This repo contains sample code for the Job Salary Prediction competition, hosted by Kaggle with Adzuna.
It contains several benchmarks marked as different git tags. This is the Random Forest Benchmark. For another example, see the Mean Benchmark.
Executing this benchmark requires Python 2.7 along with the following packages:
- pandas (version >=10.1)
- sklearn
- numpy
To run the benchmark,
- Download the data
- Modify SETTINGS.json to point to the training and validation data on your system, as well as a place to save the trained model and a place to save the submission
- Train the model by running
python train.py - Make predictions on the validation set by running
python predict.py - Make a submission with the output file
This benchmark took approximately 1.5 hours to execute on a Windows 8 laptop with 8GB of RAM and 4 cores at 2.7GHz.
If you run into issues with execution time or memory usage, you can make the model run faster and use less memory by doing the following:
- Reduce the n_estimators parameter on RandomForestRegressor in train.py to 10
- Reduce the max_features parameter on CountVectorizer in train.py to 20
- Set the nrows parameter to the pd.read_csv function in data_io.py to 10000 (currently not set)