BMI local Implementation

Installation

1. git clone https://github.com/HTAustin/CAL.git
2. Intall Sofia-ML package: https://code.google.com/archive/p/sofia-ml/
3. Make the kisssdb indexer. cd CAL && make
4. Change the path for Sofia-ML in doAll_Baseline

SOFIA="/the/path/to/sofia-ml-read-only/src/sofia-ml"

Usage

1. Run CAL Auto TAR: bash doAll_Baseline
2. Configure behaviour through environment variables

MODE            - default tfidf. Valid values: 4gram, tfidf
MAXTHREADS      - number of threads. Default: 4
SOFIA           - path to sofia ml binary. Default: ./sofia-ml/src/sofia-ml
CORP            - corpus to use. Default: oldreut
CACHE           - if set, enable caching of corpus specific pre computations. Default not set.

eg.
$ MODE=4gram MAXTHREADS=16 SOFIA=/home/nghelani/sofia-ml/sofia-ml CORP=aquaint bash doAll_Baseline

3. Important files assumed by the script

Corpus/<CORP>.tgz                       - Corpus
judgement/<CORP>.topic.stemming.txt     - Topics separated by newline (each line is "<topic_id>:<query>")
judgement/qrels.<JUDGECLASS>.list       - Relevance judgements for topics (each line is "<topic> 0 <doc> <score>")

4. The output of BMI are stored in result/ folder.
5. The gain curve can be plotted by analyzing result/baseline/<corp>/<topic>/<topic>.record.list
6. Plot gain curves with gainCurve.py (see python2 gainCurve.py -h)

Speedup Tips

1. Comment out the ./dofast line if you already completed fine the last time
2. If using qrels for assessment, consider quitting the iterations when you have found the desired number of relevant documents (See the sample snippet)

    NUM_REL=$(cat rel.$TOPIC.fil | sort | uniq | wc -l)
    TOT_REL=$(grep "^$TOPIC.*[1-9]$" ../judgement/qrels.$JUDGECLASS.list | cut -d' ' -f3 | sort | uniq | wc -l)
    if [ $NUM_REL -eq $TOT_REL ]; then
        break
    fi

3. Lower the number of iterations. The default number of iterations (=100) might be too high for your purpose.

Contribute

Please feel free to open issues and report bugs.

License