
Data_Labels = np.load(path+'/'+ all_label)
Data_Labels = tf.one_hot(Data_Labels,2,1,0)
Data_Labels = tf.to_float(Data_Labels,name = 'Data_Labels')
return Data_Examples,Data_Labels
def get_batch_data(t_enh,t_enh_label,t_pro,t_pro_label):
t_enh = tf.cast(t_enh, tf.float32)
t_enh_label = tf.cast(t_enh_label, tf.float32)
t_pro = tf.cast(t_pro, tf.float32)
t_pro_label = tf.cast(t_pro_label, tf.float32)
input_queue = tf.train.slice_input_producer([t_enh,t_enh_label,t_pro,t_pro_label], shuffle=False)
t_enh_b,t_enh_label_b,t_pro_b,t_pro_label_b = tf.train.batch(input_queue, batch_size=batch_size, num_threads=2, capacity=20000)
return t_enh_b,t_enh_label_b,t_pro_b,t_pro_label_b
# train_batch_holder#
train_enh_batch_holder = tf.placeholder(tf.float32,shape=(None,10001,4,1),name = 'x-input')
train_pro_batch_holder = tf.placeholder(tf.float32,shape=(None,10001,4,1),name = 'x2-input')
train_enh_label_holder = tf.placeholder(tf.float32,shape=(None,2),name = 'y-input')
def cnn_1(train_batch_holder,filter_size_1,filter_number_1,pool_1_size):
filters_1 = tf.Variable(tf.random_normal([filter_size_1,4,1,filter_number_1],dtype=tf.float32,seed=1))
biases_1 = tf.Variable(tf.random_normal([filter_number_1],dtype=tf.float32,seed=1))
conv_1 =tf.nn.conv2d(train_batch_holder,filters_1,strides=[1,1,1,1],padding='VALID') #####train_batch_holder#####
bias_1 = tf.nn.bias_add(conv_1,biases_1)
actived_conv_1 = tf.nn.relu(bias_1)
pool_1 = tf.nn.max_pool(actived_conv_1,ksize=[1,pool_1_size,1,1],strides=[1,pool_1_size,1,1],padding="VALID")
return pool_1
def cnn_2(pool_1,filter_size_2,filter_number_1,filter_number_2,pool_2_size):
filters_2 = tf.Variable(tf.random_normal([filter_size_2,1,filter_number_1,filter_number_2],seed=1))
biases_2 = tf.Variable(tf.random_normal([filter_number_2],dtype=tf.float32))
conv_2 = tf.nn.conv2d(pool_1,filters_2,strides=[1,1,1,1],padding='VALID') # input pool_1
bias_2 = tf.nn.bias_add(conv_2,biases_2)
actived_conv_2 = tf.nn.relu(bias_2)
pool_2 = tf.nn.max_pool(actived_conv_2,ksize=[1,pool_2_size,1,1],strides=[1,pool_2_size,1,1],padding="VALID")
return pool_2
def cat_cnn(pool_2_enh,pool_2_pro): # cat the enh_pro for next rnn
pool_cat = tf.concat([pool_2_enh,pool_2_pro],1)
pool_cat_1_ndim = pool_cat.get_shape()[0].value
pool_cat_2_ndim = pool_cat.get_shape()[1].value
pool_cat_4_ndim = pool_cat.get_shape()[3].value
pool_cat_3ndim = tf.reshape(pool_cat,[-1,pool_cat_2_ndim,pool_cat_4_ndim])
x = pool_cat_3ndim
return x,pool_cat_2_ndim,pool_cat_4_ndim
def rnn_1(x, pool_cat_2_ndim,pool_cat_4_ndim,filter_size_1,filter_size_2):
n_input = pool_cat_2_ndim # w
n_step = pool_cat_4_ndim # h
n_hidden = 128
n_classes = 2
weights = {'in': tf.Variable(tf.random_normal([n_input, n_hidden])),
'out': tf.Variable(tf.random_normal([n_hidden, n_classes]))}
biases = {'in': tf.Variable(tf.constant(0.1, shape=[n_hidden])),
'out': tf.Variable(tf.constant(0.1, shape=[n_classes]))}
X = tf.reshape(x, [-1, n_input])
X_in = tf.matmul(X, weights['in'])+biases['in']
X_in = tf.reshape(X_in, [-1, n_step, n_hidden])
#define LSTMcell
lstm_cell = tf.contrib.rnn.BasicLSTMCell(n_hidden)
with tf.variable_scope("lstm_cell"+str(filter_size_1)+'_'+str(filter_size_2), reuse=None):
init_state = lstm_cell.zero_state(batch_size, dtype=tf.float32)
outputs, final_state = tf.nn.dynamic_rnn(lstm_cell, X_in, initial_state=init_state, time_major=False)
results = tf.matmul(final_state[1], weights['out']) + biases['out'] # 2 ndim
dim = results.get_shape()[1].value
return results,dim#以上是对.npy数据的读入操作，以及接入CNN，RNN等操作，后续部分接入全连接层与计算 ACC等是常规流程，在此不做赘述。