An extension of pandas for efficient representation of nested associated datasets.
Nested-Pandas extends the pandas package with tooling and support for nested dataframes packed into values of top-level dataframe columns. Pyarrow is used internally to aid in scalability and performance.
Nested-Pandas allows data like this:
To instead be represented like this:
Where the nested data is represented as nested dataframes:
# Each row of "object_nf" now has it's own sub-dataframe of matched rows from "source_df"
object_nf.loc[0]["nested_sources"]
Allowing powerful and straightforward operations, like:
# Compute the mean flux for each row of "object_nf"
import numpy as np
def mean_flux(row):
"""Calculates the mean flux for each object"""
return np.mean(row["nested_sources.flux"])
object_nf.map_rows(mean_flux, output_names="mean_flux")
Nested-Pandas is motivated by time-domain astronomy use cases, where we see
typically two levels of information, information about astronomical objects and
then an associated set of N measurements of those objects. Nested-Pandas offers
a performant and memory-efficient package for working with these types of datasets.
Core advantages being:
- hierarchical column access
- efficient packing of nested information into inputs to custom user functions
- avoiding costly groupby operations
This is a LINCC Frameworks project - find more information about LINCC Frameworks here.
This project is supported by Schmidt Sciences.