TypeError with GPU training

[ShijianXu]

Dear authors,

I am running the following code and get the TypeError:

model = LassoNetRegressor()

model.fit(X=train_df.drop('target', axis=1), 
          y=train_df['target'],
          X_val=val_df.drop('target', axis=1),
          y_val=val_df['target']
          )

model.score(test_df.drop('target', axis=1), test_df['target'])

The error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[12], line 9
      1 model = LassoNetRegressor()
      3 model.fit(X=train_df.drop('target', axis=1), 
      4           y=train_df['target'],
      5           X_val=val_df.drop('target', axis=1),
      6           y_val=val_df['target']
      7           )
----> 9 model.score(test_df.drop('target', axis=1), test_df['target'])
     11 # print("Best model scored", model.score(test_df.drop('target', axis=1), test_df['target']))
     12 # print("Lambda =", model.best_lambda_)

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/base.py:849, in RegressorMixin.score(self, X, y, sample_weight)
    846 from .metrics import r2_score
    848 y_pred = self.predict(X)
--> 849 return r2_score(y, y_pred, sample_weight=sample_weight)

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/utils/_param_validation.py:213, in validate_params.<locals>.decorator.<locals>.wrapper(*args, **kwargs)
    207 try:
    208     with config_context(
    209         skip_parameter_validation=(
    210             prefer_skip_nested_validation or global_skip_validation
    211         )
    212     ):
--> 213         return func(*args, **kwargs)
    214 except InvalidParameterError as e:
    215     # When the function is just a wrapper around an estimator, we allow
    216     # the function to delegate validation to the estimator, but we replace
    217     # the name of the estimator by the name of the function in the error
    218     # message to avoid confusion.
    219     msg = re.sub(
    220         r"parameter of \w+ must be",
    221         f"parameter of {func.__qualname__} must be",
    222         str(e),
    223     )

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/metrics/_regression.py:1180, in r2_score(y_true, y_pred, sample_weight, multioutput, force_finite)
   1039 @validate_params(
   1040     {
   1041         "y_true": ["array-like"],
   (...)
   1059     force_finite=True,
   1060 ):
   1061     """:math:`R^2` (coefficient of determination) regression score function.
   1062 
   1063     Best possible score is 1.0 and it can be negative (because the
   (...)
   1178     -inf
   1179     """
-> 1180     y_type, y_true, y_pred, multioutput = _check_reg_targets(
   1181         y_true, y_pred, multioutput
   1182     )
   1183     check_consistent_length(y_true, y_pred, sample_weight)
   1185     if _num_samples(y_pred) < 2:

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/metrics/_regression.py:104, in _check_reg_targets(y_true, y_pred, multioutput, dtype)
    102 check_consistent_length(y_true, y_pred)
    103 y_true = check_array(y_true, ensure_2d=False, dtype=dtype)
--> 104 y_pred = check_array(y_pred, ensure_2d=False, dtype=dtype)
    106 if y_true.ndim == 1:
    107     y_true = y_true.reshape((-1, 1))

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/utils/validation.py:997, in check_array(array, accept_sparse, accept_large_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, ensure_min_samples, ensure_min_features, estimator, input_name)
    995         array = xp.astype(array, dtype, copy=False)
    996     else:
--> 997         array = _asarray_with_order(array, order=order, dtype=dtype, xp=xp)
    998 except ComplexWarning as complex_warning:
    999     raise ValueError(
   1000         "Complex data not supported\n{}\n".format(array)
   1001     ) from complex_warning

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/sklearn/utils/_array_api.py:521, in _asarray_with_order(array, dtype, order, copy, xp)
    519     array = numpy.array(array, order=order, dtype=dtype)
    520 else:
--> 521     array = numpy.asarray(array, order=order, dtype=dtype)
    523 # At this point array is a NumPy ndarray. We convert it to an array
    524 # container that is consistent with the input's namespace.
    525 return xp.asarray(array)

File ~/miniconda3/envs/pytorch/lib/python3.9/site-packages/torch/_tensor.py:1062, in Tensor.__array__(self, dtype)
   1060     return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
   1061 if dtype is None:
-> 1062     return self.numpy()
   1063 else:
   1064     return self.numpy().astype(dtype, copy=False)

TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

Since I am using the interface of LassoNet, I don't think I have much flexibility to modify the code.
Do you have any idea what might cause this error and how should I fix it?

Thank you very much!

[louisabraham]

We could handle this in the library but in the meantime, you can probably call predict yourself, then move the result to CPU before calling the scoring function on the output.

[ShijianXu]

Thank you very much for such a quick reply! This suggestion temporarily solves my problem.