Seldon Demo

Seldon Demo

Classifies rows of a data table with the Seldon-Mini model and compares it against an AutoGluon baseline on the same split. Use a sample table or your own CSV. Suits data scientists and ML engineers.

Made by Neuralk-AI on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space lets you try Seldon-Mini, a model for tabular classification. You pick a sample table or upload a CSV, choose the column to predict, and the model labels a held-out test split. An AutoGluon baseline runs on the same split, so you can compare the two side by side.

Who it suits: It suits data scientists and ML engineers who want to see how a tabular model behaves on their own data before adopting it.

Industry | Who uses it | How they use it | Example
Finance | Data scientists | Test loan and credit tables to see how the model sorts good from bad | Run the credit-g sample table
Healthcare and nonprofits | Analysts | Try a donor table with few features | Run the blood-transfusion sample table
Retail and telecom | Analysts | Check churn prediction on customer rows | Run the churn sample table
Machine learning | ML engineers | Compare a tabular model with a standard baseline | Read ROC AUC and accuracy for both models

You can skip it if you need a hosted service for production predictions, since this is a demo that recomputes every run.

Where to use it:

• A web browser: open the Space page on Hugging Face
• Windows, macOS or Linux: run a copy with Python 3.12
• A computer without a GPU also works locally, with the model running on the CPU

Use case | Who | What they get back
Classify a sample table | Anyone curious | Scores and a prediction for each test row
Classify your CSV | Data scientists | The same results for a table you upload
Compare with a baseline | ML engineers | Both models' ROC AUC, accuracy, timing and confusion matrices

How to set it up: The online page needs nothing. A local copy needs more, because the model is gated.

• A web browser for the online demo
• Python 3.12 for a local copy
• A Hugging Face account that has accepted the gate of the Seldon-Mini model
• The packages listed in requirements.txt

1. To use it online, open the Space page on Hugging Face.
2. To run a copy, install the Hugging Face hub package.
3. Download the Space into a local folder.
4. Move into the folder and install Gradio 6.27.0 and the listed packages.
5. Start the app. The README also shows a run with uv, after signing in once with the account that accepted the gate.

pip install -U huggingface_hub
hf download Neuralk-AI/seldon_demo --repo-type space --local-dir seldon_demo
cd seldon_demo
pip install gradio==6.27.0
pip install -r requirements.txt
python app.py

hf auth login  # once, with an account that has accepted the gate of Neuralk-AI/seldon-mini
uv run --python 3.12 --with gradio==6.27.0 --with-requirements requirements.txt python app.py

How to use it: Open the page, pick one of the five sample tables or upload a CSV, and choose the target column. Press the single Run button, and both models run at the same time. Each model's card shows its stages as they finish, and the page ends with the scores and a preview of the data.

Example: You pick the credit-g table and the target column named class, then press Run. The Seldon-Mini card reads the training rows and predicts in one pass. The AutoGluon card fits its tree models one by one. You get ROC AUC, accuracy, timing, confusion matrices and a prediction for each test row.

For newcomers and experts: If you are new to this, start with a sample table and leave everything else alone. Press Run once and compare the two cards. A higher ROC AUC and accuracy mean the model sorted the test rows better.

If you know the field, Seldon-Mini reads the training rows as context in a Fit step and predicts in one pass with no tuning. The baseline is AutoGluon at its medium quality preset on the CPU, capped at 60 seconds, which trains tree models one after another and blends them into an ensemble. Seldon-Mini's stages flip to done together once its GPU call finishes.

Pick a table or upload a CSV
Choose the target column
The data is split into train and test rows
Seldon-Mini and AutoGluon run side by side
Scores, matrices and predictions appear

Good to know: Nothing is cached or shared between visitors, so every run computes both models again. The five sample tables come from OpenML, and the chart shows how many rows each one has.

Title: Rows in each bundled table
credit-g: 1000 rows
blood-transfusion: 748 rows
credit-approval: 690 rows
vehicle: 846 rows
churn: 5000 rows

• License: MIT
• Cost to run: a Hugging Face account that accepted the model gate and its read token, plus ZeroGPU hardware for the hosted Space
• Data it reaches: the CSV you upload, the bundled OpenML tables, and the model weights loaded from Hugging Face
• Not for: reusing results across visitors, since nothing is cached

On Hugging Face, 13 November 2026

• Likes: 2
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/Neuralk-AI/seldon_demo
• Files: https://huggingface.co/spaces/Neuralk-AI/seldon_demo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.