Handwritten Number Prediction
Number Prediction Web Apps with tflearn, flask, d3.js (https://riki-num-pred.herokuapp.com)
The CNN (3conv + 2fc) is trained on the MNIST dataset in tensorflow. The recognition error on the test data set is 1.02% (9898/10000 classified correctly)
Requirement
numpy, tensorflow, tflearn, flask
Usage
$ python app.py
deploy to heroku
$ pyenv virtualenv 3.5.2 venv_DL_num-pred
$ pyenv local venv_DL_num-pred
$ echo .python-version >> .gitignore
$ cat .gitignore # for check
$ pip install gunicorn
# install tensorflow @ Mac OS X
$ export TF_BINARY_URL=https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.10.0-py3-none-any.whl
$ pip install $TF_BINARY_URL
$ pip install numpy flask tflearn
$ pip freeze > requirements.txt
requirements.txt の tensorflowの箇所を書き換え
- tensorflow==0.10.0
+ https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow-0.10.0-cp35-cp35m-linux_x86_64.whl
$ echo python-3.5.2 > runtime.txt
$ echo web: gunicorn app:app --log-file=- > Procfile
$ heroku create riki-num-pred
$ heroku buildpacks:set heroku/python
$ git add -A
$ git commit -m "deploy heroku"
$ git push heroku master
$ heroku ps:scale web=1
$ heroku open
NN Archtecture
