carbonium

Manage a list of names with several properties and (overlapping) order criteria


Keywords
pandas, pandas-dataframe, pipeline, python
License
MIT
Install
pip install carbonium==0.10.4

Documentation

carbonium

Easily manage a list of names with several properties and (overlapping) order criteria.

Installation

Install carbonium is as easy as run pip install carbonium.

Usage

As first step you should define a name list:

name_list = [
    {
        "domains": ["raw", "output"],
        "name": "var1",
        "alias": "column_1_name",
        "output_order": 1,
        "filling_value": 10,
    },
    {
        "domains": ["raw"],
        "name": "var2",
        "alias": "column_2_name",
        "output_order": 2,
    },
    {
        "domains": ["new", "output"],
        "name": "new_var",
        "alias": "new_column_name",
    },
]

Each name definition is a dictionary that contains some common, mandatory key, and some other keys, domain-specific or name-specific.

Mandatory keys are only three:

  • domains, a list of string, each representing a domain
  • name, a string, uniquely identifiers of a name
  • alias, a string that can be used to refers to the name in context where it is named with this alternative string.

Then, each name belongs to some domains. Domains are used to perfom names selection (give me all names belonging to domain). Names that belongs to the same domain should have the same optional attributes.

After name list definition, you can instantiate the structure class:

from carbonium import Structure

structure = Structure(name_list)

Internally, the Structure class iteratively instantiate a Name class for each name definition. After this step you can access to each Name and its properties through c object, but you can also use one of property or method of the class.

print(structure.names)
# returns:  ['var1', 'var2', 'new_var']

print(structure.domains)
# returns: {'new', 'output', 'raw'}

print(
    structure.var1.name,
    structure.var1.domains,
    structure.var1.output_order
)

Calling structure.var1.name you have access to the string associated to var1... and so on.

ordered_raw_columns = [
        (
            i,
            structure.get(i).output_order,
            structure.get(i).get("filling_value")
        )
        for i in structure.get_names('raw')
]

ordered_raw_columns = sorted(
    ordered_raw_columns,
    key=lambda x: x[1]
)

In this example all the names belonging to raw domain are extracted with some other properties. In this way the same name can be used in different domains or contexts by referring to contexctual relevant properties.

import pandas as pd
df = pd.DataFrame([
    {"var1": 100, "var2": 200},
    {"var2": 220},
])

for name in structure.get_names('raw'):
    alias = structure.get(name).alias
    filling = structure.get(name).get("filling_value")
    if filling:
        df[alias].fillna(filling, inplace=True)

for name in structure.get_names('new'):
    alias = structure.get(name).alias
    df[alias] = "arbitrary"    

output_columns = structure.get_names('new')
df[output_columns].to_parquet('output.parquet')

As you can see, whithout modify the code but only the taxonomy described in name_list, you can affect different columns.