Improving the ecotoxicological hazard assessment of chemicals by pairwise learning employed on a large set of ecotoxicity test data
One needs to first download and extract the ADORE data set in the current folder.
Then download and compile the LibFM library. The Python codes in the Jupyter notebook simply call the LibFM binary for model fitting and prediction. First the ADORE dataset is transformed to the LibSVM format, then the command line arguments are constructed, and finally LibFM is called on the relevant data sets with the given parameters.
You can also call LibFM directly. The following example is given in the LibFM manual:
./libFM -task r -train ml1m-train.libfm -test ml1m-test.libfm -dim ’1,1,8’ -iter 1000 -method mcmc -init_stdev 0.1
You can then run all of the code blocks in experiments.ipynb to reproduce the results in the paper. Set the PATH_LIBFM variable to the directory that contains the LibFM binaries and the PATH_CACHE variable to a directory that can be used to create temporary files for LibFM.
The predictions for all possible (species, chemical, duration) triplets are saved in the file predictions.csv. This file is also uploaded to Zenodo and can be downloaded from there.
Markus Viljanen ([email protected])