ai_library

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

Package Overview

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.
  • returns predictions, first_predicted, last_predicted, data_frequency, conformal_q
  • conformal_q is the safety margin in order to be 1-alpha accurate for the predicted time window

• ai_library.record

  • 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.

How to Use

Install

pip install ai-library-swch==0.1.1 

a safe python version is 3.12.3

Configuration

this is an example of how config can be printed and updated.

from ai_library import validate_config, update_config, show_config

update_config(None, updates=["parquet_train_size", 100000])
show_config()
validate_config()

Metric Recorder Function

Make sure that .env is reachable in the folder where this is executed, and that the specified port is reachable. Sample .env configuration:

MON_CLIENT_STOMP_HOST=127.0.0.1
MON_CLIENT_STOMP_PORT=61622

Execution command:

nohup python3 -c "from ai_library import record; record.main()" > record.log 2>&1 &

nohup is not necessary, but make sure to run the recorder in the background.

Train and Infer

import the functions:

from ai_library import train, infer

Manually use train (no cron)

the standard way is to use the cron functions to train in intervals. This might still be useful, for example when switching models.

train()

Cron Scheduling

add_to_cron creates a cron job that runs the train function in intervals.

import ai_library.codebase.setup.cron_manager as cron_manager

cron_manager.add_to_cron()
cron_manager.remove_from_cron()

Inference

after a trained model is accessible via the path, that is given in the config file, inference can be applied:

predictions, first_predicted, last_predicted, data_frequency, conformal_q = infer()