add ground truth plot

explainers.py

@@ -69,7 +69,6 @@ def _attribute_by_index(
                 )['mu_star'].data
                 attr[:, pred_index, t] = torch.tensor(morris_index, device=device)
 
-        print(attr.shape)
         return attr
 
     def attribute(

utils/interpreter.py

@@ -56,7 +56,7 @@ def batch_compute_attr(
         baselines = get_baseline(inputs, mode=baseline_mode)
 
         # get attributions
-        attr = compute_attr(
+        attr = compute_regressor_attr(
             inputs, baselines, explainer, additional_forward_args, exp.args
         )
         attr_list.append(attr)
@@ -155,62 +155,6 @@ def reshape_over_output_horizon(attr, inputs, args):
 
     return attr
 
-def compute_attr(
-    inputs, baselines, explainer,
-    additional_forward_args, args
-):
-    assert type(inputs) == torch.Tensor, \
-        f'Only input type tensor supported, found {type(inputs)} instead.'
-    name = explainer.get_name()
-
-    # these methods don't support having multiple outputs at the same time
-    if name in ['Deep Lift', 'Lime', 'Integrated Gradients', 'Gradient Shap']:
-        attr_list = []
-        for target in range(args.pred_len):
-            score = explainer.attribute(
-                inputs=inputs, baselines=baselines, target=target,
-                additional_forward_args=additional_forward_args
-            )
-            attr_list.append(score)
-
-        attr = torch.stack(attr_list)
-        # pred_len x batch x seq_len x features -> batch x pred_len x seq_len x features
-        attr = attr.permute(1, 0, 2, 3)
-
-    elif name == 'Feature Ablation':
-        attr = explainer.attribute(
-            inputs=inputs, baselines=baselines,
-            additional_forward_args=additional_forward_args
-        )
-    elif name == 'Occlusion':
-        attr = explainer.attribute(
-            inputs=inputs,
-            baselines=baselines,
-            sliding_window_shapes = (1,1),
-            additional_forward_args=additional_forward_args
-        )
-    elif name == 'Augmented Occlusion':
-        attr = explainer.attribute(
-            inputs=inputs,
-            sliding_window_shapes = (1,1),
-            additional_forward_args=additional_forward_args
-        )
-    elif name == 'Morris Sensitivity':
-        attr = explainer.attribute(
-            inputs=inputs,
-            additional_forward_args=additional_forward_args
-        )
-    else:
-        print(f'{name} not supported.')
-        raise NotImplementedError
-
-    attr = attr.reshape(
-        # batch x pred_len x seq_len x features
-        (inputs.shape[0], args.pred_len, args.seq_len, attr.shape[-1])
-    )
-
-    return attr
-
 def get_total_data(dataloader, device, add_x_mark=False):        
     if add_x_mark:
         return (