ZenoML-Next

This fork of ZenoML aims to maintain the existing functionality while adding small, practical features for personal use.

pip install zenoml_next

Features

• New entry point: zeno_next() simplifies usage in notebooks with fixed inputs.
• S3 URI support: Enables S3 paths as the image column (requires aws configure on the host machine).

Example: Using zeno_next

import pandas as pd
from zeno import zeno_next

df = pd.read_parquet("tmp_df.parquet")
print(len(df))

# Use local paths:
# data_column = "filename"
# MOUNT_PATH = "/root/danbooru_images"

# Use S3 URI:
data_column = "s3_uri"
MOUNT_PATH = ""

ZENO_PORT = 8019

# Clear the cache if needed:
# !rm -rf .zeno_cache/
zeno_next(df, data_column, MOUNT_PATH, ZENO_PORT)

Upkeeps

• bump pydantic to version 2.x

Building Repo

# see DEVELOPMENT.md for setup repo
python -m pip install build twine
python -m build
twine check dist/*
twine upload dist/*

Original README

Zeno is a general-purpose framework for evaluating machine learning models. It combines a Python API with an interactive UI to allow users to discover, explore, and analyze the performance of their models across diverse use cases. Zeno can be used for any data type or task with modular views for everything from object detection to audio transcription.

Demos

Image Classification	Audio Transcription	Image Generation	Dataset Chatbot	Sensor Classification
Imagenette	Speech Accent Archive	DiffusionDB	LangChain + Notion	MotionSense
code	code	code	code	code

teaser.mp4

Quickstart

Install the Zeno Python package from PyPI:

pip install zenoml

Command Line

To get started, run the following command to initialize a Zeno project. It will walk you through creating the zeno.toml configuration file:

zeno init

Take a look at the configuration documentation for additional toml file options like adding model functions.

Start Zeno with zeno zeno.toml.

Jupyter Notebook

You can also run Zeno directly from Jupyter notebooks or lab. The zeno command takes a dictionary of configuration options as input. See the docs for a full list of options. In this example we pass the minimum options for exploring a non-tabular dataset:

import pandas as pd
from zeno import zeno

df = pd.read_csv("/path/to/metadata/file.csv")

zeno({
    "metadata": df, # Pandas DataFrame with a row for each instance
    "view": "audio-transcription", # The type of view for this data/task
    "data_path": "/path/to/raw/data/", # The folder with raw data (images, audio, etc.)
    "data_column": "id" # The column in the metadata file that contains the relative paths of files in data_path
})

You can pass a list of decorated function references directly Zeno as you add models and metrics.

Citation

Please reference our CHI'23 paper

@inproceedings{cabrera23zeno,
  author = {Cabrera, Ángel Alexander and Fu, Erica and Bertucci, Donald and Holstein, Kenneth and Talwalkar, Ameet and Hong, Jason I. and Perer, Adam},
  title = {Zeno: An Interactive Framework for Behavioral Evaluation of Machine Learning},
  year = {2023},
  isbn = {978-1-4503-9421-5/23/04},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {https://doi.org/10.1145/3544548.3581268},
  doi = {10.1145/3544548.3581268},
  booktitle = {CHI Conference on Human Factors in Computing Systems},
  location = {Hamburg, Germany},
  series = {CHI '23}
}

Community

Chat with us on our Discord channel or leave an issue on this repository if you run into any issues or have a request!