Text Classification Sandbox
Teach a small model to label text using your examples. Check it on separate test text and download the predictions and results.
How can I train a model to label text?
Add examples with labels, and mark which ones to use for training or testing. Train the model, then check how it handles the separate test examples and new text. Review the mistakes before using its labels. Download the report and examples together.
Teach it with your examples.
Label a small collection, reserve separate examples for testing, then try new passages. See the model’s scores and the mistakes it makes.
Import a collection from CSV
Use exactly these headers: id,label,role,text. Set role to train or test. Test labels must also appear in training. Quoted multiline text is supported. Review five sample rows before replacing the collection.
Your passages
18 matching passages · Page 1 of 1
| ID | Label | Role | Text preview | Edit |
|---|---|---|---|---|
| garden-1 | garden | train | Plant tomato seeds in warm garden soil. | |
| garden-2 | garden | train | Water the flowers and pull weeds from the garden. | |
| garden-3 | garden | train | The gardener grows fresh vegetables near the greenhouse. | |
| garden-4 | garden | train | Add compost to improve soil around the plants. | |
| garden-5 | garden | train | Rose bushes need sunlight and regular watering. | |
| garden-6 | garden | test | Garden tools help plant seeds and trim flowers. | |
| cooking-1 | cooking | train | Bake the bread in a hot kitchen oven. | |
| cooking-2 | cooking | train | Cook pasta in boiling water and add tomato sauce. | |
| cooking-3 | cooking | train | This recipe combines fresh vegetables and olive oil. | |
| cooking-4 | cooking | train | The chef prepares soup with salt and herbs. | |
| cooking-5 | cooking | train | Mix flour and butter to make pastry dough. | |
| cooking-6 | cooking | test | Kitchen recipes explain how to roast and bake food. | |
| travel-1 | travel | train | Book a flight and pack a suitcase for the trip. | |
| travel-2 | travel | train | The train journey crosses mountains and small towns. | |
| travel-3 | travel | train | Tourists can reserve a hotel near the airport. | |
| travel-4 | travel | train | A passport is needed for international travel. | |
| travel-5 | travel | train | The bus station has tickets for the coastal route. | |
| travel-6 | travel | test | A travel guide helps visitors plan their holiday. |
Choose how to compare text
Words use Unicode letters, digits and underscores, including single-character tokens. Word pairs also include adjacent two-word phrases. Character fragments use 3–5 characters inside whitespace-separated words, with boundary padding. There is no stemming or semantic language model. English stop words only match lower-case forms when Match case is selected.
Running sends all passages and settings to Bookify’s isolated processing service. Temporary job files are cleared after processing; nothing is automatically saved in this page.
Three simple steps
How to use Text Classification Sandbox
Edit the examples or import a CSV and review its rows.
Choose a method and send the experiment to the processing service.
Inspect the results, then download the full analysis and reusable project.
Common questions
Good to know
Understand the result.
Keep the original.
Is this tool free, and are my files uploaded?
This tool is free with no account required. When you run the tool, your input goes to a separate processing service. Temporary files are deleted when the job ends. You can download the result to your device.
What are the limits and details?
Text is uploaded to an isolated processing worker using pinned scikit-learn 1.7.2. Up to 200,000 characters across all passages, 8,000 characters per passage, 80-character unique IDs and 10,000 retained vocabulary features. Projects must fit within 500,000 encoded JSON characters; imports are UTF-8 files up to 2 MB. Processing stops after two minutes. Words use Unicode letters, digits and underscores; character fragments use 3–5 characters inside whitespace-separated words with boundary padding. No stemming, translation or semantic model is used. The vocabulary cap retains the most frequent features. English stop words are optional and match only lower-case forms when case matching is enabled. Nothing is automatically saved. JSON preserves exact source text; CSV prefixes formula-like strings and spreadsheets can reinterpret numeric-looking text. Training requires 2–20 labels. Test labels must appear in training; repeated example text is rejected after Unicode, case and whitespace normalization. Related paraphrases can still bias the evaluation. No automatic split or cross-validation is performed. Scores are uncalibrated model outputs, not verified probabilities. Text matching no training features remains unmatched. Logistic regression uses C=1 and at most 500 iterations. Projects preserve source and settings; models are rebuilt rather than exported as executable files.