Fix np.float and np.float32 to float. #1257
Conversation
|
np.float is indeed depreacated, but I'm not sure np.float32 is also deprecated. I am pretty sure float is np.float64. So you should keep np.float32 expressions since GPU usually works with 32 bits floatting point numbers. |
| @@ -470,7 +470,7 @@ def _set_parameters( | |||
| if isinstance(y_center, torch.Tensor): | |||
| eps = torch.finfo(y_center.dtype).eps | |||
| else: | |||
| eps = np.finfo(np.float).eps | |||
There was a problem hiding this comment.
32 or 64 bits? Why not y_center.dtype?
| @@ -785,7 +785,7 @@ def fit(self, y: pd.Series, X: pd.DataFrame): | |||
| self | |||
| """ | |||
| y = self.preprocess(y) | |||
| eps = np.finfo(np.float).eps | |||
There was a problem hiding this comment.
32 or 64 bits? We cannot use y.dtype because it is possible to have different type for y and X.
You are absolutely right, I have assumed the inverse, that float was np.float32. Therefore, I should not have changed np.float32 to float, but it is not clear to me when we are doing |
|
After some thought, I decided to change np.float to np.float32 following the assumption that we usually work with GPUs, so eps from np.float64 can be 0 if we use np.float32, but eps from np.float32 will never be 0. |
|
Is this update still in progress? |
Description
This PR fixes #1236 and #1207.
Following the numpy error message,
np.floatwas a deprecated alias for the builtinfloat. The aliases was originally deprecated in NumPy 1.20.Checklist
pre-commit install.To run hooks independent of commit, execute
pre-commit run --all-filesMake sure to have fun coding!