CoffeeBrain

You can turn this

into this

{
  "left_pot": "empty",
  "right_pot": "little"
}

And when you hook it up with https://github.com/ktkiiski/coffee-pot you get this

How to install

pip install --upgrade pip && pip install -r requirements.txt

How to fetch data

Get your labeled dataset in CSV format. CSV format is: imageurl,left_label,right_label

When you have yours CSV then run python dataloader <dataset.csv> and enjoy. The dataset is split into testing and training data onload.

How to generate new models

• First fetch data. See above.
• python generate_models.py generates two classifiers. One for each side.
• Classifiers are serialized to classifiers/

How to run the predictor web server

• First fetch data. See above.
• Then generate models. See above.
• Then run python web.py and your server should be running.
• Predictor is listening at endpoint /predict

How to get a prediction Request

curl -X POST -H "Content-Type: application/json"  -d '{"image_url":"https://s3.eu-central-1.amazonaws.com:443/coffee-pot/media/snapshots/2016-10-14/2016-10-14_07.49.09.899294.jpg"}' http://localhost:5000/predict/

Response

{
  "left_pot": "empty",
  "right_pot": "little"
}

You can request prediction from with

How to develop

• First fetch data. See above
• run jupyter notebook and load the dev notebook
• Fiddle with the code as you wish

Fiddle with code

See results

I want different images to my training and testing data

• Remove directories training_data and testing_data
• run dataloader.py <dataset.csv>. It will not redownload any new data but will random new datasets from the existing files