1108622 - [back-end] Implement Folding Tests for URL classification on Moreover corpus

Status: RESOLVED FIXED

(Content Services Graveyard :: Classification Engine, defect)

Description

[maxim zhilyaev]

Need to test classification accuracy/recall on Moreover corpus by folding corpus into older and recent chunks. Whereby, the training occurs on older chunk and is applied to the recent chunk.

If user history is incorporated into URL classification via model fitting (bug# 1104335) the testing should be extended to recency folding as well. However, a user model needs to be developed to synthesize history from sites popularity.

Comment 1

[maxim zhilyaev]

We should have a general way to build rule set from a subset of corpus and apply it to yet another subset of corpus to allow for folding testing.  However, there's currently no algorithmic way to compute rule set - all manual updates to Matthew payload.  The bug# 1104364 is filed for automation of rule generation.

Comment 2

[maxim zhilyaev]

This generic way was implemented by allowing DFR generation and Rule generation be done on specific chunk of the corpus.  Both scripts allow -f date and -t date arguments to enable selection by date range.

Comment 3

[maxim zhilyaev]

./generateCorpusStats.js -h
USAGE: generateCorpusStats.js [OPTIONS]
Generates Corpus URL and Title stats

  -h, --help          display this help
  -v, --verbous       display debug info
  -d, --dbHost=ARG    db hosts: default=localhost
  -p, --dbPort=ARG    db port: default=27017
  -f, --fromDate=ARG  starting from date in yyyy/mm/dd format (like 2014/10/01): default none
  -t, --toDate=ARG    ending from date in yyyy/mm/dd format (like 2014/10/01): default none
  -l, --limit=ARG     docs limit: default none


./generateDFRStats.js -h
USAGE: generateDFRStats.js [OPTIONS] [DFR FILES]
Generates DFR matching stats

  -h, --help          display this help
  -v, --verbous       display debug info
  -d, --dbHost=ARG    db hosts: default=localhost
  -p, --dbPort=ARG    db port: default=27017
  -f, --fromDate=ARG  starting from date in yyyy/mm/dd format (like 2014/10/01): default none
  -t, --toDate=ARG    ending from date in yyyy/mm/dd format (like 2014/10/01): default none
  -l, --limit=ARG     docs limit: default none

Comment 4

[[github robot]]

Commit pushed to master at https://github.com/mzhilyaev/pfeed

https://github.com/mzhilyaev/pfeed/commit/5318f4805ae2d2894f0a238fb54d336b29202dd3
Closes Bug 1108622 - Fix missing from/to search params