SDK 0.13.1Python 3.11–3.13

From a DataFrame
to a prediction.

Upload a table to your private Welt workspace, fit a qualified model and predict from Python. The same SDK reopens predictors and runs native causal discovery.

Make your first prediction Python reference

Set up in three steps

Full walkthrough

Install

Setup guide

uv pip install 'welt-client[parquet]'

pip install 'welt-client[parquet]'

Connect

Connection guide

python

from welt import Client

Client().connect()

Approve the code, workspace and permissions in your browser. Running remotely? Open the printed link on your own computer. Access stays in this Python process.

welt login

Approve in your browser. The CLI saves access privately for this API origin. Run Python on the same computer.

python

import os
from getpass import getpass

os.environ["WELT_API_KEY"] = getpass("Welt API key: ")

Use your own read/write key from Workspace → API keys. The key stays in this Python process environment; it is never saved in code.

Train, predict, evaluate

Full walkthrough

Load Iris below: 150 rows, four features, classes 0–2. The target column is target. Before downloading or fitting, open Workspace → Billing and check your credits and spending limits.

python

from welt import Client

data = Client().load_research_dataset(
    "asset-d839644b6de36fc5cafdda18c5daeca3"
)
target = "target"
data.head()

Row	sepal-length	sepal-width	petal-length	petal-width
0	4.6	3.6	1.0	0.2
1	4.3	3.0	1.1	0.1
2	5.0	3.2	1.2	0.2
3	5.8	4.0	1.2	0.2
4	4.4	3.0	1.3	0.2

python

from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, classification_report
from welt import Classifier

train, test = train_test_split(
    data, test_size=0.2, random_state=9, stratify=data[target]
)
X_train, y_train = train.drop(columns=target), train[target]
X_test, y_test = test.drop(columns=target), test[target]

model = Classifier(model="tabicl-v2", random_state=9)
model.fit(X_train, y_train)
predictions = model.predict(X_test)

print("accuracy:", accuracy_score(y_test, predictions))
print(classification_report(y_test, predictions, zero_division=0))

              precision    recall  f1-score   support

           0       1.00      1.00      1.00        10
           1       1.00      1.00      1.00        10
           2       1.00      1.00      1.00        10

    accuracy                           1.00        30
   macro avg       1.00      1.00      1.00        30
weighted avg       1.00      1.00      1.00        30

Earlier recorded Iris run · TabICL v2 · seed 9 · 120 training / 30 held-out rows. This report is one observed result, not a promised score. Fit uploads the training table; prediction runs remotely.

For your own table, replace target with your target column.

What do you want to do?

Classify rowsPredict a category for each row of a table.Predict a numberRegression that returns values in your target's units.Use your own tableDataFrames, CSV and Parquet with a named target.Discover a causal graphNative graphs from observational data.Research datasetsSearch, inspect and load real tables to practise on.Reuse a predictorPredict again later without another fit.

Available models

Models and limits

Model	Task	Context (rows × columns)	Prediction rows per call	Classes
Kumo Tabular Medium `kumo-medium`	Classification	8,000 × 20, 4,000 × 40, 2,000 × 80	up to 100	up to 10
Kumo Tabular Medium `kumo-medium`	Regression	8,000 × 20, 4,000 × 40, 2,000 × 80	up to 100	—
Mitra-v2 `mitra-v2`	Classification	4,000 × 20, 2,000 × 40, 500 × 80	up to 100	up to 10
Mitra-v2 `mitra-v2`	Regression	4,000 × 20, 2,000 × 40, 500 × 80	up to 100	—
TabICLv2 `tabicl-v2`	Classification	8,000 × 20, 4,000 × 40, 2,000 × 80	up to 100	up to 10
TabICLv2 `tabicl-v2`	Regression	8,000 × 20, 4,000 × 40, 2,000 × 80	up to 100	—
TabDPT 1.3 `tabdpt-1.3`	Classification	500 × 20	up to 100	up to 10
TabDPT 1.3 `tabdpt-1.3`	Regression	500 × 20	up to 100	—
AVICI `avici`	Causal discovery	1,000 × 20, 500 × 40	—	—
CDFM `cdfm`	Causal discovery	8,000 × 20, 4,000 × 40, 2,000 × 80	—	—
Arrow / ArrowFM `arrow`	Causal discovery	8,000 × 20, 4,000 × 40, 2,000 × 80	—	—

Public catalogue snapshot from 2026-10-15. Context is the training table the model conditions on, or the whole table for causal discovery. It fits when its rows and columns are both within one listed shape, so wider tables take fewer rows. Supervised columns count features without the target; causal discovery counts every variable. Causal discovery needs at least 2 rows and 2 variables. Batch prediction handles larger prediction sets. The live catalogue is authoritative: Client().models() reports current limits and readiness.

Row	sepal-length	sepal-width	petal-length	petal-width
0	4.6	3.6	1.0	0.2
1	4.3	3.0	1.1	0.1
2	5.0	3.2	1.2	0.2
3	5.8	4.0	1.2	0.2
4	4.4	3.0	1.3	0.2

Row	sepal-length	sepal-width	petal-length	petal-width
0	4.6	3.6	1.0	0.2
1	4.3	3.0	1.1	0.1
2	5.0	3.2	1.2	0.2
3	5.8	4.0	1.2	0.2
4	4.4	3.0	1.3	0.2

From a DataFrameto a prediction.

Set up in three steps

What do you want to do?

Available models

From a DataFrame
to a prediction.

Row	sepal-length	sepal-width	petal-length	petal-width
0	4.6	3.6	1.0	0.2
1	4.3	3.0	1.1	0.1
2	5.0	3.2	1.2	0.2
3	5.8	4.0	1.2	0.2
4	4.4	3.0	1.3	0.2