diagonalpy 0.2.0

deploy linear models from scikit-learn to diagonal.sh

diagonalpy

diagonalpy is a Python library for deploying scikit-learn linear models to the inference platform diagonal.sh

Features

• Save scikit-learn linear models locally, as onnx file
• Deploy scikit-learn linear models to the diagonal.sh inference platform
• Delete models deployed on the diagonal.sh inference platform

Installation

Currently, diagonalpy is available for Python 3.9, 3.10, 3.11 and 3.12. Support for 3.13 will be added as soon as dependencies allow it.

pip install diagonalpy

torch is a dependency of diagonalpy, so if it isn't installed in the installation environment, you'll also have to run

pip install torch

Quick Start

Save a Model

import numpy as np
from sklearn.linear_model import LinearRegression
from diagonalpy.save import save

# Train a scikit-learn model
model = LinearRegression()
X = np.random.randn(100, 10)
y = np.sum(X, axis=1) + np.random.randn(100)
model.fit(X, y)

save(model, "/my-path/model.onnx")

You can then upload the model manually at console.diagonal.sh.

Deploy a Model

Once you have an account and created a console key, you can set it to the env variable DIAGONALSH_API_KEY, and deploy directly. You'll also have to set the env variable DIAGONALSH_REGION to eu-west-3. This is the snippet:

import os
os.environ['DIAGONALSH_API_KEY'] = 'dia_console_...'
os.environ['DIAGONALSH_REGION'] = 'eu-west-3'

Then, you can deploy directly from diagonalpy:

from diagonalpy.deploy import deploy

deploy(model, "my-wonderful-model")

The deployed model will be available for inference usually after 10 to 15 minutes.

Inferring your deployed model

If you want to get the model output for the model you deployed, you can do this via a HTTP POST request, like this:

curl -X POST -H "Content-Type: application/json" -H "X-API-Key: YOUR_INFERENCE_API_KEY" -d "{\"data\": YOUR_DATA, \"model\":\"YOUR_MODEL_ID\"}" "https://infer.diagonal.sh/YOUR_MODEL_ROUTE" &

The data for a model like the one trained above would look like this:

[-5.0, 2.0, 0.0, 1.0, 3.0, -2.0, 1.0, 0.0, 2.0, 5.0]

YOUR_DATA should be replaced with an unnested list containing comma-separated values that can be integers or floats, but will be interpreted as floats either way.

The inference_api_key can be created under Settings on console.diagonal.sh. The model_id and model_route are visible on the models overview on console.diagonal.sh.

Delete a deployed model

Deleting a model on diagonal.sh is easy with diagonalpy:

from diagonalpy.delete import delete

delete("model-id-from-deployment")

Supported Models

Regression Models:

• LinearRegression
• Ridge
• RidgeCV
• Lasso
• LassoCV
• ElasticNet
• ElasticNetCV
• Lars
• LarsCV
• LassoLars
• LassoLarsCV
• LassoLarsIC
• OrthogonalMatchingPursuit
• OrthogonalMatchingPursuitCV
• BayesianRidge
• ARDRegression
• HuberRegressor
• QuantileRegressor
• TheilSenRegressor
• TweedieRegressor

Classification Models

• LogisticRegression
• LogisticRegressionCV
• SGDClassifier
• Perceptron
• PassiveAggressiveClassifier
• RidgeClassifier
• RidgeClassifierCV

Environment Variables

DIAGONALSH_API_KEY: Your Diagonal.sh API key (required)

DIAGONALSH_REGION: AWS region for deployment (required) - currently, only "eu-west-3" is valid

Environment Setup

export DIAGONALSH_API_KEY="your_api_key"
export DIAGONALSH_REGION="eu-west-3"

License

This package is distributed under CC BY-ND license, which allows commercial use of the unmodified software and prohibits the distribution of any modifications of this software.