-
Rainbow-66
cost曲线和准确率曲线贴图如下,蓝线为训练集,橙线为测试集!
运行环境:
python==3.6
Tensorflow==1.8.0
出图程序是自己加的,大致如下是否打印成本
if print_cost: # 每5代打印一次损失值和训练集准确度 if epoch % 5 == 0: # 当前的预测情况 corrent_prediction = tf.equal(tf.argmax(Z3, 1), tf.argmax(Y, 1)) # 当前的准确度 corrent_accuracy = tf.reduce_mean(tf.cast(corrent_prediction, "float")) Ac_train = corrent_accuracy.eval({X: X_train, Y: Y_train}) # 当前训练集准确度 Ac_test = corrent_accuracy.eval({X: X_test, Y: Y_test}) # 当前测试集准确度 # 当前测试集损失值 test_cost = sess.run(cost, feed_dict={X: X_test, Y: Y_test}) print("当前是第 " + str(epoch) + " 迭代,成本值为:" + str(minibatch_cost) + "训练集准确度为:" + str(Ac_train)) # 记录成本 if epoch % 5 == 0: costs.append(minibatch_cost) accuracys_train.append(Ac_train) accuracys_test.append(Ac_test) test_costs.append(test_cost)
如图,测试集的cost值在增大,而准确率在缓慢上升,是数据集的问题还是模型的问题,还是出图程序的问题?
提前感激大佬了!!!