Unexpected content length doc_length and duplicate_ngram_chr_fraction #357
Additionally the results indicate This is the source text exact as used: I am at a loss to understand why the duplicate flag does not pass and why doc length says it is just 700+ when it is really about 3000+ characters. |
Replies: 3 comments 2 replies
|
... I can solve the first issue with something like: Apparently it has some troubles with some invisible characters or newlines? However the second issue persists even after above cleanse, which results in a single line string text. |
I believe it is the number of spacy tokens. So yes the docs are wrong. For the other error how does the duplicate_ngram_chr_fraction_5 change when applying the fix? The n_gram used are spacy n_grams and sometimes it does treat e.g. double white spaces as a separate token. I could imagine it could do something similar with special tokens. |
|
yes, I can solve the ngram issue with the "fix" I mention (basically just strip out whatever is not expected in a "normal" text. Fix means the value becomes much lower, as expected. I guess this is solved then, pending DOC update! |
I believe it is the number of spacy tokens. So yes the docs are wrong.
For the other error how does the duplicate_ngram_chr_fraction_5 change when applying the fix? The n_gram used are spacy n_grams and sometimes it does treat e.g. double white spaces as a separate token. I could imagine it could do something similar with special tokens.