Type Errors with Numpy on CPU

[nsutphen1]

I am attempting to build a model that takes in Array2D for the features and Array1D for the labels.

The feature vectors are numpy arrays stored in a python list, and the labels are one-hot encoded arrays in a numpy array.

The error comes when I attempt to use the parametric attention layer. It says that y must be a ragged list, but there aren't any layers that affect y to my knowledge.

Been working on this for a while, any input would be great.

[svlandeg]

Can you provide a minimal code snippet that exhibits the error, and paste the error trace?

[nsutphen1]

Model:

with Model.define_operators({'>>': chain, '|': concatenate}):
  model = (thinc.api.list2ragged()
          >>thinc.api.ParametricAttention()
          >>reduce_max()
          >>expand_window(window_size=1)
          >>reduce_max()
          >>Softmax(nO=nr_classes))

optimizer = Adam(learn_rate)

Where the error occurs:
model.initialize(dummy_x,dummy_y_2)

Traceback:

ValidationError                           Traceback (most recent call last)

/usr/local/lib/python3.6/dist-packages/thinc/util.py in validate_fwd_input_output(name, func, X, Y)
    420     try:
--> 421         ArgModel.parse_obj(args)
    422     except ValidationError as e:

14 frames

ValidationError: 1 validation error for ArgModel
Y -> 0
  instance of Ragged expected (type=type_error.arbitrary_type; expected_arbitrary_type=Ragged)


During handling of the above exception, another exception occurred:

DataValidationError                       Traceback (most recent call last)

/usr/local/lib/python3.6/dist-packages/thinc/util.py in validate_fwd_input_output(name, func, X, Y)
    421         ArgModel.parse_obj(args)
    422     except ValidationError as e:
--> 423         raise DataValidationError(name, X, Y, e.errors())
    424 
    425 

DataValidationError: 

Data validation error in 'para-attn'
X: <class 'thinc.types.Ragged'> Y: <class 'list'>

Y -> 0   instance of Ragged expected

[rgommers]

This issue is 6 years old now, and doesn't have a self-contained reproducer. So I'll close this as inactionable. If this is still relevant, please comment and add a reproducer that can actually be run to reproduce the error shown above.