def train(epoch):
model.train()
loss_queue = [0] * display #存储
correct_num_queue = [0] * display
for i, (images, labels) in enumerate(train_loader):
# 把训练数据和标签放到cuda
images = images.cuda()
labels = labels.cuda()
# 清除累积梯度
optimizer.zero_grad()
#前向传播
outputs = model(images)
#计算loss
loss = loss_fn(outputs, labels)
#存储loss
loss_queue.append(loss.data.item())
loss_queue.pop(0)
#反向传播,计算出梯度矩阵
loss.backward()
if args.sparsity_lambda:
for m in model.modules():
if isinstance(m, nn.BatchNorm2d):
#将γ的原本梯度矩阵加上由L1稀疏化引入的梯度
m.weight.grad.data.add_(args.sparsity_lambda *
torch.sign(m.weight.data))
#根据梯度更新参数
optimizer.step()
#当前正确个数
_, prediction = torch.max(outputs.data, 1)
current_correct_num = torch.sum(prediction == labels.data)
correct_num_queue.append(current_correct_num)
correct_num_queue.pop(0)
if i != 0 and i % display == 0:
print_and_write(
current_time() + ' Epoch {}, Iteration {}, lr = {}'.format(
epoch, i, optimizer.param_groups[0]['lr']), log_file)
#print_and_write(current_time()+' Training loss cur = {:.3f}'.format(loss.data.item()), log_file)
print_and_write(
current_time() +
' Training loss ave = {:.3f}'.format(mean(loss_queue)),
log_file)
#print_and_write(current_time()+' Training accu cur = {:.3f}'.format(current_correct_num/batch_size), log_file)
print_and_write(
current_time() + ' Training accu ave = {:.3f}'.format(
sum(correct_num_queue) / (batch_size * display)), log_file)
def update_user(self, id, name, key, table="users"):
conditions = {"id": id}
user = {"name": name}
current_time = common.current_time()
user["timestamp"] = current_time
user["key"] = key
self.update(table=table, values=user, conditions=conditions)
def update_datas(self,
id,
name,
conditions,
value,
condition_extension={},
value_extension={}):
# 出去变化值之外的所有条件值
conditions.update({"id": id, "name": name})
if condition_extension:
conditions["condition_extension"] = common.gen_json_str(
condition_extension)
# 变化值
current_time = common.current_time()
uuid = "%s-%s" % (hash_utils.hash(
conditions, length=8), hash_utils.hash(current_time, length=8))
values = {"uuid": uuid}
values["timestamp"] = current_time
values["value"] = value
if value_extension:
values["value_extension"] = common.gen_json_str(value_extension)
self.update(table=self.table_datas,
values=values,
conditions=conditions)
values.update(conditions)
self.insert(table=self.table_history, values=values)
def user():
# 获取cookie中的值，已登录用户
id = request.cookies.get('id')
name = request.cookies.get('name')
form_update_data = UpdateData()
if form_update_data.validate_on_submit():
common.print_info("提交表单数据")
city = form_update_data.city.data
project = form_update_data.project.data
value = form_update_data.value.data
common.print_info("city: %s, project: %s, value: %s" %
(city, project, value))
conditions = {"city": city, "project": project}
config.update_datas(id, name, conditions, value)
common.print_info("get id:%s, name:%s from cookies" % (id, name))
# return render_template('user.html', id=id, name=name, form=form_update_data)
if request.method == 'POST':
file = request.files['file']
filename = file.filename
common.print_info("上传文件: %s" % filename)
try:
# file.save(os.path.join(app.config['UPLOAD_FOLDER'], secure_filename(filename)))
excel = os.path.join(
app.config['UPLOAD_FOLDER'],
"%s_%s_%s_%s" % (common.current_time(fmt='%Y%m%d-%H%M%S-%f'),
id, name, filename))
file.save(excel)
# 检查
ret, info = excel_check(excel, titles_std=titles_std)
if ret:
project_name, date_power = info
flash("%s 上传成功，项目名称：%s，时间力度：%s" %
(filename, project_name, date_power))
# 重命名文件
filename_new = "%s_%s_%s_%s" % (date_power, project_name, id,
name)
excel_new = file_utils.replace_filename(excel, filename_new)
file_utils.copy_file(excel, excel_new)
# 源文件备份
excel_backup = os.path.join("backups", os.path.basename(excel))
file_utils.move_file(excel, excel_backup)
else:
flash(info)
except Exception as e:
flash("ERROR! %s" % e)
# return render_template('user.html', form=form_update_data)
return render_template('user.html',
id=id,
name=name,
form=form_update_data)
info.get(subsidy_str), info.get(gmv_str),
",".join(tmp))
csv_contents.append(line)
for city, info in result.citys.items():
line = "%s,%s,%s,%s,%s" % (city, info.get(battle_zone_str),
info.get(needed_date),
info.get(subsidy_str), info.get(gmv_str))
values = []
for project_name in result.project_names:
project_value = info.get("projects").get(project_name)
values.append(
"%s,%s" %
(project_value.get("power"), project_value.get("subsidy")))
line = "%s,%s" % (line, ",".join(values))
csv_contents.append(line)
csv_contents = "\n".join(csv_contents)
csv_path = os.path.join(
excel_path,
"result_%s.csv" % common.current_time(fmt='%Y%m%d_%H%M%S%f'))
with open(csv_path, "w") as f:
f.write(csv_contents)
print("")
print("")
print("==============================================")
print("")
print("报表输出到：\n%s" % os.path.abspath(csv_path))
print("")
print("==============================================")
# return x
# model = torchvision.models.resnet18(num_classes=10)
# model.conv1 = nn.Conv2d(3,64,kernel_size=3, stride=1, padding=1)
# model._forward_impl = types.MethodType(_forward_impl,model)
# 2.mobilenet_v2
# model = torchvision.models.mobilenet_v2(num_classes=200)
# 3.vgg19_bn去掉多余全连接
model = torchvision.models.vgg19_bn(num_classes=10)
model.avgpool = nn.AdaptiveAvgPool2d((1, 1))
model.classifier = nn.Linear(512, 10)
print_and_write(
current_time() + ' Epochs:{}, Batch size:{}, Init lr:{}'.format(
max_epoch, batch_size, base_lr), log_file)
#如果有指定预训练模型
if not args.pretrain_weights == '0':
print_and_write(
current_time() +
' Loading from {}'.format(args.pretrain_weights), log_file)
pretrained_dict = torch.load(args.pretrain_weights)
model_dict = model.state_dict()
pretrained_dict = {
k: v
for k, v in pretrained_dict.items() if k in model_dict
} #过滤掉不用的层
pretrained_dict = {
k: v