Open GoogleCodeExporter opened 8 years ago
this is really a plague, especially if you're trying to train 'real' models
where transitions probabilities are potentially very low (in which case, just
hacking the Opdf write methods to add more decimals doesn't make any sense...)
Pick the attached class, place it in the 'io' directory and replace all
references to StreamTokenizer for CustomStreamTokenizer. This attached class
is a simple modification of the original Sun StreamTokenizer implementation: it
now supports the scientific notation for the 'number' class.
Original comment by jean.phi...@gmail.com
on 12 Jul 2011 at 8:28
Attachments:
A fix that worked for me is adding the following into line 92 of
ObservationVectorReader.java
st.whitespaceChars(',',',');
Original comment by chris.mi...@gmail.com
on 24 Aug 2015 at 11:51
Original issue reported on code.google.com by
vamos.be...@gmail.com
on 15 May 2010 at 6:46