threshold parameter

[Gracelek]

After the installation, I am playing with QuickUMLS.

from quickumls import * 
matcher_thres_default = QuickUMLS('/users/gracelee/documents/quickumls_data')
text = 'the authors studied the diagnostic and prognostic factors by using stepwise logistic regression analysis.'
results = matcher_thres_default.match(text, best_match=True, ignore_syntax=False)
for res in results : print(res)

The code above returns three extracted terms. Their similarities are 1.0, 1.0, and 0.6 for each.
Isn't the threshold parameter (default 0.7) handling the similarity values ?

[{'start': 85, 'end': 104, 'ngram': 'regression analysis', 'term': 'regression analysis', 'cui': 'C0034980', 'similarity': 1.0, 'semtypes': {'T170'}, 'preferred': 1}]
[{'start': 39, 'end': 49, 'ngram': 'prognostic', 'term': 'prognostic', 'cui': 'C0220901', 'similarity': 1.0, 'semtypes': {'T170'}, 'preferred': 1}]
[{'start': 24, 'end': 34, 'ngram': 'diagnostic', 'term': 'Diagnostic', 'cui': 'C0011900', 'similarity': 0.6, 'semtypes': {'T033'}, 'preferred': 1}, {'start': 24, 'end': 34, 'ngram': 'diagnostic', 'term': 'Diagnostic', 'cui': 'C0358514', 'similarity': 0.6, 'semtypes': {'T130'}, 'preferred': 1}, {'start': 24, 'end': 34, 'ngram': 'diagnostic', 'term': 'Diagnostic', 'cui': 'C0430022', 'similarity': 0.6, 'semtypes': {'T060'}, 'preferred': 1}, {'start': 24, 'end': 34, 'ngram': 'diagnostic', 'term': 'Diagnostic', 'cui': 'C1547424', 'similarity': 0.6, 'semtypes': {'T170'}, 'preferred': 1}]

If a parameter threshold=0.7 is specified when instantiation, it returns the same result with the same code. But when threshold=0.8, it gives the first two terms.

[soldni]

Hi,
Could you provide me with a bit more information about your setup? (e.g. which platform/python version/UMLS subset/etc.) Thanks!

-Luca

[Gracelek]

Hi Luca,
Here is the setup info.
Python3.6, UMLS 2016AB, and QuickUMLS 1.2 latest version and I am running them on Mac Sierra.
Thanks
-G

[soldni]

Hey,

Thanks for the info. I found a discrepancy between how similarity was being calculated by QuickUMLS and how it was computed in the underlying string matching library. The issue you were experiencing is now fixed in the latest release (v. 1.2.1).

-Luca