Machine Learning basics for Bitcoin Price Prediction

1. Tick Data Scraping

Download the data from coinbase using the script coinbase_scrapper.py. When run the script will create a gdax.h5 file using the pandas.HDFStore format to save all the ticks. Because the gdax api returns the data in 100 tick batches and the total number of ticks across all pairs is 100 million++ it will take more than a week to download all the data. You may want to change the pairs to be requested to a smaller subset of pairs within the coinbase_scrapper.py.

You can parse the ticks downloaded into candles using the efficient_candle_load from the utils.py file.

2. Tweet Data Scraping

To gather tweet data you need to first setup a twitter developer account create an app and create a secrets.json file in the same folder as the live_tweet_gather.py script with your api key, secret, token and token_secret. An example of this file is as follows:

{
"twitter": {
    "key": "xxxxxxxxxxxxxxxxxxxx",
    "secret": 
    "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    "token": 
    "xxxxxxxxxx-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    "token_secret": 
    "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
  }
}

(Replace the xxxxxx with your own accounts info)

After creating the secrets.json file you can run live_tweet_gather.py to start gathering tweets with the queries specified within the script ('bitcoin', 'BTC', 'cryptocurrency', 'crypto' by default)

Whenever you want to prepare that data to be run with a model you need to run the create_dataset function within tweets.py . This will create a tweets.h5 file with the tokenized tweet words.

Run

At this point if you have gathered some data you can run jupyter notebook and open the Slides.ipynb notebook from the webui and start executing the notebook cells. Of course you can do that regardless of having data if you just want to have a nice ui to look through the code.