Skip to main content

Configuration file

Zeno options are set using a TOML file such as the following:

Example TOML configuration for an image classification task
view = "image-classification"
metadata = "./metadata.csv"
functions = "./tests/"
models = [
    "./model_epochs_2.pth",
    "./model_epochs_5.pth",
    "./model_epochs_10.pth",
    "./model_epochs_20.pth"
]
data_path = "./data/"
data_column = "id"
data_type = "path"

Which is used to run zeno with zeno config.toml.

Configuration options​

The configuration file has the the following options.

view​

The Zeno View used for the given data and task, e.g. image-classification. Required

See Instance views for a list of available views.

metadata​

Path to a csv or parquet file with metadata for data instances. Required

Must have a column with unique identifiers for each instance, specificed by id_column.

Jupyter: A Pandas DataFrame.

functions​

Path to a directory with Python files implementing Zeno functions.

Jupyter: A list of functions with Zeno decorators.

models​

List of models to test.

Can either be paths or strings which are passed directly to the load_model function.

id_column​

The column with unique identifiers in the metadata file. Optional

By default it is id.

data_column​

The column with either file names for data instances or raw data (e.g. text). Optional

By default it is id.

label_column​

The column with either file names for data instances or raw data (e.g. text). Optional

By default it is label.

data_path​

The directory with data instances. Optional

label_path​

The directory with label files. Optional

batch_size​

The batch size used when running inference. Optional

cache_path​

Where cache outputs are stored. Optional

By default it is .zeno_cache in the same directory as the TOML file. Cache stores preprocesssing and model outputs.

samples​

Number of samples to show by default in view. Default 30. Optional

calculate_histogram_metrics​

Whether to calculate metrics for each histogram's bins to color them by metric. Set to false if you have a large dataset where calculating metrics can be costly. Default True. Optional

editable​

Whether users should be able to create/delete/edit slices, folder, and reports. If set to false any changes to the frontend will not be saved in the backend. Default True. Optional

serve​

Whether Zeno should be served on localhost. If False, the FastAPI instance is returned. Useful for hosting on services such as AWS Lambda. Default True. Optional

host​

The host to run Zeno on, should be changed to run on custom domains. Default localhost. Optional

port​

The port to run Zeno on. Optional