PCA with dataframe can't real explained_variance_ratio_ except nan

What happened:

when use dask_ml.decomposition.PCA fit the dask.dataframe， then will get the nan of pca Attributes pca.explained_variance_ratio_/explained_variance_ ; maybe can’t get the dataframe shape lead to the error.

What you expected to happen: [0.99244289 0.00755711]

Minimal Complete Verifiable Example: [nan nan]

# Put your MCVE code here
import dask.dataframe as dd
import dask.array as da
import numpy as np
from dask_ml.decomposition import PCA

X = np.array([[-1, -1], [-2, -1], [-3, -2], [1, 1], [2, 1], [3, 2]])
dX = da.from_array(X, chunks=X.shape)
pca = PCA(n_components=2)
pca.fit(dX)
print(pca.explained_variance_ratio_)
#the result is [0.99244289 0.00755711]


df = dd.from_dask_array(dX, columns=['a','b'])
pca = PCA(n_components=2, svd_solver='full')
pca.fit(df)
print(pca.explained_variance_ratio_)
#the result is [nan nan]

Anything else we need to know?:

Environment: python 3.6 dask 2.18.1 dask-glm 0.2.0 dask-ml 1.5.0 dask-xgboost 0.1.10

• Dask version:
• Python version:
• Operating System:
• Install method (conda, pip, source):

[TomAugspurger]

DataFrame.to_dask_array(lengths=True) will give you a Dask Array with known partition lengths, at the cost of computing them.

[TomAugspurger]

Thanks for following up.