https://github.com/titu1994/kaggle
Kaggle competition library. Uses Python 3.4.1 with almost all known python libraries for Machine Learning
https://github.com/titu1994/kaggle
Last synced: 3 months ago
JSON representation
Kaggle competition library. Uses Python 3.4.1 with almost all known python libraries for Machine Learning
- Host: GitHub
- URL: https://github.com/titu1994/kaggle
- Owner: titu1994
- Created: 2015-12-27T13:44:44.000Z (almost 10 years ago)
- Default Branch: master
- Last Pushed: 2016-11-18T08:43:42.000Z (almost 9 years ago)
- Last Synced: 2025-04-11T08:32:44.612Z (6 months ago)
- Language: Python
- Size: 83.8 MB
- Stars: 7
- Watchers: 4
- Forks: 5
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Kaggle
A few python scripts that perform well on Kaggle competions.
# Competition
## MNIST
Contains various scripts using either XGBoost, PCA + SVM or Convolutional Neural Networks.
- CNNs tend to perform the best, with the VGG-like net performing the best but requiring the largest amount of time.
- SqueezeNet is fast to train but does not perform as well as VGG
- The current best script I posses is the DCCNN MNIST architecture, but since it has been trained on the entire data set it will
over fit in this Kaggle dataset and give 100% accuracy. (Original accuracy = 99.77 %)
## Titanic
Starter dataset to teach about the basics of Machine Learning principles such as data cleaning / preprocessing and feature construction.
- Best performing model is XGBoost
## Bike Sharing Demand
A time series data set which is very useful to understand how to manipulate and train time series datasets.
- Best performing model is an ensemble of two XGBoost which learn the two different time factored outputs and then merge them into a single output.
## BNP
Bank BNP Paribas contest.
- Tried various combinations of stacking and neural nets. Best was a combination of neural nets and XGBoost stacked with Logistic Regression as final layer.