Environment:Shiyu coder Kronos Comet ML Logging

Knowledge Sources	Kronos Comet ML
Domains	Experiment_Tracking, Infrastructure
Last Updated	2026-02-09 13:47 GMT

Overview

Optional Comet ML experiment tracking integration for logging training metrics, hyperparameters, and model performance during Qlib finetuning.

Description

The Qlib finetuning pipeline integrates Comet ML for experiment tracking. When enabled (`use_comet=True` in config), training metrics (loss, learning rate, validation loss) are logged to a Comet ML workspace. The integration is optional and can be disabled by setting `use_comet=False`. The Comet API key and workspace must be configured in `finetune/config.py` or loaded from environment variables.

Usage

Use this environment when you want to track and visualize Qlib finetuning experiments. Disable it for local development or when Comet ML access is unavailable. This environment is only relevant to the Qlib Finetuning Pipeline workflow.

System Requirements

Category	Requirement	Notes
Network	Internet access	Required for sending metrics to Comet ML servers

Dependencies

Python Packages

`comet_ml` (latest version)

Credentials

The following must be configured in `finetune/config.py` or via environment variables:

`COMET_API_KEY`: Comet ML API key for authentication
Comet workspace name (configured in `comet_config['workspace']`)
Comet project name (configured in `comet_config['project_name']`)

Quick Install

pip install comet_ml

# Set API key via environment variable (recommended)
export COMET_API_KEY="your_api_key_here"

Code Evidence

Comet ML configuration with placeholder values from `finetune/config.py:75-84`:

self.use_comet = True # Set to False if you don't want to use Comet ML
self.comet_config = {
    # It is highly recommended to load secrets from environment variables
    # for security purposes. Example: os.getenv("COMET_API_KEY")
    "api_key": "YOUR_COMET_API_KEY",
    "project_name": "Kronos-Finetune-Demo",
    "workspace": "your_comet_workspace" # TODO: Change to your Comet ML workspace name
}

Comet ML experiment initialization from `finetune/train_predictor.py:199-208`:

if config['use_comet']:
    comet_logger = comet_ml.Experiment(
        api_key=config['comet_config']['api_key'],
        project_name=config['comet_config']['project_name'],
        workspace=config['comet_config']['workspace'],
    )

Common Errors

Error Message	Cause	Solution
`ModuleNotFoundError: No module named 'comet_ml'`	Comet ML not installed	`pip install comet_ml` or set `use_comet=False`
`CometRestApiException: Invalid API key`	Wrong or placeholder API key	Replace `YOUR_COMET_API_KEY` with actual key or use `os.getenv("COMET_API_KEY")`
`ConnectionError`	No internet access	Set `use_comet=False` for offline training

Compatibility Notes

Optional: Set `use_comet=False` in config to skip entirely. Training proceeds normally without it.
Security: The config file contains placeholder API keys. Always load secrets from environment variables in production.
Scope: Only used by Qlib finetuning scripts (`train_tokenizer.py`, `train_predictor.py`). Not used by CSV finetuning or inference.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment