eval code update

utils/TSTR.py

@@ -4,13 +4,14 @@
 from sklearn.metrics import accuracy_score, mean_squared_error
 
 class GeneralRNN(torch.nn.Module):
-    r"""A general RNN model for time-series prediction
+    """
+    A general RNN model for time-series prediction
     """
 
     def __init__(self, args):
         super(GeneralRNN, self).__init__()
-        self.model_type = args['model_type']
 
+        self.model_type = args['model_type']
         self.input_size = args['in_dim']
         self.hidden_size = args['h_dim']
         self.output_size = args['out_dim']
@@ -86,13 +87,13 @@ def one_step_ahead_prediction(train_data, test_data):
     args["task"] = "regression"
     args["model_type"] = "gru"
     args["bidirectional"] = False
-    args["epochs"] = 20
+    args["epochs"] = 2000
     args["batch_size"] = 256
     args["in_dim"] = dim
-    args["h_dim"] = dim
+    args["h_dim"] = 30 # dim
     args["out_dim"] = dim
     args["n_layers"] = 3
-    args["dropout"] = 0.5
+    args["dropout"] = 0.3
     args["max_seq_len"] = 30  # only 29 is used for prediction
     args["learning_rate"] = 1e-3
     args["grad_clip_norm"] = 5.0
@@ -110,6 +111,7 @@ def one_step_ahead_prediction(train_data, test_data):
         batch_size=no,
         shuffle=True
     )
+
     # Initialize model
     model = GeneralRNN(args)
     model.to(args["device"])
@@ -142,14 +144,25 @@ def one_step_ahead_prediction(train_data, test_data):
 
     # Evaluate the trained model
     with torch.no_grad():
-        perf = 0
+        perf = np.zeros((8789,2668))
+        ind = 0
         for test_x in test_dataloader:
             test_x = test_x.to(args["device"])
             test_p = model(test_x[:,0:29,:]).cpu()
 
-            test_p = np.reshape(test_p.numpy(), [-1])
-            test_y = np.reshape(test_x[:,1:30,:].cpu().numpy(), [-1])
-
-            perf += rmse_error(test_y, test_p)
-
-    return perf
+            test_p = np.reshape(test_p.numpy(), [test_p.numpy().shape[0],-1])
+            test_y = np.reshape(test_x[:,1:30,:].cpu().numpy(), [test_x[:,1:30,:].cpu().numpy().shape[0],-1])
+
+            perf[ind:ind+test_x[:,1:30,:].cpu().numpy().shape[0]] = test_p-test_y
+            ind = ind+test_x[:,1:30,:].cpu().numpy().shape[0]
+            #perf += rmse_error(test_y, test_p)
+            #perf2.append(rmse_error(test_y, test_p))
+
+        mae = np.abs(perf).mean(1)
+        rmse = (perf * perf).mean(1)
+        rmse = np.sqrt(rmse)
+
+        print(f'MAE : {mae.mean(0)}')
+        print(f'RMSE : {rmse.mean(0)}')
+
+    return perf, rmse.mean(0), mae.mean(0)

utils/visual_vrae.py

@@ -5,7 +5,7 @@
 import pandas as pd
 
 
-def visualization(original_data, generated_data, analysis):
+def visualization(original_data, generated_data, analysis, name, x_zoom=False, y_zoom=False):
     """Using PCA or tSNE for generated and original data visualization.
 
     Args:
@@ -44,14 +44,18 @@ def visualization(original_data, generated_data, analysis):
 
         ax.legend()
 
-        plt.title('PCA plot')
-        plt.xlabel('x-pca')
-        plt.ylabel('y-pca')
+        if x_zoom != False and y_zoom != False:
+            plt.xlim(x_zoom[0],x_zoom[1])
+            plt.ylim(y_zoom[0],y_zoom[1])
+
+        plt.title(f'{name} PCA')
+        plt.xlabel('PC_1')
+        plt.ylabel('PC_2')
         plt.show()
 
     elif analysis == 'tsne':
         # TSNE Analysis
-        tsne = TSNE(n_components=2, verbose=1, perplexity=40, n_iter=300)
+        tsne = TSNE(n_components=2, verbose=1, perplexity=50, n_iter=1000)
 
         # conat data
         concat_data = np.concatenate((original_data, generated_data), axis=0)
@@ -69,7 +73,11 @@ def visualization(original_data, generated_data, analysis):
 
         ax.legend()
 
-        plt.title('tsne plot')
+        if x_zoom != False and y_zoom != False:
+            plt.xlim(0,x_zoom)
+            plt.ylim(0,y_zoom)
+
+        plt.title(f'{name} T-SNE')
         plt.xlabel('x-tsne')
         plt.ylabel('y-tsne')
         plt.show()

vrae_experiment.sh

@@ -5,10 +5,10 @@
 python run_vrae.py --scale_type 'Standard' --hidden_layer_depth 1
 python run_vrae.py --scale_type 'Standard' --hidden_layer_depth 2
 
-# MinMax
-python run_vrae.py --scale_type 'MinMax' --hidden_layer_depth 1
-python run_vrae.py --scale_type 'MinMax' --hidden_layer_depth 2
+# # MinMax
+# python run_vrae.py --scale_type 'MinMax' --hidden_layer_depth 1
+# python run_vrae.py --scale_type 'MinMax' --hidden_layer_depth 2
 
-# Robust
-python run_vrae.py --scale_type 'Robust' --hidden_layer_depth 1
-python run_vrae.py --scale_type 'Robust' --hidden_layer_depth 2
+# # Robust
+# python run_vrae.py --scale_type 'Robust' --hidden_layer_depth 1
+# python run_vrae.py --scale_type 'Robust' --hidden_layer_depth 2