Ignore NaN values?

[WHBSpeiser]

Hello,

I am trying to run EOF over a large collection of images where, on some dates, values are missing (NaN) at a given pixel position due to cloud cover. When I run df_eof on this collection, I get an error as "input contains NaN, infinity or a value too large for dtype('float64')" as some pixels are omitted on certain dates because of cloud masking, but df_eof will work if I change the NaN value to a numeric value. Is there a way to go around this, or can I only use datasets in which data is never missing? Thank you!