mxnet卷积神经网络训练MNIST数据集测试

mxnet框架下超全手写字体识别—从数据预处理到网络的训练—模型及日志的保存

import numpy as np
import mxnet as mx
import logging  

logging.getLogger().setLevel(logging.DEBUG)  

batch_size = 100
mnist = mx.test_utils.get_mnist()
train_iter = mx.io.NDArrayIter(mnist['train_data'], mnist['train_label'], batch_size, shuffle=True)
val_iter = mx.io.NDArrayIter(mnist['test_data'], mnist['test_label'], batch_size)  

data = mx.sym.var('data')
# first conv layer
conv1= mx.sym.Convolution(data=data, kernel=(5,5), num_filter=20)
tanh1= mx.sym.Activation(data=conv1, act_type="tanh")
pool1= mx.sym.Pooling(data=tanh1, pool_type="max", kernel=(2,2), stride=(2,2))
# second conv layer
conv2= mx.sym.Convolution(data=pool1, kernel=(5,5), num_filter=50)
tanh2= mx.sym.Activation(data=conv2, act_type="tanh")
pool2= mx.sym.Pooling(data=tanh2, pool_type="max", kernel=(2,2), stride=(2,2))
# first fullc layer
flatten= mx.sym.Flatten(data=pool2)
fc1= mx.symbol.FullyConnected(data=flatten, num_hidden=500)
tanh3= mx.sym.Activation(data=fc1, act_type="tanh")
# second fullc
fc2= mx.sym.FullyConnected(data=tanh3, num_hidden=10)
# softmax loss
lenet= mx.sym.SoftmaxOutput(data=fc2, name='softmax')  

# create a trainable module on GPU 0
lenet_model = mx.mod.Module(
                symbol=lenet,
                context=mx.cpu())  

# train with the same
lenet_model.fit(train_iter,
                eval_data=val_iter,
                optimizer='sgd',
                optimizer_params={'learning_rate':0.1},
                eval_metric='acc',
                batch_end_callback = mx.callback.Speedometer(batch_size, 100),
                num_epoch=10)  

INFO:root:Epoch[0] Batch [100] Speed: 1504.57 samples/sec accuracy=0.113564
INFO:root:Epoch[0] Batch [200] Speed: 1516.40 samples/sec accuracy=0.118100
INFO:root:Epoch[0] Batch [300] Speed: 1515.71 samples/sec accuracy=0.116600
INFO:root:Epoch[0] Batch [400] Speed: 1505.61 samples/sec accuracy=0.110200
INFO:root:Epoch[0] Batch [500] Speed: 1406.21 samples/sec accuracy=0.107600
INFO:root:Epoch[0] Train-accuracy=0.108081
INFO:root:Epoch[0] Time cost=40.572
INFO:root:Epoch[0] Validation-accuracy=0.102800
INFO:root:Epoch[1] Batch [100] Speed: 1451.87 samples/sec accuracy=0.115050
INFO:root:Epoch[1] Batch [200] Speed: 1476.86 samples/sec accuracy=0.179600
INFO:root:Epoch[1] Batch [300] Speed: 1409.67 samples/sec accuracy=0.697100
INFO:root:Epoch[1] Batch [400] Speed: 1379.52 samples/sec accuracy=0.871900
INFO:root:Epoch[1] Batch [500] Speed: 1374.88 samples/sec accuracy=0.901000
INFO:root:Epoch[1] Train-accuracy=0.925556
INFO:root:Epoch[1] Time cost=42.527
INFO:root:Epoch[1] Validation-accuracy=0.936900
INFO:root:Epoch[2] Batch [100] Speed: 1376.59 samples/sec accuracy=0.936436
INFO:root:Epoch[2] Batch [200] Speed: 1379.29 samples/sec accuracy=0.948100
INFO:root:Epoch[2] Batch [300] Speed: 1375.07 samples/sec accuracy=0.953400
INFO:root:Epoch[2] Batch [400] Speed: 1369.65 samples/sec accuracy=0.958600
INFO:root:Epoch[2] Batch [500] Speed: 1371.79 samples/sec accuracy=0.960900
INFO:root:Epoch[2] Train-accuracy=0.966667
INFO:root:Epoch[2] Time cost=43.660
INFO:root:Epoch[2] Validation-accuracy=0.972900
INFO:root:Epoch[3] Batch [100] Speed: 1230.74 samples/sec accuracy=0.969505
INFO:root:Epoch[3] Batch [200] Speed: 1335.27 samples/sec accuracy=0.970800
INFO:root:Epoch[3] Batch [300] Speed: 1264.43 samples/sec accuracy=0.972600
INFO:root:Epoch[3] Batch [400] Speed: 1242.03 samples/sec accuracy=0.974100
INFO:root:Epoch[3] Batch [500] Speed: 1322.77 samples/sec accuracy=0.974600
INFO:root:Epoch[3] Train-accuracy=0.976465
INFO:root:Epoch[3] Time cost=46.860
INFO:root:Epoch[3] Validation-accuracy=0.980700
INFO:root:Epoch[4] Batch [100] Speed: 1342.42 samples/sec accuracy=0.978020
INFO:root:Epoch[4] Batch [200] Speed: 1339.98 samples/sec accuracy=0.980600
INFO:root:Epoch[4] Batch [300] Speed: 1344.36 samples/sec accuracy=0.981000
INFO:root:Epoch[4] Batch [400] Speed: 1338.13 samples/sec accuracy=0.980000
INFO:root:Epoch[4] Batch [500] Speed: 1343.76 samples/sec accuracy=0.979000
INFO:root:Epoch[4] Train-accuracy=0.983535
INFO:root:Epoch[4] Time cost=44.694
INFO:root:Epoch[4] Validation-accuracy=0.985700
INFO:root:Epoch[5] Batch [100] Speed: 1333.50 samples/sec accuracy=0.981584
INFO:root:Epoch[5] Batch [200] Speed: 1342.07 samples/sec accuracy=0.985400
INFO:root:Epoch[5] Batch [300] Speed: 1339.04 samples/sec accuracy=0.984300
INFO:root:Epoch[5] Batch [400] Speed: 1323.42 samples/sec accuracy=0.983500