1. Run setup.py

2. Comment out sending email and logging to twitter. This functionality will only work if you have my passwords.

3. Open initial_fit.py and select the parameters you desire (i.e. num_timesteps and num_targets). num_timesteps is an input to the neural network where num_targets represents an output.

4. Run initial_fit.py.

5. You can now run online_fit.py. This will call the online_fit() function. This function will return the future predictions (how many will depend on your selected value for num_targets). Since this will no longer log to Twitter you'll need to edit this file to do some sort of logging yourself. I would suggest calling print() on the function itself (online_fit).

got3/ - Directrory containing an open source tool to get tweets without using Twitter's official API. Twitter's official API only returns tweets up to two weeks old which will not do.

model/model.py - Contains a function to build the model.

date_handler.py - Contains function to bypass the official Twitter API by scraping the website with rotating proxies.

fit_functions.py - Contains functions to initial/online fit the model along with other utility functions.

get_db_info.py - Prints every record of embedded sqlite3 database to stdout.

initial_fit.py - Fits the model on the past few years of price/sentiment data.

online_fit.py - Fits the model since the last performed fit (be it initial or online).

proxy_selector.py - Contains function to select a proxy from a list of free proxies.

requirements.txt - Contains all the dependencies needed to run the project. Use pip to install the dependencies.

send_email.py - Contains a function to send an email notification containing a graph representing the initial fit. Called from the initial_fit() function.

setup.py - Run this script to setup the project for use.

twitter_logger.py - Contains function to log predictions to twitter. This function is called after every online_fit.

1. Perform an initial fit on a model that takes in more tweets and only predicts for one timestep.

2. Start logging to twitter and observe winrate.

3. Fix online_fit learning rate.

4. Add features to dataset.

5. Implement q learning.

6. Link to some exchange's trading API (either ETH/USD or ETH/Tether).

7. Smart contract implementation.