Learning to rank resources for selective search
- SVM Rank
- Python sklean library
- Download /bos/tmp11/zhuyund/LeToRankResource/data.zip. Put under the same dir with source code and unzip.
- Modify SVMRank path in the source code. Modify training basedir (./data/aol-train/ or ./data/mqt-train/) in pairwise-train-AOL-cleaned.py.
- python ./pairwise-train-AOL-cleaned.py
- python ./pairwise-test-clean.py
- pairwise-test-clean.py will print out number of relevant documents retrieved by each method (when selecting 4 shards).
- Shard list will be written into ./data/cwb-test/.
- Use fedsearch/run_job_writer.rb and fedsearch/make_runs.sh
- shardmap: /bos/usr0/zhuyund/fedsearch/data/cent1-qw160-split-new-ext/
- central: /bos/usr0/zhuyund/fedsearch/data/cwB-split-sdm-nospam-central/
- qrels: /bos/usr0/zhuyund/fedsearch/data/cwb-[2009-2012].qrels /bos/usr0/zhuyund/fedsearch/data/cwb.qrels
- upload AOL and MQT training data.