🧠 AI Training Assistant

A full-stack machine learning workbench that runs entirely on your local machine. Upload a CSV, explore and clean your data, configure a classical ML model or a custom deep neural network, train it with real-time progress streaming, and export the results — all through a browser UI backed by a Python/Flask API.

---

✨ Features

📂 Data Loading

• Upload any CSV file (chunked reading — handles large files)
• Load built-in sklearn datasets: Iris, Wine, Diabetes, Breast Cancer
• Automatic column type detection (numeric / categorical / text)

🔍 Data Explorer

• Paginated table view with sort and filter on any column
• Per-column statistics (min, max, mean, std, median, quartiles, top values)
• Inline column operations: rename, drop, change type, set as target
• Row operations: delete selected rows, remove duplicates
• Full dataset reset to original uploaded state

⚙️ Preprocessing Pipeline

Option	Choices
Missing value strategy	Mean, Median, Mode, Zero fill, Drop rows, KNN Imputation
Categorical encoding	One-Hot, Label Encoding
Feature scaling	StandardScaler, MinMaxScaler, RobustScaler, MaxAbsScaler
Outlier removal	IQR-based automatic removal
Duplicate removal	✓
Feature selection	SelectKBest (top-K)
Dimensionality reduction	PCA (configurable variance threshold)

🤖 Classical ML (17 algorithms)

Category	Algorithms
Classification	Random Forest, Gradient Boosting, XGBoost, SVM, KNN, Logistic Regression, Decision Tree, Naive Bayes, AdaBoost
Regression	Linear Regression, Ridge, Lasso, ElasticNet, SVR, Random Forest, Gradient Boosting, KNN
Clustering	K-Means, DBSCAN
Anomaly Detection	Isolation Forest

🧬 Deep Learning (Keras/TensorFlow)

• Visual layer builder — add/remove Dense layers with per-layer neurons, activation, dropout
• Quick architecture presets: Tiny, Small, Medium, Large, XLarge
• 10 activations: ReLU, GELU, Swish, Sigmoid, Tanh, LeakyReLU, ELU, Softmax, Linear, PReLU
• 7 optimizers: Adam, AdamW, SGD, RMSprop, Adagrad, Nadam, Adadelta
• 6 weight initializers: Xavier/Glorot, He Normal, He Uniform, LeCun Normal, Orthogonal, Random Normal
• LR schedulers: ReduceOnPlateau, CosineAnnealing, StepDecay, CyclicLR, WarmupLinear
• Early stopping with configurable patience
• L2 regularization, bias toggle, configurable batch size

📊 Real-Time Training

• Live loss/accuracy curves streamed epoch-by-epoch via SSE (Server-Sent Events)
• Staged progress for Gradient Boosting (per-estimator updates)
• Final metrics panel: Accuracy, F1, Precision, Recall, R², MAE, RMSE, Silhouette Score
• Confusion matrix + full classification report

💾 Exports

• Trained model file (.pkl for classical, .h5 for Keras)
• Training history as CSV
• Processed/cleaned dataset as CSV

---

🗂️ Project Structure

AI Training Assistant/
├── backend.py              # Flask API server
├── requirements.txt        # Python dependencies
├── ml_uploads/             # Uploaded CSV files (auto-created)
├── ml_models/              # Saved trained models (auto-created)
└── frontend/               # Vite + React app
    ├── src/
    │   ├── App.jsx         # Main frontend (full UI)
    │   └── main.jsx        # React entry point
    ├── package.json
    └── vite.config.js

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+
• npm 9+

---

1. Clone the repository

git clone https://github.com/your-username/ai-training-assistant.git
cd ai-training-assistant

---

2. Set up the Python backend

# Create and activate a virtual environment
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS / Linux
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

---

3. Set up the React frontend

cd frontend
npm install

---

4. Run the app

Open two terminals:

Terminal 1 — Backend (from the project root):

# Windows
.venv\Scripts\python.exe backend.py

# macOS / Linux
python backend.py

You should see:

=======================================================
  AI Training Assistant — Backend
=======================================================
  TensorFlow : ✓  2.21.0
  XGBoost    : ✓
  Uploads    : /path/to/ml_uploads
  Models     : /path/to/ml_models
=======================================================
  Running at http://localhost:5000

Terminal 2 — Frontend (from the frontend/ folder):

cd frontend
npm run dev

Then open http://localhost:5173 in your browser.

---

🖥️ How to Use

Step 1 — Load Data

• Click Upload CSV to load your own dataset, or
• Click one of the built-in sample datasets (Iris, Wine, Diabetes, Breast Cancer)

Step 2 — Explore & Clean

• Browse your data in the Data Explorer table
• Click any column header to sort; use the filter bar to search
• Right-click a column name to rename, drop, change its type, or view statistics
• Select rows and delete them, or use Remove Duplicates

Step 3 — Set Target Column

• In the column list, click the column you want to predict and mark it as Target

Step 4 — Preprocess

• Go to the Preprocess tab
• Choose your missing value strategy, encoding, scaling, and any optional steps (outlier removal, feature selection, PCA)
• Click Apply Preprocessing — a log shows exactly what was done

Step 5 — Configure Model

• Choose Problem Type (Classification, Regression, Clustering, etc.)
• Toggle between Classical ML and Deep Learning
  • Classical: pick an algorithm from the dropdown
  • Deep Learning: build your layer stack visually or choose a preset
• Set hyperparameters (epochs, learning rate, batch size, optimizer, scheduler, etc.)

Step 6 — Train

• Click ▶ Start Real Training
• Watch the loss/accuracy curves update in real time
• When training completes, metrics are shown automatically

Step 7 — Export

• Download the trained model (.pkl or .h5)
• Download training history as CSV
• Download the processed dataset as CSV

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/api/upload	Upload a CSV file
GET	/api/sample/<name>	Load a built-in sklearn dataset
GET	/api/dataset/preview	Paginated table data
GET	/api/dataset/stats/<col>	Column statistics
POST	/api/column/drop	Drop a column
POST	/api/column/rename	Rename a column
POST	/api/column/target	Set target column
POST	/api/column/type	Change column dtype
POST	/api/column/fillnull	Fill nulls in a column
POST	/api/rows/delete	Delete rows by index
POST	/api/rows/dedup	Remove duplicate rows
POST	/api/dataset/reset	Reset to original data
POST	/api/preprocess	Run preprocessing pipeline
POST	/api/train/stream	Start training (SSE stream)
GET	/api/model/export	Download trained model
GET	/api/history/csv	Download training history
GET	/api/dataset/export	Download processed dataset
GET	/api/health	Backend health check
GET	/api/session/info	Current session metadata

---

⚡ Tech Stack

Layer	Technology
Frontend	React 19, Vite, Recharts
Backend	Python, Flask, Flask-CORS
ML — Classical	scikit-learn 1.8, XGBoost 3.2
ML — Deep Learning	TensorFlow 2.21 / Keras
Data	pandas, NumPy
Streaming	Server-Sent Events (SSE)

---

🛠️ Troubleshooting

Backend shows 404 on / This is normal. The backend has no homepage — all routes are under /api/. Visit http://localhost:5000/api/health to confirm it's running.

recharts not found error in frontend

cd frontend
npm install recharts

TensorFlow not detected Deep learning is disabled but classical ML still works. To enable TF:

pip install tensorflow==2.21.0

Port already in use Change the backend port at the bottom of backend.py:

app.run(debug=False, host="0.0.0.0", port=5001, threaded=True)

And update frontend/src/App.jsx line 8:

const API = "http://localhost:5001/api";

---

📄 License

MIT License — free to use, modify, and distribute.

---

🙋 Contributing

Pull requests are welcome. For major changes, open an issue first to discuss what you'd like to change.