Initial commit

lib/metrics.py

@@ -89,6 +89,7 @@ def masked_mae_np(preds, labels, null_val=np.nan):
         mask /= np.mean(mask)
         mae = np.abs(np.subtract(preds, labels)).astype('float32')
         mae = np.nan_to_num(mae * mask)
+        # print(mae)
         return np.mean(mae)
 
 
@@ -104,7 +105,39 @@ def masked_mape_np(preds, labels, null_val=np.nan):
         mape = np.nan_to_num(mask * mape)
         return np.mean(mape)
 
-
+def masked_pcc(preds, labels, null_val=np.nan):
+    with np.errstate(divide='ignore', invalid='ignore'):
+        if np.isnan(null_val):
+            mask = ~np.isnan(labels)
+        else:
+            mask = np.not_equal(labels, null_val)
+        mask = mask.astype('float32')
+        mask /= np.mean(mask)
+        # pred = preds.cpu().detach().numpy()
+        # label = labels.cpu().detach().numpy()
+        # print(labels.shape)
+        # print(preds.shape)
+        labels = np.nan_to_num(mask * labels)
+        preds = np.nan_to_num(mask *  preds)
+        # pcc = np.corrcoef(preds,labels)
+        # print(pcc.shape)
+        # pred,label = preds.reshape(-1),labels.reshape(-1)
+        preds=preds.T
+        labels=labels.T
+        # pcc = np.corrcoef(preds,labels)
+        # n = pcc.shape[0]//2
+        # for i in range(n):
+        #     print(pcc[i][i+n])
+        # print(pcc.shape)
+        # print(pcc)
+        
+        pred,label = preds.reshape(-1),labels.reshape(-1)
+        print(pred.shape)
+        print(label.shape)
+        return np.corrcoef(pred,label)[0][1]
+
+
+        
 # Builds loss function.
 def masked_mse_loss(scaler, null_val):
     def loss(preds, labels):

lib/utils.py

@@ -196,7 +196,8 @@ def count_parameters(model):
 def load_dataset(dataset_dir, batch_size, test_batch_size=None, **kwargs):
     data = {}
     for category in ['train', 'val', 'test']:
-        cat_data = np.load(os.path.join(dataset_dir, category + '.npz'))
+        cat_data = np.load(os.path.join(dataset_dir,'client_2_' + category + '.npz'))
+        # cat_data = np.load(os.path.join(dataset_dir, category + '.npz'))
         data['x_' + category] = cat_data['x']
         data['y_' + category] = cat_data['y']
     scaler = StandardScaler(mean=data['x_train'][..., 0].mean(), std=data['x_train'][..., 0].std())

model/dcrnn_cell.py

@@ -40,9 +40,12 @@ class DiffusionGraphConv(BaseModel):
         # Reshape input and state to (batch_size, num_nodes, input_dim/state_dim)
         batch_size = inputs.shape[0]
         inputs = torch.reshape(inputs, (batch_size, self._num_nodes, -1))
+        # print(inputs.shape)
         state = torch.reshape(state, (batch_size, self._num_nodes, -1))
+        # print(state.shape)
         inputs_and_state = torch.cat([inputs, state], dim=2)
         input_size = inputs_and_state.shape[2]
+        # print(input_size)
         # dtype = inputs.dtype
 
         x = inputs_and_state
@@ -61,11 +64,14 @@ class DiffusionGraphConv(BaseModel):
                     x2 = 2 * torch.sparse.mm(support, x1) - x0
                     x = self._concat(x, x2)
                     x1, x0 = x2, x1
-
+        # print(x.shape)
         x = torch.reshape(x, shape=[self.num_matrices, self._num_nodes, input_size, batch_size])
+        # print(x.shape)
         x = torch.transpose(x, dim0=0, dim1=3)  # (batch_size, num_nodes, input_size, order)
+        # print(x.shape)
         x = torch.reshape(x, shape=[batch_size * self._num_nodes, input_size * self.num_matrices])
-
+        # print(x.shape)
+        # print(self.weight.shape)
         x = torch.matmul(x, self.weight)  # (batch_size * self._num_nodes, output_size)
         x = torch.add(x, self.biases)
         # Reshape res back to 2D: (batch_size, num_node, state_dim) -> (batch_size, num_node * state_dim)