https://github.com/mtchavez/kaggle-digit-recognizer
Kaggle Digit Recognizer Competition
https://github.com/mtchavez/kaggle-digit-recognizer
Last synced: about 2 months ago
JSON representation
Kaggle Digit Recognizer Competition
- Host: GitHub
- URL: https://github.com/mtchavez/kaggle-digit-recognizer
- Owner: mtchavez
- Created: 2014-12-02T06:01:40.000Z (over 10 years ago)
- Default Branch: master
- Last Pushed: 2014-12-02T06:16:23.000Z (over 10 years ago)
- Last Synced: 2025-02-04T15:32:51.984Z (4 months ago)
- Size: 19.1 MB
- Stars: 0
- Watchers: 2
- Forks: 3
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Digit Recognizer
Basic Kaggle competition for recognizing digits using the MNIST
("Modified National Institute of Standards and Technology") dataset.
[Competition details](http://www.kaggle.com/c/digit-recognizer)
## Requirements
* Julia >= 3.0
* DataFrames package
* DecisionTree package
* iJulia
* Datasets in `./data` or on the [competition page](http://www.kaggle.com/c/digit-recognizer/data)
## Running
Start up ipython notebook `ipython notebook --profile julia` and open
the `DigitRecognizer.ipynb` notebook
## Results
Using a random forest of 10 random features, 10 trees, and 0.5 portion of
samples per tree builds a model of:
```julia
Ensemble of Decision Trees
Trees: 20
Avg Leaves: 3426.75
Avg Depth: 24.4
```
Using a 5 fold cross validation the mean accuracy for prediction the correct
digit is `Mean Accuracy: 0.9388095238095238`
This is relatively good using a decision tree on digit recognition versus
a k nearest neighbors approach which might yield better accuracy.
At time of submission this accuracy ranked `#368` on the [leaderboard](http://www.kaggle.com/c/digit-recognizer/leaderboard).