[BUG] IncrementalPCA: transform() call fails when using whitening

[lowener]

Describe the bug
Trying to use transform() on IPCA when using whitening fails.

Stacktrace:

Traceback (most recent call last):
  File "a.py", line 11, in <module>
    cu_t = cu_ipca.transform(X)
  File "/opt/conda/envs/rapids/lib/python3.8/site-packages/cuml/internals/api_decorators.py", line 587, in inner_get
    ret_val = func(*args, **kwargs)
  File "/opt/conda/envs/rapids/lib/python3.8/site-packages/cuml/experimental/decomposition/incremental_pca.py", line 408, in transform
    return super().transform(X)
  File "/opt/conda/envs/rapids/lib/python3.8/site-packages/cuml/internals/api_decorators.py", line 587, in inner_get
    ret_val = func(*args, **kwargs)
  File "cuml/decomposition/pca.pyx", line 664, in cuml.decomposition.pca.PCA.transform
  File "/opt/conda/envs/rapids/lib/python3.8/site-packages/cuml/internals/api_decorators.py", line 596, in inner
    ret_val = func(*args, **kwargs)
  File "cuml/decomposition/pca.pyx", line 626, in cuml.decomposition.pca.PCA._sparse_transform
  File "cuml/decomposition/pca.pyx", line 629, in cuml.decomposition.pca.PCA._sparse_transform
  File "cuml/common/base.pyx", line 262, in cuml.common.base.Base.__getattr__
AttributeError

Steps/Code to reproduce bug

import cupy as cp
import cupyx
from cuml.datasets import make_blobs
from cuml.experimental.decomposition import IncrementalPCA as cuIPCA

X, _ = make_blobs(n_samples=500, n_features=15, random_state=10)
cu_ipca = cuIPCA(n_components=2, whiten=True, batch_size=100)
cu_ipca.fit(X)
cu_t = cu_ipca.transform(X)

This happens when whiten=True.

Expected behavior
The call to transform() is successful with cuml PCA and sklearn IncrementalPCA.

Environment details (please complete the following information):

• Docker image used: rapidsai/rapidsai-dev-nightly:0.17-cuda11.0-devel-ubuntu18.04-py3.8

[JohnZed]

@divyegala can you take a look and/or share some pointers with @lowener so he could dig further?

[divyegala]

@lowener looks like it's occurring because attribute n_rows

cuml/python/cuml/decomposition/pca.pyx

Line 629 in f1abcb8

self.components_ *= cp.sqrt(self.n_rows - 1)

is not being set for IncrementalPCA. It will likely be fixed if you set that attribute during IncrementalPCA.fit, if you want to try?

[lowener]

Solved with #3190