MLAB · Machine Learning Analytics Bench

MLAB·

Machine Learning Analytics Bench

🔒

Task

Predict a measured quantity from contemporaneous inputs. Scored by leave-one-group-out — never a forecast.

Data

no file — load a sample

0 rows Decline — 5 wells x 31 months, rate against time. For forecasting and the soft sensor.
Screening — 64 wells, one row each, static properties against the gain a job delivered. Still regression, but the question asked of it is a yes/no call, so it is the one that exercises the decision matrix.
Diagnosis — 90 wells with rising water, and whether a production log later found channelling. The label is recorded, not cut from a number, so Classify takes it with threshold mode "already 0/1".
Stimulation — 96 fracturing jobs across four campaigns. Ten parameters against the rate gain that followed; four of them carry nearly all of it.

Columns

Target (y)

Group (held out whole)

Features (X)

load data first

Model families

Run

Every result is audited before it is returned. A blocked run reports its findings and withholds the numbers.

Machine Learning Analytics Bench

Choose a task, load a table, pick the target and the grouping column, then Run.
Nested cross-validation by default · Arps baseline required for forecasts.

Decision matrix — using the model to make a yes/no call

Correlation — features against each other and against the target

Leaderboard — nested score is the honest one

Nested vs naive score