resnet18: AttributeError: 'SVDDecomposedConvLayer' object has no attribute 'min_rank'

[aupuzikov]

I've tried to compress my resnet18 and failed. SVDDecomposedConvLayer has no min_rank field. I'm getting exception.

Anaconda3\envs\musco\lib\site-packages\musco\pytorch\compressor\decompositions\svd_layer.py in __init__(self, layer, layer_name, rank_selection, rank, pretrained, vbmf_weaken_factor, param_reduction_rate)
    157 
    158         if rank_selection == 'vbmf':
--> 159             self.rank = estimate_vbmf_ranks(self.weight, vbmf_weaken_factor, min_rank = self.min_rank)
    160         elif rank_selection == 'manual':
    161             self.rank = rank

AttributeError: 'SVDDecomposedConvLayer' object has no attribute 'min_rank'

because of 6e4c947

[aupuzikov]

hi everybody! can anyone fix it? doesn't seem like a big issue more like a bug.

[aupuzikov]

https://github.com/musco-ai/musco-pytorch/blob/master/musco/pytorch/compressor/decompositions/svd_layer.py#L168
looks like out_features/in_features should be considered too....

[aswanthkrishna]

any fix?

[juliagusak]

Hi @aupuzikov and @aswanthkrishna, thank you for noticing this bug! Now it is fixed

[aswanthkrishna]

thanks a lot :)

[styler00dollar]

It seems not to be fixed if it gets installed with pip. I still have:

~/anaconda3/lib/python3.8/site-packages/musco/pytorch/compressor/decompositions/svd_layer.py in __init__(self, layer, layer_name, rank_selection, rank, pretrained, vbmf_weaken_factor, param_reduction_rate)
    157 
    158         if rank_selection == 'vbmf':
--> 159             self.rank = estimate_vbmf_ranks(self.weight, vbmf_weaken_factor, min_rank = self.min_rank)
    160         elif rank_selection == 'manual':
    161             self.rank = rank

AttributeError: 'SVDDecomposedConvLayer' object has no attribute 'min_rank'

I used musco-pytorch 1.0.6. Solution:

pip uninstall musco-pytorch -y
git clone https://github.com/musco-ai/musco-pytorch
cd musco-pytorch
python setup.py install

This results in:

~/anaconda3/lib/python3.8/site-packages/numpy/linalg/linalg.py in _raise_linalgerror_svd_nonconvergence(err, flag)
    104 
    105 def _raise_linalgerror_svd_nonconvergence(err, flag):
--> 106     raise LinAlgError("SVD did not converge")
    107 
    108 def _raise_linalgerror_lstsq(err, flag):

LinAlgError: SVD did not converge

or

~/anaconda3/lib/python3.8/site-packages/numpy/lib/function_base.py in asarray_chkfinite(a, dtype, order)
    495     a = asarray(a, dtype=dtype, order=order)
    496     if a.dtype.char in typecodes['AllFloat'] and not np.isfinite(a).all():
--> 497         raise ValueError(
    498             "array must not contain infs or NaNs")
    499     return a

ValueError: array must not contain infs or NaNs

Which would be a new issue though, but I thought I at least mention it. The output seems random. Used code from Github readme.