mercadolibre
2nd place solution for the MeLi Challenge 2019
This solution uses three models SGD (scikit-learn), multinomialNB (scikit-learn) and GRU (Keras) trained on both char level and word level, and in two different (but quite similar) datasets.
Instructions to generate submission
Required packages and softwares:
Python3 with the libraries Numpy, Pandas, Scikit-Learn, MatplotLib, Keras, NLTK. (No other library was used. No pre-trained model was used.)
Follow the steps:
-
Unpack train, test and sample submission in the root folder with the respective names: train.csv, test.csv, sample_submission.csv
-
Run all notebooks in MeLi_BaseGen/
-
Run all notebooks in MeLi_scripts2/
-
Run all notebooks in MeLi_scripts3/
-
Run the notebook MeLi_Ensembles/MeLi_Ensemble_06.ipynb
-
Run the notebook MeLi_Ensembles/MeLi_Ensemble_11.ipynb
-
Run the notebook MeLi_Ensembles/MeLi_FinalEnsemble.ipynb
The final submission will be in the root folder with the name 'submission_MegaEnsemble02.csv' :)
PS: Some of these notebooks might require more than 64 GB of memory to run. Each notebook takes about 1 to 2 hours to run copletely in a 4 cores computer.