Simple, Light-Weight Data Frames for Python

Dataiter's DataFrame is a class for tabular data similar to R's data.frame, implementing all common operations to manipulate data. It is under the hood a dictionary of NumPy arrays and thus capable of fast vectorized operations. You can consider it to be a light-weight alternative to Pandas with a simple and consistent API. Performance-wise Dataiter relies on NumPy and PyArrow and is close to the level of Pandas. Currently the notable slow part is string data and operations (Dataiter's NumPy StringDType being slower than Pandas' Arrow strings).

Installation

# Latest stable version
pip install -U dataiter

# Latest development version
pip install -U git+https://github.com/otsaloma/dataiter

Quick Start

>>> import dataiter as di
>>> data = di.read_csv("data/listings.csv")
>>> data.filter(hood="Manhattan", guests=2).sort(price=1).head()
.
        id      hood zipcode guests    sqft price
     int64    string  string  int64 float64 int64
  ──────── ───────── ─────── ────── ─────── ─────
0 42279170 Manhattan   10013      2     nan     0
1 42384530 Manhattan   10036      2     nan     0
2 18835820 Manhattan   10021      2     nan    10
3 20171179 Manhattan   10027      2     nan    10
4 14858544 Manhattan              2     nan    15
5 31397084 Manhattan   10002      2     nan    19
6 22289683 Manhattan   10031      2     nan    20
7  7760204 Manhattan   10040      2     nan    22
8 43292527 Manhattan   10033      2     nan    22
9 43268040 Manhattan   10033      2     nan    23
.

Documentation

https://dataiter.readthedocs.io/

If you're familiar with either dplyr (R) or Pandas (Python), the comparison table in the documentation will give you a quick overview of the differences and similarities in common operations.

https://dataiter.readthedocs.io/en/stable/comparison.html

For Coding Agents

Dataiter's API is small enough that the complete documentation fits in about 50k tokens. Coding agents can read all of it in one go and know the whole API before writing any code. This differs from bigger packages, like Pandas and Polars, which must fight against outdated training data and introduce their huge API piecemeal, hoping the agent knows what to ask for. For best results, have your agent download llms-full.txt and read it in full, rather than fetch it with a tool that summarizes web pages. Each documentation page is also available as Markdown by replacing .html with .md in the URL.

• https://dataiter.readthedocs.io/en/stable/llms-full.txt
• https://dataiter.readthedocs.io/en/stable/data-frame.md

Development

Install uv from https://docs.astral.sh/uv/ and then run

uv sync