datamazing 5.2.0

Package for working with pandas Dataset, but with specialized functions used for Energinet

Datamazing

The Datamazing package provides an interface for various transformations of data (filtering, aggregation, merging, etc.)

Interface

The interface is very similar to those of most DataFrame libraries (pandas, pyspark, SQL, etc.). For example, a group-by is implemented as group(df, by=["..."]), and a merge is implemented as merge([df1, df2], on=["..."], how="inner"). So, why not just use native pandas, pyspark, etc.?

1. The native libraries have some parts, with a little annoying interface (such as pandas inconsistent use of indexing)
2. Ability to add custom operations, used specifically for the Energinet domain.

Backends

The package contains methods with the same interface, but for different backends. Currently, 2 backends are supported: pandas and pyspark (though not all methods are available for both). For example, when working with pandas DataFrames, one would use

import pandas as pd
import datamazing.pandas as pdz

df = pd.DataFrame([
    {"animal": "cat", "time": pd.Timestamp("2020-01-01"), "age": 1.0},
    {"animal": "cat", "time": pd.Timestamp("2020-01-02"), "age": 3.0},
    {"animal": "dog", "time": pd.Timestamp("2020-01-01"), "age": 5.0},
])

pdz.group(df, by="animal") \
    .resample(on="time", resolution=pd.Timedelta(hours=12)) \ 
    .agg("interpolate")

whereas, when working with pyspark DataFrame, one would instead use

import datetime as dt
import pyspark.sql as ps
import datamazing.pyspark as psz

spark = ps.SparkSession.getActiveSession()

df = spark.createDataFrame([
    {"animal": "cat", "time": dt.datetime(2020, 1, 1), "age": 1.0},
    {"animal": "cat", "time": dt.datetime(2020, 1, 2), "age": 3.0},
    {"animal": "dog", "time": dt.datetime(2020, 1, 1), "age": 5.0},
])

psz.group(df, by="animal") \
    .resample(on="time", resolution=pd.Timedelta(hours=12)) \ 
    .agg("interpolate")

Development

To setup the Python environment, run

$ pip install poetry
$ poetry install

To run test locally one needs java. This can be installed using the following:

$ sudo apt install default-jdk

To execute unit tests, run

$ pytest .