machine learning @ python
Python
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data
feature
ref
submit
.gitignore
README.md
circle.yml
conv_softmax.py
convolutional.py
displayer.py
extractor.py
finalizer.py
finalizer2.py
input_data.py
optimizer.py
plotter.py
random_forest.py
showdata.py

README.md

digit-recognition

Circle CI

Commit History

out-of-bag accuracy

Development Setup

Folder Structure

  • data : raw data for training / testing
  • ref : collected sample code from the internet
  • feature : extracted feature set from data
  • submit : submit to contest history

File Usage

  • extractor.py : pre-process, only run 1 time for reading raw data
  • random_forest.py: train, test, create .csv for submission (main function)
  • mnist_softmax.py: train, test, create .csv for submission (main function)
  • convolutional.py: train, test, create .csv for submission (main function)
  • optimizer.py : calculate then plot out-of-bag accuracy for the range of trees (n_estimators)
  • plotter.py : plot history accuracy to plotly, run at CircleCI
  • displayer.py : display train/test file
  • finalizer.py : transform probability to 0 or 1 if certain enough

Feature Extraction

  • deskew_ : remove skew from image (28*28=784)
  • erosion_ : erosion of image (28*28=784)
  • skeleton_ : skeleton of image (28*28=784)
  • xy_sum_ : pixel color density in x & y direction (28+28=56)
  • bbox_ : bounding box of digit, including center coordinate, rectangle width, rectangle length, width/length ratio (2+2+1=5)
  • contour
  • color difference / slope
  • histogram of oriented gradients