daniel-cortez-stevenson / crypto-predict / 103 / 1

70%

master: 70%

push

travis-pro

Introduced connexion framework to define and execute API
1. Removed root index endpoint and added /greeting and /description
2. json.py functionality handled by connexion so removed
3. This broke loaded Keras models. Created new class in models.py
  - inits a tf graph with model loading
  - the graph provides context for model to predict
4. Low volatility resulted in 0. values after DWT smoothing
  - reduced threshold for including a change by a factor of 2
5. Reduced standard time to train on to 6000 hours
  - more recent market data
  - faster
  - probably is some dependence of the model on magnitude of data
    - this is not good for future - address this
6. TBD:
  - the SavedRegressionModel should be removed
  - whole model module should be checked
  - Add more tests!
  - simplify testing interface
  - simplify crypr structure
  - 'Do one thing' for many functions is currently violated
  - Create api/Docker specific requirements.txt
  - Maybe go to tensorflow instead of Keras? Abstraction is not helping.
  - Generic model for all coins? Hmm...

66 of 134 branches covered (49.25%)

Branch coverage included in aggregate %.

551 of 966 relevant lines covered (57.04%)

0.57 hits per line