Fix autoencoder

README.md

@@ -4,10 +4,11 @@
 
 # candock
 
-| English | [中文版](./README_CN.md) |<br>
+| English | [中文版](./README_CN.md) |<br><br>
 A time series signal analysis and classification framework.<br>
 It contain multiple network  and provide data preprocessing, reading, training, evaluation, testing and other functions.<br>
-Some output examples: [heatmap](./image/heatmap_eg.png)  [running_err](./image/running_err_eg.png)  [log.txt](./docs/log_eg.txt)<br>Supported network:<br>
+Some output examples: [heatmap](./image/heatmap_eg.png)  [running_loss](./image/running_loss_eg.png)  [log.txt](./docs/log_eg.txt)<br>
+Supported network:<br>
 
 >1d
 >

README_CN.md

@@ -4,10 +4,10 @@
 
 # candock
 
-| [English](./README.md) | 中文版 |<br>
+| [English](./README.md) | 中文版 |<br><br>
 一个通用的一维时序信号分析,分类框架.<br>
 它将包含多种网络结构，并提供数据预处理,读取,训练,评估,测试等功能.<br>
-一些训练时的输出样例: [heatmap](./image/heatmap_eg.png)  [running_err](./image/running_err_eg.png)  [log.txt](./docs/log_eg.txt)<br>
+一些训练时的输出样例: [heatmap](./image/heatmap_eg.png)  [running_loss](./image/running_loss_eg.png)  [log.txt](./docs/log_eg.txt)<br>
 目前支持的网络结构:<br>
 >1d
 >

models/core.py

@@ -34,15 +34,15 @@ class Core(object):
 
         self.net = creatnet.creatnet(self.opt)
         self.optimizer = torch.optim.Adam(self.net.parameters(), lr=self.opt.lr)
-        self.criterion_class = nn.CrossEntropyLoss(self.opt.weight)
-        self.criterion_auto = nn.MSELoss()
+        self.criterion_classifier = nn.CrossEntropyLoss(self.opt.weight)
+        self.criterion_autoencoder = nn.MSELoss()
         self.epoch = 1
         self.plot_result = {'train':[],'eval':[],'F1':[]}
         self.confusion_mats = []
         self.test_flag = True
 
         if printflag:
-            #util.writelog('network:\n'+str(self.net),self.opt,True)
+            util.writelog('network:\n'+str(self.net),self.opt,True)
             show_paramsnumber(self.net,self.opt)
 
         if self.opt.pretrained != '':
@@ -97,31 +97,33 @@ class Core(object):
     def forward(self,signal,label,features,confusion_mat):
         if self.opt.model_name == 'autoencoder':
             out,feature = self.net(signal)
-            loss = self.criterion_auto(out, signal)
-            features[i*self.opt.batchsize:(i+1)*self.opt.batchsize,:self.opt.feature] = (feature.data.cpu().numpy()).reshape(self.opt.batchsize,-1)
-            features[i*self.opt.batchsize:(i+1)*self.opt.batchsize,self.opt.feature] = label.data.cpu().numpy()
+            loss = self.criterion_autoencoder(out, signal)
+            label = label.data.cpu().numpy()
+            feature = (feature.data.cpu().numpy()).reshape(self.opt.batchsize,-1)
+            for i in range(self.opt.batchsize):
+                features.append(np.concatenate((feature[i], [label[i]])))
         else:
             out = self.net(signal)
-            loss = self.criterion_class(out, label)
+            loss = self.criterion_classifier(out, label)
             pred = (torch.max(out, 1)[1]).data.cpu().numpy()
-            label=label.data.cpu().numpy()
+            label = label.data.cpu().numpy()
             for x in range(len(pred)):
                 confusion_mat[label[x]][pred[x]] += 1
-        return loss,features,confusion_mat
+        return out,loss,features,confusion_mat
 
     def train(self,signals,labels,sequences):
         self.net.train()
         self.test_flag = False
+        features = []
         epoch_loss = 0
         confusion_mat = np.zeros((self.opt.label,self.opt.label), dtype=int)
-        features = np.zeros((len(sequences)//self.opt.batchsize*self.opt.batchsize,self.opt.feature+1))
         
         np.random.shuffle(sequences)
         self.process_pool_init(signals, labels, sequences)
         for i in range(len(sequences)//self.opt.batchsize):
             signal,label = self.queue.get()
             signal,label = transformer.ToTensor(signal,label,gpu_id =self.opt.gpu_id)
-            loss,features,confusion_mat=self.forward(signal, label, features, confusion_mat)
+            output,loss,features,confusion_mat = self.forward(signal, label, features, confusion_mat)
 
             epoch_loss += loss.item()     
             self.optimizer.zero_grad()
@@ -136,27 +138,27 @@ class Core(object):
 
     def eval(self,signals,labels,sequences):
         self.test_flag = True
-        confusion_mat = np.zeros((self.opt.label,self.opt.label), dtype=int)
-        features = np.zeros((len(sequences)//self.opt.batchsize*self.opt.batchsize,self.opt.feature+1))
+        features = []
         epoch_loss = 0
+        confusion_mat = np.zeros((self.opt.label,self.opt.label), dtype=int)
 
         self.process_pool_init(signals, labels, sequences)
         for i in range(len(sequences)//self.opt.batchsize):
             signal,label = self.queue.get()
             signal,label = transformer.ToTensor(signal,label,gpu_id =self.opt.gpu_id)
             with torch.no_grad():
-                loss,features,confusion_mat = self.forward(signal, label, features, confusion_mat)
+                output,loss,features,confusion_mat = self.forward(signal, label, features, confusion_mat)
                 epoch_loss += loss.item()
 
-        if self.opt.model_name != 'autoencoder':
+        if self.opt.model_name == 'autoencoder':
+            plot.draw_autoencoder_result(signal.data.cpu().numpy(), output.data.cpu().numpy(),self.opt)
+            print('epoch:'+str(self.epoch),' loss: '+str(round(epoch_loss/i,5)))
+            plot.draw_scatter(features, self.opt)
+        else:
             recall,acc,sp,err,k  = statistics.report(confusion_mat)         
             #plot.draw_heatmap(confusion_mat,self.opt,name = 'current_eval')
             print('epoch:'+str(self.epoch),' macro-prec,reca,F1,err,kappa: '+str(statistics.report(confusion_mat)))
             self.plot_result['F1'].append(statistics.report(confusion_mat)[2])
-        else:
-            plot.draw_autoencoder_result(signal.data.cpu().numpy(), out.data.cpu().numpy(),self.opt)
-            print('epoch:'+str(self.epoch),' loss: '+str(round(epoch_loss/i,5)))
-            plot.draw_scatter(features, self.opt)
         
         self.plot_result['eval'].append(epoch_loss/i) 
 

tools/client.py

@@ -45,7 +45,7 @@ return: {'return' : 'done',
         }
 """
 data = {'token':opt.token,'mode': 'train'}
-r = requests.post(opt.url, data)
+r = requests.post(opt.url, data ,timeout=60)
 rec_data = r.json()
 print(rec_data['report'])
 

tools/server.py

@@ -32,6 +32,7 @@ core.network_init(printflag=True)
 
 # -----------------------------train-----------------------------
 def train(opt):
+    core.network_init(printflag=True)
 
     categorys = os.listdir(opt.rec_tmp)
     categorys.sort()

util/dsp.py

@@ -94,15 +94,6 @@ def signal2spectrum(data,window_size, stride, n_downsample=1, log = True, log_al
         spectrum = spectrum_new
         spectrum = (spectrum-0.05)/0.25
 
-        # spectrum = np.log1p(spectrum)
-        # h = window_size//2+1
-        # tmp = np.linspace(0, h-1,num=h,dtype=np.int64)
-        # index = np.log2(tmp+1)*(h/np.log2(h+1))
-        # spectrum_new = np.zeros_like(spectrum)
-        # for i in range(h-1):
-        #     spectrum_new[int(index[i]):int(index[i+1])] = spectrum[i]
-        # spectrum = spectrum_new
-        # spectrum = (spectrum-0.05)/0.25
     else:
         spectrum = (spectrum-0.02)/0.05
 

util/plot.py

@@ -9,11 +9,11 @@ markers = ['o','^','.',',','v','<','>','1','2','3','4','s','p','*','h','H','+','
 
 #---------------------------------heatmap---------------------------------
 
-'''
+"""
 heatmap: https://matplotlib.org/gallery/images_contours_and_fields/image_annotated_heatmap.html#sphx-glr-gallery-images-contours-and-fields-image-annotated-heatmap-py
 choose color:https://matplotlib.org/tutorials/colors/colormaps.html?highlight=wistia
       recommend:  YlGn  Wistia Blues YlOrBr
-'''
+"""
 def create_heatmap(data, row_labels, col_labels, ax=None,
             cbar_kw={}, cbarlabel="", **kwargs):
     """
@@ -193,6 +193,8 @@ def label_statistics(labels):
     return label_cnt,label_cnt_per,label_num
 
 def draw_scatter(data,opt):
+    data = np.array(data)
+    data = data[np.argsort(data[:,-1])]
     label_cnt,_,label_num = label_statistics(data[:,-1])
     fig = plt.figure(figsize=(12,9))
     cnt = 0
@@ -205,23 +207,36 @@ def draw_scatter(data,opt):
         data_dimension = 3
     
     if data_dimension == 2:
+        plt.xlim(-1.5,1.5)
+        plt.ylim(-1.5,1.5)
         for i in range(label_num):
-            plt.scatter(data[cnt:cnt+label_cnt[i],0], data[cnt:cnt+label_cnt[i],1],
+            plt.scatter(
+                (data[cnt:cnt+label_cnt[i],0])[:100], 
+                (data[cnt:cnt+label_cnt[i],1])[:100],
+                label=str(i),
             )
             cnt += label_cnt[i]
 
-
     elif data_dimension == 3:
         ax = fig.add_subplot(111, projection='3d')
+        ax.set_zlim3d(-1.5, 1.5)
+        ax.set_ylim3d(-1.5, 1.5)
+        ax.set_xlim3d(-1.5, 1.5)
         for i in range(label_num):
-            ax.scatter(data[cnt:cnt+label_cnt[i],0], data[cnt:cnt+label_cnt[i],1], data[cnt:cnt+label_cnt[i],2],
+            ax.scatter(
+                (data[cnt:cnt+label_cnt[i],0])[:100], 
+                (data[cnt:cnt+label_cnt[i],1])[:100], 
+                (data[cnt:cnt+label_cnt[i],2])[:100],
+                label=str(i),
             )
             cnt += label_cnt[i]
-
+    plt.title('Autoencoder Embedding Result')
+    plt.legend(loc=2)
     plt.savefig(os.path.join(opt.save_dir,'feature_scatter.png'))
     np.save(os.path.join(opt.save_dir,'feature_scatter.npy'), data)
     plt.close('all')
 
+
 def draw_autoencoder_result(true_signal,pred_signal,opt):
     plt.subplot(211)
     plt.plot(true_signal[0][0])