Quick start

This walkthrough takes a scikit-learn classifier from a Python session to a saved, reviewable inference. It uses the demo model behind the screenshots in this manual: a decision tree that approves or declines a synthetic loan application.

1. Export a bundle

MLsuite accepts models only as .mlbundle files written by MLbundle. Export with the MLbundle and scikit-learn versions the instance pins in backend/pyproject.toml (MLbundle 0.2.1 and scikit-learn 1.9.1 at the time of writing); a scikit-learn bundle loads only with the version that wrote it.

python -m pip install "mlbundle[sklearn,schema]==0.2.1" "scikit-learn==1.9.1" pandas

export_demo.py

import numpy as np
import pandas as pd
from sklearn.tree import DecisionTreeClassifier

from mlbundle import export

rng = np.random.default_rng(7)
frame = pd.DataFrame({
    "age": rng.integers(21, 70, 400),
    "income": rng.normal(42000, 12000, 400).round(0),
    "has_guarantor": rng.random(400) < 0.3,
    "employment": rng.choice(["salaried", "self_employed", "unemployed"], 400),
})
score = (
    frame.income / 1000
    + frame.has_guarantor * 15
    - (frame.employment == "unemployed") * 25
)
y = np.where(score > 40, "approve", "decline")
X = pd.get_dummies(frame, columns=["employment"], prefix_sep="__", dtype=int)
model = DecisionTreeClassifier(max_depth=4, random_state=7).fit(X, y)

export(model, "demo-loan-tree.mlbundle", training_data=X)

Passing training_data= lets MLbundle infer the form fields from the training frame: a number, a true/false choice and one employment choice built from the three one-hot columns. Without it, every feature becomes a number field.

2. Upload, describe, run

1. Upload the model. Sign in, open Models and choose New model. Drop demo-loan-tree.mlbundle on the page, check its name and choose Save. See Upload models.

2. Create a schema. Open Schemas, choose New schema, select the model, name the schema and choose Create schema. Its first snapshot, v1, is published at once with the fields and report the bundle carried. See Schemas.

3. Bookmark the snapshot. Open the schema’s Snapshots tab, open the actions of v1 and choose Bookmark. Name it, for example loan-approval, and choose Save bookmark. See Bookmarks.

4. Run it. Open Predict and choose the bookmark. Fill the form and choose Run schema. The result appears in the Results tab and the run joins the Session panel.

5. Save the run. Give the run a name in the session panel and choose Save. It is now an inference, listed under Inferences with its inputs, outputs and the snapshot it ran on. See Run predictions.

Every saved run, one row each, with the schema's inputs and outputs as columns.

Next steps

• Give the fields readable labels and add review questions by publishing a change.
• Run many records at once with a CSV or XLSX upload.
• Ask reviewers to check the results, then export everything.
• Publish the bookmark so anyone can run it, or give it to an API client.