crypto_forecaster

Generates predictions for a list of cryptocurrencies. Made to be just a framework, performance can be improved with more ML models (LSTM, Reinforcement, SVR..) google-site-verification: google480abb259e214884.html

---

Project maintained by ksink7 Hosted on GitHub Pages — Theme by mattgraham

Cryptocurrency Forecaster

Update Coming soon adding a tikinter application for executing various parts of the forecaster. Future updates will build in live graphs as well as a backtesting function for custom trading strategies. Currently working on a reinforcement learning agent and will add into the backtesting function upon completion to be used as a potential strategy

Prediction Builder: Maintains a database of historical pricing data, builds out features to be used for ML models, trains those ML models, and logs predictions to csv file.

Predition_grapher: Graphs above predictions against actual close prices.

Prediction Builder Main Function does the follwing in order, all default to No

1. Updates Database (CURRENTLY JUST BITTREX DATA, MORE EXCHANGES WILL BE ADDED IN THE FUTURE)
2. Rebuilds Feature Set (CURRENTLY ONLY FEATURES FOR LINEAR REGRESSION, MORE TO COME)
    2a. As more ML models are added to section 3, feature set builders need added here
3. Rebuilds and saves out models to be used for projections
    3a. Input other models here (LSTM, SVR..)
4. Update Prediction Logs (CURRENTLY JUST OUTPUTS TO CSV FILES, MORE    EFFICIENT TECHNIQUES TO COME)

Arguments to run prediction builder from command line:

-D - Update Database (ex. D=Yes will update the database for all coins set in prediction_builder file)
-F - Update Feature datasets
-M - Update all models
-L - Update prediction logs

***All default to No, can change within the main function in prediction_builder***

Command Line Example:
crypto_forecaster $ python prediction_builder.py D=Yes F=Yes M=Yes L=Yes
    -This will rerun all sections of the program from database updating to building the logs

Arguments to run prediction_grapher from command line:

### Argument
1. historical  :   graphs predictions up to last date in raw data files
2. -h  :   Help option, brings up supported arguments
3. future  :   graphs predictions up to last date, and projected out forecast periods (NOT TESTED)

### Keyword Arguments
-b :   base currency (USDT is default)
-c  :   coin (XRP is default)
-i  :   interval (fiveMin is default)
-f  :   future periods to predict (5 is default)
-m  :   model (Linear Regression is default)

--Command Line Example:
crypto_forecaster $ python prediction_grapher historical c=ETH i=hour

Improvements to come:

-requirements.txt file\ -Replace function arguments with correct arg/kwargs syntax (main function corrected)\ -Ability to pass arguments from command line (currently only functional for prediction_grapher.py)\ -Accurate prediction log times(time zone issues have been corrected and defaults to US/Mountain)\ -Improved graphing options(Will be a dash or flask app that graphs future projections LIVE on any timeframe)\ -Backtesting file for testing model projections with various strategies (eventual reinforcement learning algo will replace this)\ -Live/paper trading program for further testing/verification -Orderbook database for more features to use in different models -Web Scraper to be used for sentiment analysis

---

-PROGRAM IS CURRENTLY JUST A FRAMEWORK FOR BUILDING PREDICTIVE ML MODELS AT SCALE -MODEL/FEATURE BUILDER HELPER FILES CONTAIN FUNCTIONS FOR A SIMPLE LINEAR REGRESSION MODEL FORECASTING AT VARIOUS PERIODS -NEW TECHNIQUES (LIKE LSTM NETWORKS) CAN BE ADDED INTO THESE HELPER FUNCTIONS AND RAN FROM THE MAIN PREDICTION BUILDER FILE *** google-site-verification: google480abb259e214884.html