simplemlplatform 1.0.1

A lightweight MLOps SDK for experiment tracking and model registry

SimpleML SDK

A lightweight Python SDK for the SimpleML MLOps platform. Track experiments, log metrics, manage artifacts, and register models with ease.

Installation

pip install simplemlplatform

Quick Start

Basic Configuration

import simpleml

# Configure with your API key (endpoint is pre-configured)
simpleml.configure(api_key="your_api_key_here")

# Start tracking
experiment = simpleml.get_or_create_experiment("my-experiment")
run = experiment.start_run()

# Log metrics
run.log_metrics({
    "accuracy": 0.95,
    "loss": 0.05
})

# Log hyperparameters
run.log_params({
    "learning_rate": 0.001,
    "batch_size": 32,
    "epochs": 100
})

# Upload artifacts
run.upload_artifact("model.pkl", artifact_type="model")

# Complete the run
run.complete()

Configuration

import simpleml

# Simple configuration - endpoint is pre-configured for the SaaS platform
simpleml.configure(api_key="your_api_key_here")

Features

• Experiment Tracking: Organize your ML experiments
• Metrics Logging: Track training and validation metrics over time
• Artifact Management: Upload and manage model files, plots, and data
• Model Registry: Register and version your models
• Simple Setup: Pre-configured production endpoint, just add your API key
• Lightweight: Minimal dependencies, fast setup

API Reference

Configuration

simpleml.configure(
    api_key="your_api_key_here"   # Required: Your API key (endpoint is pre-configured)
)

Experiments

# Get or create experiment
experiment = simpleml.get_or_create_experiment(
    name="image-classification",
    description="CIFAR-10 classification experiment",
    tags={"dataset": "cifar10", "model": "resnet"}
)

# Get existing experiment
experiment = simpleml.get_experiment("exp-123")

# List experiments
experiments = simpleml.list_experiments()

Runs

# Start a run
run = experiment.start_run(
    name="run-1",
    hyperparameters={"lr": 0.001, "batch_size": 32},
    tags={"optimizer": "adam"}
)

# Get existing run
run = simpleml.get_run("run-123")

Metrics

# Log metrics (always use batch method)
run.log_metrics({
    "train_loss": 0.1,
    "val_loss": 0.15,
    "accuracy": 0.95
}, step=100)

Artifacts

# Upload file
run.upload_artifact("model.pkl", artifact_type="model")

# Upload with custom name
run.upload_artifact("./plots/confusion_matrix.png", 
                   filename="confusion_matrix.png",
                   artifact_type="plot")

# List artifacts
artifacts = run.list_artifacts()

# Download artifact
run.download_artifact("artifact-123", "./downloads/model.pkl")

Models

# Register model from run
model = run.register_model(
    model_name="cifar10-classifier",
    artifacts=["model.pkl", "config.json"],
    description="ResNet model for CIFAR-10",
    stage="staging"
)

# List models
models = simpleml.list_models()

# Update model stage
simpleml.update_model_stage("cifar10-classifier", "v1", "production")

Context Manager Usage

import simpleml

simpleml.configure(api_key="your_api_key_here")

with simpleml.start_run("my-experiment") as run:
    # Your training code here
    for epoch in range(10):
        loss = train_epoch()
        accuracy = validate()
        
        run.log_metrics({
            "loss": loss,
            "accuracy": accuracy
        }, step=epoch)
    
    # Upload model
    run.upload_artifact("model.pkl", artifact_type="model")
    
    # Register model if performance is good
    if accuracy > 0.9:
        run.register_model("my-model", ["model.pkl"])

Environment Variables

You can also configure using environment variables:

export SIMPLEML_API_KEY="your_api_key_here"

Error Handling

from simpleml.exceptions import SimpleMLError, AuthenticationError, NotFoundError

try:
    run = simpleml.get_run("invalid-run-id")
except NotFoundError:
    print("Run not found")
except AuthenticationError:
    print("Invalid API key")
except SimpleMLError as e:
    print(f"SimpleML error: {e}")

License

MIT License - see LICENSE file for details.