That would probably be good.
I think we should really think about how we want to approach a pipeline with imputation and one-hot encoding. It's a very common setting but we don't really have a good solution for it, also see #6967 and #8539.

We probably want imputation to be different on continuous and categorical data.
I think imputation first and one-hot-encoding later is the way to go, which would make your issues somewhat of a non-issue [which we could still fix though].

Would that also be an acceptable solution for you? Or is there a particular reason you want to do one-hot-coding first and then imputation?

The reason why I did one-hot encoding first is exactly because of #6967, and the fact that I had NaNs in both categorical and numeric features. It did not occur to me, that I can first remove NaNs in the categorical features, then apply the imputer, and finally the one-hot encoder. It's a bit convoluted, but it works.

I might try to submit a pull-request myself in the next few days, if you think this is easy enough for a beginner in scikit.

Actually, var.dropna() on the pd data frame worked fine for me to solve the problem....

I do not understand, why has this been closed? I am using this in late summer of 2019 and this error still seems to be there. Also please be careful: you should NOT simply dropna, that does not solve the problem but it makes you blind w.r.t. the symptoms. From a Data Science perspective it may be really harmful to remove NAs (because the absence of information might be an important information for the model!).

Steps/Code to Reproduce

import pandas as pd
from sklearn.preprocessing import OneHotEncoder
one_hot_encoder = OneHotEncoder(handle_unknown='ignore')
df = pd.DataFrame({'boolean_as_string':["True", "False", None]})
one_hot_encoder.fit(df)

import math
df = pd.DataFrame({'numeric':[0.1, 0.2, None, math.nan, np.nan]})
one_hot_encoder.fit(df)

Expected Results
in the first case:
pd.DataFrame({'boolean_as_string_TRUE':[1, 0, 0], 'boolean_as_string_FALSE':[0, 1, 0]})
(boolean column is being one hot encoded and NA values will result in all 0)

and in the second case:
pd.DataFrame({'numeric':[0.1, 0.2, None, math.nan, np.nan]})
(one hot encoder should not touch the column as it is numeric already)

Versions
Python 3.7.3 (default, Mar 27 2019, 16:54:48)
scikit-learn: 0.21.2-py37h27c97d8_0

This is closed because ColumnTransformer can be used to fully disregard columns not being transformed by the OneHotEncoder.