ai_library

ai_library is a Python package that exposes model training, inference, metrics recording, cron scheduling, and package-based configuration management.

Package overview

The package is structured so that import ai_library is safe and does not execute heavy runtime logic. Core user-facing functionality is available through lazy exports exposed from ai_library/__init__.py.

Available top-level APIs

• ai_library.validate_config(config_path=None)

  • Loads and validates the package default ai_library/config.yaml if no path is provided.
  • Returns the parsed configuration dictionary.

• ai_library.update_config(config_path=None, updates=...)

  • Updates the package config file by default.
  • Supports a dictionary, a [key, value] pair, or a list of [key, value] pairs.

• ai_library.show_config(config_path=None)

  • Prints the current configuration to stdout.

• ai_library.train()

  • Loads configuration from the package config.
  • Reads data, builds the selected pipeline, and trains the model.

• ai_library.infer()

  • Loads package configuration.
  • Loads a saved model and performs inference.

• ai_library.record

  • Lazy-imported recorder module.
  • Use ai_library.record.main() or run the module directly to start metric collection.

• ai_library.add_to_cron() / ai_library.remove_from_cron()

  • Manage cron scheduling for recurring training runs.
  • Adds or removes a cron job that runs python3 -m ai_library.codebase.setup.train.

Installation

From source:

python3 -m pip install -e .

From a built wheel or sdist:

python3 -m pip install dist/ai-library-swch-0.1.0-py3-none-any.whl

Configuration

The default configuration file lives inside the package at ai_library/config.yaml. This package is designed so that the default config is editable through the helper API:

from ai_library import validate_config, update_config, show_config

config = validate_config()
print(config)

update_config(None, {"pipeline_type": "gmlp"})
show_config()

Note: editing ai_library/config.yaml via update_config(None, ...) works cleanly during development or editable installs. If the package is installed from a read-only wheel, consider using an explicit config path or a custom location.

Usage

Training

python3 -m ai_library.codebase.setup.train

Or programmatically:

from ai_library import train
train()

Inference

python3 -m ai_library.codebase.setup.infer

Or programmatically:

from ai_library import infer
infer()

Metric recording

python3 -m ai_library.codebase.setup.record --out-dir ./data --run-seconds 60

Or programmatically:

from ai_library import record
record.main()

Cron scheduling

from ai_library import add_to_cron, remove_from_cron
add_to_cron()
remove_from_cron()

File structure

• ai_library/
  • __init__.py — lazy exports and package-level API surface.
  • config.yaml — default package config.
  • codebase/
    • helpers/ — helper utilities such as config_helper.py.
    • models/ — model pipeline classes.
    • setup/ — training, inference, recording, and cron management modules.

Notes

• ai_library.record is lazy-loaded so importing ai_library does not import heavy recorder dependencies until you actually access it.
• The package currently relies on package-relative config loading, so the package config file is the main runtime configuration source.
• Model training and inference rely on the dataset, saved model files, and other paths configured in ai_library/config.yaml.

Recommended next improvements

• Add a runtime override for external config files.
• Add automated tests for training and inference flows.
• Document environment dependencies and the exact requirements.txt contents.