Apppublic

dualdatarate/ipoFastApi

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes

App README

IPO Classifier

ML-based cluster classification for Indian Mainboard IPOs — a data-driven alternative to Gray Market Premium (GMP) speculation.

Instead of relying on GMP (which is unregulated and easily manipulated), this matches an upcoming IPO against historical cluster profiles built from 6 years of Mainboard IPO data.

How It Works

Input fields (subscription data, reservations, recommendations, issue size)
        ↓
Pydantic validation + computed features
        ↓
StandardScaler → UMAP (dimensionality reduction) → K-Means clustering
        ↓
Cluster profile lookup → historical outcome distribution + avg listing gain

The model was trained on 100+ scraped fields, reduced to ~30 high-correlation features, across 6+ years of Mainboard IPO history.

Cluster Profiles

Hit GET / on the running API to see the current cluster breakdown:

json

{
  "0": { "Stonks": "61.5%", "Hmm": "23.1%", "Meh": "7.7%", "Loss": "7.7%", "Avg Gain": "34.2%", "Sample Size": 26 },
  "1": { ... }
}

Outcome buckets: Loss (<0%), Meh (0–10%), Hmm (10–25%), Stonks (>25% listing gain).

Project Structure

├── api.py               # FastAPI app — POST /test to classify, GET / for cluster profiles
├── model.py             # Pydantic input schema (EntryStruct) with computed fields
├── preprocess_data.py   # StandardScaler transform → UMAP → cluster prediction
├── frontend.py          # Streamlit UI for entering IPO details
├── frontHome.py         # Streamlit homepage showing cluster profiles
├── get_schema.json      # JSON schema used to generate model.py fields
├── models.pkl           # Serialised pipeline: K-Means + UMAP + scaler (not in repo — generate below)
└── ipo_history.pkl      # Training data — 6 years of scraped Mainboard IPO history (not in repo)

Setup

bash

git clone https://github.com/sunny9n9/MainboardIpoClustering
cd MainboardIpoClustering
pip install -r requirements.txt

models.pkl and ipo_history.pkl are not committed (size). To regenerate the model, run your training notebook/script that produces these two files, then place them in the project root.

Run the API:

bash

uvicorn api:app --reload

Run the frontend:

bash

streamlit run frontHome.py

Tech Stack

—FastAPI — REST API
—Streamlit — frontend
—scikit-learn — K-Means, StandardScaler
—umap-learn — dimensionality reduction
—Pydantic v2 — input validation and computed fields
—pandas / numpy — data handling