【AI达人创造营第二期】电机异音AI诊断

! unzip -oq data/data31822/Motor_tain.zip! unzip -oq data/data31822/Motor_testP.zip

import os

train_pos_dir='Motor_tain/╒¤╤∙▒╛/'train_neg_dir='Motor_tain/╕║╤∙▒╛/'# 生成三份文件便于按照类别比例划分with open('train_pos.txt','w') as f:    for item in list(set([item[:len(item)-6] for item in os.listdir(train_pos_dir)])):        if '.ipynb' not in item:
            f.write(train_pos_dir+item+'\t1\n')with open('train_neg.txt','w') as f:    for item in list(set([item[:len(item)-6] for item in os.listdir(train_neg_dir)])):        if '.ipynb' not in item:
            f.write(train_neg_dir+item+'\t0\n')with open('train.txt','w') as f:    for item in list(set([item[:len(item)-6] for item in os.listdir(train_pos_dir)])):        if '.ipynb' not in item:            for i in range(int(500/30)):
                f.write(train_pos_dir+item+'\t1\n')    for item in list(set([item[:len(item)-6] for item in os.listdir(train_neg_dir)])):        if '.ipynb' not in item:
            f.write(train_neg_dir+item+'\t0\n')

import paddleimport numpy as npimport paddle.vision.transforms as Tfrom PIL import Imageimport pandas as pdclass MyDateset(paddle.io.Dataset):
    def __init__(self,txt_dir):
        super(MyDateset, self).__init__()
        
        self.path=[]
        self.label=[]        with open(txt_dir,'r') as f:            for line in f.readlines():
                self.path.append(line.split('\t')[0])
                self.label.append(line.split('\t')[1][0])    def __getitem__(self, index):
        path = self.path[index]
        label = int(self.label[index])

        F=pd.read_csv(path+'_F.csv')
        B=pd.read_csv(path+'_B.csv')

        F=paddle.to_tensor(F.values).flatten(0).astype('float32')
        B=paddle.to_tensor(B.values).flatten(0).astype('float32')

        label = np.array(label).astype('int64')        return F,B,label    def __len__(self):
        return len(self.label)

train_dataset=MyDateset('train.txt')

train_dataloader = paddle.io.DataLoader(
    train_dataset,
    batch_size=16,
    shuffle=True,
    drop_last=False)for step, data in enumerate(train_dataloader):
    F, B, label = data    print(step, F.shape, B.shape, label.shape)    break

class MyNet(paddle.nn.Layer):
    def __init__(self):
        super(MyNet,self).__init__()
        self.fc_F = paddle.nn.Linear(in_features=159998, out_features=2000)
        self.fc_B = paddle.nn.Linear(in_features=159998, out_features=2000)

        self.fc_1  = paddle.nn.Linear(in_features=4000, out_features=1000)
        self.fc_2  = paddle.nn.Linear(in_features=1000, out_features=200)
        self.fc_3  = paddle.nn.Linear(in_features=200, out_features=2)    def forward(self,F,B):
        F = self.fc_F(F)
        B = self.fc_B(B)
        x = paddle.concat([F,B],axis=-1)
        x = self.fc_1(x)        # x = paddle.nn.functional.relu(x)
        x = self.fc_2(x)        # x = paddle.nn.functional.relu(x)
        x = self.fc_3(x)        return x

model = MyNet()
model.train()

max_epoch=10opt = paddle.optimizer.SGD(learning_rate=0.001, parameters=model.parameters())

now_step=0for epoch in range(max_epoch):    for step, data in enumerate(train_dataloader):
        now_step+=1

        F, B, label = data
        pre = model(F,B)
        loss = paddle.nn.functional.cross_entropy(pre,label,weight=paddle.to_tensor([1,1.1]),reduction='mean')
        loss.backward()
        opt.step()
        opt.clear_gradients()        if now_step%100==0:            print("epoch: {}, batch: {}, loss is: {}".format(epoch, step, loss.mean().numpy()))

paddle.save(model.state_dict(), 'model.pdparams')

# mydict = paddle.load("model_1.pdparams")# model.set_state_dict(mydict)record=np.zeros([2,2])for i in range(len(train_dataset)):
    F,B,label=train_dataset[i]
    pre=model(F,B)    # print(f'real label: {label} pre label: {np.argmax(pre.numpy())}')
    record[label.tolist()][np.argmax(pre.numpy())]+=1record