def train(sel):
print('start training...\n')
import lc_CNN_backward_FC as backward
bw=backward.CNN_backward( MODEL_SAVE_PATH=os.path.join(sel.model_path,sel.save_model_name),
BATCH_SIZE = 10,
LEARNING_RATE_BASE = 0.0001,
LEARNING_RATE_DECAY = 0.99,
REGULARIZER = pow(10,-4) ,
STEPS = 3000,
MOVING_AVERAGE_DECAY=0.99,)
Loss_train,Loss_test=bw.main_train_and_test(sel.x_train,sel.y_train,sel.x_test, sel.y_test)
return Loss_train,Loss_test
)
import time
import tensorflow as tf
import lc_CNN_forward_FC
import numpy as np
import pandas as pd
import os
from sklearn.metrics import classification_report
from sklearn.metrics import accuracy_score
from sklearn.metrics import roc_auc_score
import matplotlib.pyplot as plt
# 导入backfoward
import lc_CNN_backward_FC as backward
bw = backward.CNN_backward()
validation_INTERVAL_SECS = 5
def validation(x_validation,
y_validation,
which_model='last',
just_fetch=True):
"""复现已有的计算图（后续程序会自动读取已经训练过的模型，即使没有也没有关系，程序会从头训练）"""
with tf.Graph().as_default() as g:
x = tf.placeholder(tf.float32, [
None, lc_CNN_forward_FC.MAT_SIZE_ROW,
lc_CNN_forward_FC.MAT_SIZE_COL, lc_CNN_forward_FC.NUM_CHANNELS
])
y_ = tf.placeholder(tf.float32, [None, lc_CNN_forward_FC.OUTPUT_NODE])
def train(sel):
import lc_CNN_backward_FC as backward
bw = backward.CNN_backward()
bw.main_train_and_test(sel.x_train, sel.y_train, sel.x_test,
sel.y_test)
def eval_prformance(sel, y_test_one_hot, y_pred, y_pred_prob):
y_test_1d = np.argmax(y_test_one_hot, 1)
accuracy = float('%.4f' % (accuracy_score(y_test_1d, y_pred)))
report = classification_report(y_test_1d, y_pred)
report = report.split('\n')
specificity = report[2].strip().split(' ')
sensitivity = report[3].strip().split(' ')
specificity = float([spe for spe in specificity if spe != ''][2])
sensitivity = float([sen for sen in sensitivity if sen != ''][2])
auc_score = float('%.4f' % roc_auc_score(y_test_one_hot, y_pred_prob))
return accuracy, sensitivity, specificity, auc_score
def main_train_and_test(sel, x_train, y_train):
import time
s = time.time()
loss_train = sel.backward(x_train, y_train)
e = time.time()
print('Running time:{}\n'.format(e - s))
return loss_train
if __name__ == '__main__':
import lc_CNN_backward_FC as backward
sel = backward.CNN_backward()