Train machine learning models on your CSV
No code and no account. Pick the column to predict, and AnalytiqAI tunes six models in this tab and scores each one on rows it never saw, for files up to 10 MB.
The lab. Your file stays in this tab.
Drop a file or open a sample, choose what to predict, then press Train models. The leaderboard, the column ranking and a CSV of predictions appear in the results panel.
Results show up here
Load a file or a sample, then press Train models. You get a ranked list of models, how far each beats the baseline, and which columns drove the predictions.
Training on your machine
The AutoML method behind six models
One row in five is set aside before training. The rest tunes each model over five rounds, and the top three are blended.
Dates are split into parts, skewed numbers are log-scaled, and busy categories become their average outcome.
The winner is checked once on the held-back rows, then each column is shuffled to rank what mattered.
AnalytiqAI reads your file with code already on this page, trains in a background worker on your own processor, and sends nothing anywhere while it works.
Privacy you can check with a request counter
It reads the browser's own resource log, the one developer tools use, from the moment you load a file.
Train a model and it stays at 0. To check it another way, open the Network tab with F12 and watch it stay empty.
Nothing is stored: no cookies, no local storage, no analytics.
- Gradient boosting1.9 s
- Random forest1.2 s
- Logistic regression0.4 s
- Nearest neighbours0.6 s
- Decision tree0.2 s
- Blend of top 30.1 s
Free Kaggle GPU runs for six more engines
Optional. Sign in with Kaggle under your results, and this page runs the same rows in a private notebook on your own account, including NVIDIA's pretrained KumoTabular model, then deletes the notebook. No install and no AnalytiqAI server.
Questions and limits
Everything runs on your processor, so AnalytiqAI takes files up to 10 MB with 200 columns, and samples 10,000 rows from anything larger.
It predicts a number, or one of 2 to 20 categories. A larger file is refused with a message, and the tab never freezes.
- churn.csvCSV, 117 KB
- tickets.tsvTSV
export.txtText
records.jsonJSON
events.jsonlJSONL
log.ndjsonNDJSON
Does my file leave my computer?
No. It is read into this tab's memory and gone when you close or reload the tab. Only a Kaggle run you start yourself uploads data.
What if my file has more than 10,000 rows?
AnalytiqAI trains on a random sample of 10,000 rows and says so in the results.
Can I use the model on new data?
Yes. After training, drop a second file with the same columns and save its predictions as a CSV.
Does it need an account or any code?
No. It is a free, no-code machine learning tool: drop a CSV, pick a column, and it tunes the models and ranks the columns by importance.
Is a GPU needed?
No. The browser models use your processor. Kaggle's GPU is an option for the six larger engines.