This document is for kaggle Cdiscount competition. All codes are written by Python3.5.
.
├── code # Code files
| ├── model # model package
| | ├── loss.py
| | ├── lr_schedule.py
| | └── xception.py
| ├── utils # utils package
| | ├── callbacks.py
| | ├── iterator.py
| | ├── sysmonitor.py # (optional) use for monitoring CPU & GPU status
| | └── utils.py # default settings
| ├── feature_extractor.py
| ├── fine_tuning.py
| ├── predict_with_snapshot.py
| ├── prediction.py
| ├── preprocessing.py
| ├── split_validation.py
| ├── train_with_branch.py
| └── training.py
├── data # Data files
| ├── input # Original data files
| | ├── category_names.csv
| | ├── sample_submission.csv
| | ├── test.bson
| | └── train.bson
| ├── logs
| | └── ...
| ├── results
| | └── ...
| ├── utils
| | └── ...
| └── weights
| └── ...
├── source_code # Unused files, only for reference
| └── ...
└── ReadMe.md
- Put your input files into /data/input folder.
- Preprocess the dataset using
preprocessing.py
- Split validation set from train.bson using
split_validation.py
- Train the model using
training.py
- Make prediction and submission using
prediction.py
orprediction_with_snapshot.py
LZM
ZJY
MTJ
- Tensorflow 1.3.0
- Keras 2.0.9 (2.0.9 support multi-gpu)
- Pymongo 3.5.1 (use for import bson, don't really import pymongo)
- Pandas
- Numpy
- Matplotlib
- h5py
All codes from internet are listed in source_code folder