mxnet-保存模型參數
阿新 • • 發佈：2018-11-14
pos war init rec .sh 模型 time urn uil
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Fri Aug 10 16:13:29 2018
@author: myhaspl
"""
import mxnet as mx
import mxnet.ndarray as nd
from mxnet import nd, autograd, gluon
from mxnet.gluon.data.vision import datasets, transforms
from time import time
def build_lenet(net):
with net.name_scope():
net.add(gluon.nn.Conv2D(channels=6,kernel_size=5,activation="relu"),
gluon.nn.MaxPool2D(pool_size=2, strides=2),
gluon.nn.Conv2D(channels=16, kernel_size=3, activation="relu"),
gluon.nn.MaxPool2D(pool_size=2, strides=2),
gluon.nn.Flatten(),
gluon.nn.Dense(120, activation="relu"),
gluon.nn.Dense(84, activation="relu"),
gluon.nn.Dense(10))
return net
mnist_train = datasets.FashionMNIST(train=True)
X, y = mnist_train[0]
print (‘X shape: ‘, X.shape, ‘X dtype‘, X.dtype, ‘y:‘, y,‘Y dtype‘, y.dtype)
#x:(height, width, channel)
#y:numpy.scalar，標簽
text_labels = [
‘t-shirt‘, ‘trouser‘, ‘pullover‘, ‘dress‘, ‘coat‘,
‘sandal‘, ‘shirt‘, ‘sneaker‘, ‘bag‘, ‘ankle boot‘
]
#轉換圖像為(channel, height, weight)格式，並且為floating數據類型，通過transforms.ToTensor。
#另外，normalize所有像素值 使用 transforms.Normalize平均值0.13和標準差0.31.
transformer = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(0.13, 0.31)])
#只轉換第一個元素，圖像部分。第二個元素為標簽。
mnist_train = mnist_train.transform_first(transformer)
#加載批次數據
batch_size = 200
train_data = gluon.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True)
#讀取本批數據
i=1
for data, label in train_data:
print i
print data,label
break#沒有這一行，會以每批次200個數據來讀取。
mnist_valid = gluon.data.vision.FashionMNIST(train=False)
valid_data = gluon.data.DataLoader(mnist_valid.transform_first(transformer),batch_size=batch_size)
#定義網絡
net = build_lenet(gluon.nn.Sequential())
net.initialize(init=mx.init.Xavier())
print net
#輸出softmax與誤差
softmax_cross_entropy = gluon.loss.SoftmaxCrossEntropyLoss()
#定義訓練器
trainer = gluon.Trainer(net.collect_params(), ‘sgd‘, {‘learning_rate‘: 0.1})
def acc(output, label):
# output: (batch, num_output) float32 ndarray
# label: (batch, ) int32 ndarray
return (output.argmax(axis=1) == label.astype(‘float32‘)).mean().asscalar()
for epoch in range(10):
train_loss, train_acc, valid_acc = 0., 0., 0.
tic = time()
for data, label in train_data:
# 前向與反饋
with autograd.record():
output = net(data)
loss = softmax_cross_entropy(output, label)
loss.backward()
# 換一批樣本數據，更新參數
trainer.step(batch_size)
# 計算訓練誤差和正確率
train_loss += loss.mean().asscalar()
train_acc += acc(output, label)
print "."
#測試正確率
for data, label in valid_data:
predict_data=net(data)
valid_acc += acc(predict_data, label)
print("Epoch %d: Loss: %.3f, Train acc %.3f, Test acc %.3f, \ Time %.1f sec" % (
epoch, train_loss/len(train_data),
train_acc/len(train_data),
valid_acc/len(valid_data), time()-tic))
#保存模型參數,非模型結構
file_name = "net.params"
net.save_parameters(file_name)mxnet-保存模型參數