Jupyter Notebook which allows you to create your own dataset with bitcoin OPEN, HIGH, LOW, CLOSE data, through the trades of different exchanges.
The data is taken from:
- Coinbase
- Bitfinex
- Kraken
- Bitstamp
- Itbit
- Okcoin
To do this, I used the trade data provided in csv format by the https://bitcoincharts.com website.
The data starts from 01-01-2012.
To create the candlesticks dataset, the data was aggregated as follows:
maximum of the interval for the HIGH datafirst of the interval for the OPEN dataminimum of the interval for the LOW datalast of the interval for the CLOSE data
The data aggregated is then averaged with the various exchanges.
The volume data of the aggregated dataset is instead the sum of the trades of the corresponding interval but to have a better evaluation of the data only the most complete bitstamp exchange is used.
You can change the interval in which to aggregate the data by changing the compression parameter in the script. Now is set to 60min interval.
There may be some holes in the data where values are missing. I therefore decided to interpolate the data of a single exchange, that is the one with the most complete Bitstamp data. The type of interpolation used is a slinear interpolation which in the case of time series like this makes the data as clean as possible if the dataset were to be used for machine learning models.
To see the related study on interpolation look at this repo: https://github.com/CristianCosci/interpolationStudy_toHandleMissingData_BTCdataset
The data are for educational purposes only, it is not recommended to use this dataset for financial purposes.
There are also some functions able to summarize some characteristics of the obtained datasets.
