Predict with your own table

Start with a labelled training DataFrame and a separate query DataFrame. Choose classification for discrete labels or regression for numeric points; Welt does not guess the task or target.

Declare the target#

fit(train, target="churn") removes the named target from features before upload. Pass exactly one of y or target. The original sklearn form fit(X, y) remains supported, including cloning, Pipeline and cross-validation within checked bounds.

import pandas as pd
from welt import Classifier

# Replace these paths with your own approved local tables.
train = pd.read_csv("labelled_customers.csv")
query = pd.read_csv("new_customers.csv")
model = Classifier(model="tabicl-v2", random_state=9)
model.fit(train, target="churn")
predictions = model.predict(query)
assert predictions.shape == (len(query),)

This template requires your files: train includes a unique string column churn; query includes the same feature columns and excludes churn. For a numeric target use Regressor explicitly. Start's complete synthetic examples run without local data files.

Match the schema#

DataFrame columns must be unique strings. Named query columns can arrive in a different order: they are safely aligned to the fitted names. Missing or extra features reject. Arrays use positional features, so preserve their order. Feature cells can be scalar numeric, categorical or missing values. Infinite values and nested objects reject. Targets cannot be missing; regression targets must be finite numeric values. Target-column convenience requires a DataFrame; read a CSV/Parquet into one before using target=. CSV/Parquet estimator paths with explicit y are convenience in-memory reads; Parquet needs the optional parquet extra. File reading is not resumable upload or arbitrary-size execution.

Keep evaluation explicit#

Separate training and query rows yourself. Respect the task's qualified limits; Welt does not silently sample your table. Preprocessing belongs inside each training fold during cross-validation. Use the same held-out rows and metric when comparing models and keep selection separate from unbiased assessment. Process-based parallel CV workers reauthenticate through WELT_API_KEY; an explicit constructor key is intentionally not serialized.

Inspect the result#

predict returns an array in query row order. Classification probabilities use classes_ order; regression returns target-unit mean points. For operation and version provenance, call model.predict_details(query); inspect pinned fit metadata with model.metadata(). Native probabilities are not conformal coverage and regression mean points are not calibrated intervals.

A new fit creates a new immutable predictor. Reuse the existing one for subsequent queries or reopen it later. Switching model, configuration or task requires a new fit; model selection explains the task profiles.