get_sparse_C function encountered index error

[nrober1122]

Describe the bug
compute_bounds encounters an error when calling the get_sparse_C function through check_prior_bounds for certain nn.Module configurations.

To Reproduce

1. Minimum example is provided in this colab
2. The pytorch state_dict file is attached, which can be unzipped then uploaded into the colab file structure.
   single_pendulum_small_controller.torch.zip
3. Error message:

IndexError                                Traceback (most recent call last)
<ipython-input-5-f8356946ec50> in <module>
     19   my_input = BoundedTensor(nominal_input, ptb)
     20 
---> 21   lb, ub = model.compute_bounds(x=(my_input,), method="backward")
     22   t_end = time.time()
     23 

6 frames
/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in compute_bounds(self, x, aux, C, method, IBP, forward, bound_lower, bound_upper, reuse_ibp, reuse_alpha, return_A, needed_A_dict, final_node_name, average_A, intermediate_layer_bounds, reference_bounds, intermediate_constr, alpha_idx, aux_reference_bounds, need_A_only, cutter, decision_thresh, update_mask)
   1337         self.final_node_name = final.name
   1338 
-> 1339         self.check_prior_bounds(final)
   1340 
   1341         if method == 'backward':

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in check_prior_bounds(self, node)
    881             return
    882         for n in node.inputs:
--> 883             self.check_prior_bounds(n)
    884         for i in getattr(node, 'requires_input_bounds', []):
    885             self.compute_intermediate_bounds(

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in check_prior_bounds(self, node)
    881             return
    882         for n in node.inputs:
--> 883             self.check_prior_bounds(n)
    884         for i in getattr(node, 'requires_input_bounds', []):
    885             self.compute_intermediate_bounds(

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in check_prior_bounds(self, node)
    881             return
    882         for n in node.inputs:
--> 883             self.check_prior_bounds(n)
    884         for i in getattr(node, 'requires_input_bounds', []):
    885             self.compute_intermediate_bounds(

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in check_prior_bounds(self, node)
    883             self.check_prior_bounds(n)
    884         for i in getattr(node, 'requires_input_bounds', []):
--> 885             self.compute_intermediate_bounds(
    886                 node.inputs[i], prior_checked=True)
    887         node.prior_checked = True

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/bound_general.py in compute_intermediate_bounds(self, node, prior_checked)
    968                         ref_intermediate_lb, ref_intermediate_ub = aux_bounds
    969 
--> 970                 sparse_C = self.get_sparse_C(
    971                     node, sparse_intermediate_bounds,
    972                     ref_intermediate_lb, ref_intermediate_ub)

/usr/local/lib/python3.9/dist-packages/auto_LiRPA-0.3.1-py3.9.egg/auto_LiRPA/backward_bound.py in get_sparse_C(self, node, sparse_intermediate_bounds, ref_intermediate_lb, ref_intermediate_ub)
    374                 del ref_intermediate_lb, ref_intermediate_ub
    375         if not reduced_dim:
--> 376             num_channel = node.output_shape[-3]
    377             # Identity patch size: (ouc_c, 1, 1, 1, out_c, 1, 1).
    378             patches = (

IndexError: tuple index out of range

5. To replicate this bug, please open the colab (optionally, save as a copy so you can edit as the above link just provides viewing permission). Unzip & upload the provided .torch model file. Run the first cell to clone & install auto-lirpa. Restart the runtime and auto-lirpa will be fully installed (no need to re-run the first cell again). Run the subsequent cells to define the model class, load the state_dict, and run the test case.

System configuration:

• OS: Ubuntu 20.04 (provided by colab)
• Python version: Python 3.9.16
• Pytorch Version: 1.13.1
• Hardware: google colab
• Have you tried to reproduce the problem in a cleanly created conda/virtualenv environment using official installation instructions and the latest code on the main branch?: implemented in a new colab file

Additional context
This is an attempt to work around issue #43 (also shares context with #42), so we are using the same model structure with multiple controller/dynamics blocks appended together. Notice this error doesn't occur for the first iteration (one controller/dynamics pair), but does for subsequent iterations. It seems to be an issue with the way BoundAdd is handled in get_sparse_C