# License: BSD-3-Clause
import copy
import numpy as np
import pandas as pd
from sklearn.preprocessing import FunctionTransformer
from ..utils import check_Xs_y
from .. import deepmodule_installed, Tensor
if deepmodule_installed:
import torch
[docs]
class RemoveMods(FunctionTransformer):
r"""
A transformer that generates block-wise missingness patterns in complete multi-modal datasets. Apply
`FunctionTransformer <https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.FunctionTransformer.html>`_ (from `Scikit-learn`) with `remove_modalities` as a function.
Parameters
----------
observed_mod_indicator: array-like of shape (n_samples, n_mods)
Boolean array-like indicating observed modalities for each sample.
See Also
--------
:class:`~imml.ampute.Amputer`
:class:`~imml.ampute.remove_mods`
Example
--------
>>> import numpy as np
>>> import pandas as pd
>>> from imml.ampute import RemoveMods
>>> Xs = [pd.DataFrame(np.random.default_rng(42).random((20, 10))) for i in range(3)]
>>> observed_mod_indicator = np.random.default_rng(42).choice(2, size=(len(Xs[0]), len(Xs)))
>>> transformer = RemoveMods(observed_mod_indicator = observed_mod_indicator)
>>> transformer.fit_transform(Xs)
"""
def __init__(self, observed_mod_indicator):
self.observed_mod_indicator = observed_mod_indicator
super().__init__(remove_mods, kw_args={"observed_mod_indicator": self.observed_mod_indicator})
[docs]
def remove_mods(Xs: list, observed_mod_indicator):
r"""
A function that generates block-wise missingness patterns in complete multi-modal datasets.
Parameters
----------
Xs : list of array-likes objects
- Xs length: n_mods
- Xs[i] shape: (n_samples, n_features_i)
A list of different mods.
Returns
-------
transformed_X : list of array-likes objects (n_samples, n_features_i)
The transformed multi-modal dataset.
See Also
--------
:class:`~imml.ampute.Amputer`
:class:`~imml.ampute.RemoveMods`
Example
--------
>>> import numpy as np
>>> import pandas as pd
>>> from imml.ampute import remove_mods
>>> Xs = [pd.DataFrame(np.random.default_rng(42).random((20, 10))) for i in range(3)]
>>> observed_mod_indicator = np.random.default_rng(42).choice(2, size=(len(Xs[0]), len(Xs)))
>>> remove_mods(Xs=Xs, observed_mod_indicator = observed_mod_indicator)
"""
Xs = check_Xs_y(Xs=Xs, ensure_all_finite="allow-nan")
transformed_Xs = []
if isinstance(observed_mod_indicator, pd.DataFrame):
observed_mod_indicator = observed_mod_indicator.values
for X_idx, X in enumerate(Xs):
idxs_to_remove = observed_mod_indicator[:, X_idx] == False
transformed_X = copy.deepcopy(X)
if isinstance(transformed_X, pd.DataFrame):
transformed_X = transformed_X.values
if isinstance(transformed_X, np.ndarray):
if np.issubdtype(transformed_X.dtype, np.integer):
transformed_X = transformed_X.astype(float)
transformed_X = np.copy(transformed_X)
transformed_X[idxs_to_remove, :] = np.nan
elif isinstance(transformed_X, Tensor):
if torch.is_floating_point(transformed_X):
transformed_X = transformed_X.float()
transformed_X[idxs_to_remove, :] = torch.nan
transformed_Xs.append(transformed_X)
if isinstance(Xs[0], pd.DataFrame):
transformed_Xs = [pd.DataFrame(transformed_X, columns=X.columns, index=X.index) for X, transformed_X in
zip(Xs, transformed_Xs)]
return transformed_Xs