Skip to content
This repository has been archived by the owner on Oct 5, 2021. It is now read-only.
/ corintick Public archive

Column-based datastore for historical timeseries data

License

Notifications You must be signed in to change notification settings

gusutabopb/corintick

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

87 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

corintick

CI status Coverage PyPI package Supported Python versions

Column-based datastore for historical timeseries data. Corintick is designed mainly to store pandas DataFrames that represent timeseries.

Instalation

In order to use Corintick you need MongoDB. See installation instructions here.

Corintick itself can be installed with pip:

$ pip install corintick

Quickstart

Initialize Corintick:

from corintick import Corintick
corin = Corintick()

Now we need a DataFrame to insert into Corintick. For demonstration purposes, we will get data from Quandl:

import quandl
df1 = quandl.get('TSE/7203')

Here, df1 looks like this:

              Open    High     Low   Close      Volume
Date
2012-08-23  3240.0  3270.0  3220.0  3260.0   4652200.0
2012-08-24  3225.0  3245.0  3210.0  3235.0   3659600.0
2012-08-27  3250.0  3280.0  3215.0  3220.0   3614600.0
2012-08-28  3235.0  3260.0  3150.0  3180.0   6759100.0
2012-08-29  3180.0  3195.0  3160.0  3175.0   2614800.0
2012-08-30  3180.0  3190.0  3160.0  3170.0   3291700.0
2012-08-31  3135.0  3155.0  3095.0  3095.0   5663800.0
...

Writing

Inserting df1 into Corintick is simple:

corin.write('7203.T', df1, source='Quandl', country='Japan')

The first argument passed to corintick.write is an UID (universal identifier) and must be unique for each timeseries inserted in a given collection. The second argument is the dataframe to be inserted. The remaining keyword arguments are optional metadata tags that can be attached to the dataframe/document for querying.

Reading

Reading from Corintick is also straightforward:

df2 = corin.read('7203.T')

You can also specify start and end as ISO-8601 datetime string...

df2 = corin.read('7203.T', start='2014-01-01', end='2014-12-31')
              Open    High     Low   Close      Volume
2014-01-06  6360.0  6400.0  6280.0  6300.0  12249300.0
2014-01-07  6270.0  6340.0  6260.0  6270.0   7891400.0
2014-01-08  6310.0  6320.0  6260.0  6300.0   7184100.0
2014-01-09  6310.0  6340.0  6260.0  6270.0   8653000.0
2014-01-10  6260.0  6310.0  6250.0  6290.0   7815900.0
...
2014-12-24  7645.0  7687.0  7639.0  7657.0  9287900.0
2014-12-25  7600.0  7655.0  7597.0  7611.0  5362700.0
2014-12-26  7629.0  7700.0  7615.0  7696.0  6069100.0
2014-12-29  7740.0  7746.0  7565.0  7662.0  9942800.0
2014-12-30  7652.0  7674.0  7558.0  7558.0  7821200.0

...and which columns you want retrieved:

df2 = corin.read('7203.T', columns=['Close', 'Volume'], start='2017-05-10')
             Close      Volume
2017-05-10  6081.0   7823700.0
2017-05-11  6123.0  13511900.0
2017-05-12  6047.0   8216600.0
2017-05-15  6009.0   5925200.0
2017-05-16  6093.0   6449300.0
...

Configuration

By default, Corintick tries to use a MongoDB instance running at localhost:27017. This can be changed through the host and port arguments of the Corintick initializer. Similarly, the database to be used by Corintick defaults to corintick and can also be changed using the db parameter. All the data in the db database is assumed to be Corintick data. Avoid having any other process/application reading/writing data to that database.

In case your MongoDB setup requires authentication, you can use the username and password arguments.

See Corintick.__init__ for details.

Collections

Corintick can use multiple collections to better organize data. A Corintick collection is the same as a MongoDB collection. In each collection, only a single dataframe/document can exist for a given UID for a given time period.

In case you need to store two different types of data for a same UID over an overlapping time frame (i.e. trade data and order book data for a given stock), you should separate the two different types of data into different collections.

By default, data is written to the corintick collection. This default collection can be changed by assigning a string to Corintick.default_collection.

>>> corin.collection = 'another_collection'

Collections can also be specified on a method call basis:

df = corin.read('7203.T', collection='orderbook')
corin.write(df, collection='another_collection')

Corintick mechanics

During writing, Corintick does the following:

  1. Takes the input DataFrame and splits into columns
  2. Serializes/compresses each using the LZ4 compression algorithm
  3. Generates a MongoDB document containing the binary blobs corresponding to each column and other metadata

During reading, the opposite takes places:

  1. Documents are fetched
  2. Data is decompressed and converted back to numpy arrays
  3. DataFrame is reconstructed and returned to the user

Background

Corintick was inspired by and aims to be a simplified version of Man AHL’s Arctic.

Differences from Arctic

Corintick has a single storage engine, which is column-based and not versioned, similar to Arctic’s TickStore. However, differently from TickStore, it does support non-numerical object dtype columns by parsing them into MessagePack string objects

Naming

Corintick aimed from the beginning to be a column-based data storage. "Corintick" is a blend of “Corinthan” (style of Roman columns) and "tick".

Benchmarks

TODO

  • vs InfluxDB
  • vs vanila MongoDB
  • vs MySQL
  • vs KDB+ (32-bit)

Contributing

To contribute, fork the repository on GitHub, make your changes and submit a pull request.
Corintick is not a mature project yet, so just simply raising issues is also greatly appreciated :)