Skip to content

Latest commit

 

History

History
17 lines (14 loc) · 2.9 KB

README.MD

File metadata and controls

17 lines (14 loc) · 2.9 KB
Raw

Example with Sparse data based on Quora Question Pairs

This example uses StackNet along with some preparation with python to score 0.3250 (logloss or better after merging with another kernel submission ) for the Quora Question Pairs challenge on kaggle

To run follow the next steps:

  1. Download the train.csv and test.csv files from here
  2. Download the StackNet.jar file from the Git
  3. Ensure you have Python installed
  4. Make certain you have Java higher than 1.6 installed and that Java is in your PATH. Have a look at this if you encounter trouble.
  5. Run the main_querry_v1.py script (also in the example section of the git). This is based on this script . It creates the same basic data as in the script plus a simple tf-idf on question1 and question2 merged as the main purpose was to test high-cardinality sparse data. Then it prints the train.sparse and test.sparse in Libsvm format in the same folder where the file is executed.
  6. Run in the command line. You may need to look at the parameters’ section on GitHub to understand more about the available models and their hyper paramaters. CD to the folder where all the files are and press: java -Xmx4048m -jar StackNet.jar train task=classification train_file=train.sparse test_file=test.sparse params=paramsv1.txt sparse=True pred_file=querry_pred.csv test_target=false verbose=true Threads=3 stackdata=false folds=5 seed=1 metric=logloss
  7. The whole process with 3 threads may take around 4-5 hours. Consider increasing the threads to 11 if you have that many, but you might need to increase the allocated memory from 4 GB (Xmx4048m) to 6 GB (Xmx6072m) or more .
  8. The structure as appears in the paramsv1.txt file contains 11 level_1 models and 1 level_2 model. You can learn more about the available models and parameters here
  9. You can get 0.31019 via averaging the output prediction file of StackNet, in this scenario querry_pred.csv with the ** xgb_seed12357_n315.csv** from this kernel . You may use the create_sub_from_preds.py to do that. If you want only the StackNet prediction, then set the second_file to “”
  10. There is an ongoing discussion with many tips and troubleshooting in the twosigma competition too.