https://github.com/264gaurav/deep-learning

Deep Learning and Neural Network learning/building and exploring
https://github.com/264gaurav/deep-learning

artificial-neural-networks dagshub deep-learning deep-neural-networks dvc experiment-tracking hyperparameter-tuning keras-tensorflow keras-tuner mlflow-tracking numpy sklearn tensorflow testing training-data

Last synced: 4 months ago
JSON representation

Deep Learning and Neural Network learning/building and exploring

• Host: GitHub
• URL: https://github.com/264gaurav/deep-learning
• Owner: 264Gaurav
• Created: 2025-08-31T12:12:47.000Z (5 months ago)
• Default Branch: main
• Last Pushed: 2025-09-20T21:57:06.000Z (4 months ago)
• Last Synced: 2025-09-20T23:35:38.591Z (4 months ago)
• Topics: artificial-neural-networks, dagshub, deep-learning, deep-neural-networks, dvc, experiment-tracking, hyperparameter-tuning, keras-tensorflow, keras-tuner, mlflow-tracking, numpy, sklearn, tensorflow, testing, training-data
• Language: Jupyter Notebook
• Homepage:
• Size: 506 KB
• Stars: 1
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: Readme.md

Awesome Lists containing this project

README

          
# Deep Learning: Underfitting, Overfitting & Regularization

## 📌 Core Libraries

- **NumPy** → Arrays & matrix operations

- **Matplotlib** → Visualization

- **Pandas** → Data handling (CSV, DataFrames)

- **TensorFlow & Keras** → Neural networks

  - **Sequential** → Linear stack of layers

  - **Dense** → Fully connected layer

  - **Dropout** → Reduces overfitting

---

## ⚖️ Underfitting vs Overfitting

### 🔹 Underfitting

Model too simple → poor performance on train & test.

**Fix:** Add layers/neurons, train longer, lower LR, reduce regularization.

### 🔹 Overfitting

Model too complex → memorizes training, fails on test.

**Fix:** Dropout, L2 regularization, early stopping, data augmentation, reduce complexity.

---

## 📊 Comparison

| Feature           | Underfitting 🟡 | Good Fit 🟢 | Overfitting 🔴 |

| ----------------- | --------------- | ----------- | -------------- |

| Training Accuracy | Low             | High        | Very High      |

| Test Accuracy     | Low             | High        | Low            |

| Model Complexity  | Too simple      | Balanced    | Too complex    |

| Generalization    | Poor            | Good        | Poor           |

---

## 🎯 Dropout

- Randomly drops neurons (prob _p_) during training.

- Prevents reliance on specific neurons → robust features.

- **Training:** scale surviving neurons `1/(1-p)`

- **Testing:** dropout off, scaling handled automatically.

- Best for large models, avoid overuse on small datasets.

---

## 🛠️ Ways to Reduce Overfitting

- Dropout

- L1/L2 Regularization

- Early Stopping

- Data Augmentation

- Batch Normalization

- Reduce Model Complexity

- Cross-Validation

---

## ✅ Rule of Thumb

- **Underfitting:** Increase complexity, train longer.

- **Overfitting:** Add regularization, dropout, early stopping, augmentation.

📌 **Goal:** Achieve balance → model should **generalize well** to unseen data.

# Keras Tuner - HyperParameter Tuning of models:

### For understanding of KerasTuner, refer to the [KerasTuner Readme](./2-kerasTuner/Readme.md).

# Embeddings with keras and tensoflow :

### For understanding of Embeddings, refer to the [Embeddings Readme](./5-WordEmbeddings/Readme.md).