"NaN" Data pit with double type when processing data

Recently, using BP neural network for text categorization, the generated text vector matrix contains "NaN" data, which causes the network to be unable to converge.

Now we finally find out why,

Mainly because in the calculation of division operations, the denominator is 0, resulting in an infinite number of results, can not be displayed with Nan instead. Java code to read the data to add a judge is good if (Double.isnan (datas[i)) | | Double.isinfinite (Datas[i]) {},//excludes data that has "NaN" and infinity in the data.