QuickRank can translate learnt tree-based models into efficient C++ source code that can be used to score documents efficiently.
This is achieved with a two step process. During the first step, a previously learnt model is translated into a C++ source. QuickRanks implements three translation strategies:
condop: translates into nested C conditional operators of the kind(x[feature]<=threshold) ? ( ...left...) : (...right...).vpred: uses the strategy described in Asadi et al [1].oblivious: an optimized strategies for oblivious regression trees [2].
The optimized source code is generated as follows:
./bin/quicklearn --model-file model.xml \
--code-file model.cc \
--generator condop
After the source code was generated it is possible to test its efficiency.
First you need to replace the file src/scoring/ranker.cc with the source file model.cc generated previously.
Then you can compile it by invoking make quickscore in your build directory.
Upon termination a new binary is compiled bin/quickscore implementing the original model.
./bin/quickscore -r 10 -d dataset.test
The result shows the time need by the model to score the document in the input dataset averaged over 10 rounds.
_____ _____
/ / /____/
/____\ / \ QuickRank has been developed by hpc.isti.cnr.it
::Quick:Rank:: [email protected]
# Dataset size: 5000 x 136 (instances x features)
# Num queries: 43 | Avg. len: 116
Total scoring time: 0.000139 s.
Avg. Dataset scoring time: 1.39e-05 s.
Avg. Doc. scoring time: 2.78e-09 s.
[1] Asadi N, Lin J, De Vries AP. Runtime optimizations for tree-based machine learning models. IEEE Transactions on Knowledge and Data Engineering. 2014. LINK
[2] Capannini, G., Lucchese, C., Nardini, F. M., Orlando, S., Perego, R., and Tonellotto, N. Quality versus efficiency in document scoring with learning-to-rank models. Information Processing & Management (2016). LINK.