CLI Reference

The make-mlops-easy command line interface wraps the distributed runtime and exposes high-level pipeline workflows. Commands are implemented with Click and share a consistent set of options such as --config, --master-url, and --poll-interval.

make-mlops-easy [COMMAND] [OPTIONS]

Run make-mlops-easy --help to inspect the top-level commands or append --help to any command for detailed usage.

Workflow commands

train

make-mlops-easy train DATA_PATH [--target COLUMN] [--config PATH] [--no-deploy]

Option	Description
DATA_PATH	Dataset to train on (CSV, JSON, or Parquet).
--target, -t	Target column name. Defaults to the last column when omitted.
--config, -c	Apply a YAML configuration file.
--no-deploy	Skip the deployment stage (training still runs).
--master-url	URL of the master service (http://127.0.0.1:8000 by default).
--poll-interval	Seconds between workflow status checks.

The command prints progress, metrics, and the deployment directory (unless deployment is disabled).

predict

make-mlops-easy predict DATA_PATH MODEL_DIR [--config PATH] [--output PATH]

Option	Description
DATA_PATH	Dataset to score.
MODEL_DIR	Deployment directory produced by train.
--output, -o	Optional path to save predictions ({"predictions": [...]}).
--config, --master-url, --poll-interval	Same as train.

Predictions are also printed to the console.

status

make-mlops-easy status MODEL_DIR [--config PATH]

Displays deployment metadata, evaluation metrics, and monitoring summaries loaded from MODEL_DIR.

observe

make-mlops-easy observe MODEL_DIR [--config PATH] [--output PATH]

Produces a formatted observability report and optionally saves it to disk.

Runtime management

master start

make-mlops-easy master start [--host HOST] [--port PORT] [--state-path PATH]

Runs the FastAPI master service that accepts workflow requests from the CLI and coordinates workers.

• --host / --port – bind address and port (defaults: 127.0.0.1:8000).
• --state-path – optional JSON file used to persist workflow state (~/.easy_mlops/master_state.json by default).

The process blocks until interrupted (Ctrl+C).

worker start

make-mlops-easy worker start [--master-url URL] [--worker-id ID] [--poll-interval SECONDS] [--capability TAG ...]

Launches a worker agent that polls the master for tasks and executes them via TaskRunner.

• --master-url – master endpoint (defaults to http://127.0.0.1:8000 or EASY_MLOPS_MASTER_URL).
• --worker-id – custom identifier; autogenerated when omitted.
• --poll-interval – seconds between polling requests (minimum 0.5s).
• --capability – repeatable flag that tags the worker (future task routing hook).

Workers stream task output (stdout) back to the master so the CLI can render it after completion.

Project scaffolding

init

make-mlops-easy init [--output PATH]

Creates a default mlops-config.yaml you can customise and pass to other commands via --config. The file mirrors the defaults baked into easy_mlops/config/config.py.

Environment variables and defaults

Variable	Purpose	Default
EASY_MLOPS_MASTER_URL	Overrides the master URL used by CLI commands.	http://127.0.0.1:8000
EASY_MLOPS_POLL_INTERVAL	Controls the default polling interval in seconds.	2.0

CLI options take precedence over environment variables.

Using the Makefile

Once you have installed the project with make install-dev, you can invoke the CLI through shortcuts that automatically resolve the virtual environment:

make train DATA=examples/sample_data.csv TARGET=approved
make predict DATA=examples/sample_data.csv MODEL_DIR=models/deployment_20240101_120000
make status MODEL_DIR=models/deployment_20240101_120000
make observe MODEL_DIR=models/deployment_20240101_120000

Custom arguments can be appended through the ARGS variable, for example make train DATA=... ARGS="--config configs/quickstart.yaml".