Applied Data Science Course 🚀

This repository contains my notes, practices, and solutions for the Applied Data Science Course offered by Sharif University of Technology (EE-879). The course has also been presented — and is currently being presented — at Teias, Khatam University. The content covers various data science concepts, hands-on exercises, and solutions to course assignments.

---

Course Information

For more details about the course, visit the official course website: Applied Data Science - Sharif University of Technology, Spring 2025 or Saloot GitHub

Course Topics

The course covers the following key topics:

1. 📚 Introduction to Pandas
2. 🧹 Data Cleaning and Preprocessing
3. 📊 Data Visualization
4. ⚙️ Feature Engineering and Dimensionality Reduction
5. 🎯 Different Problem Types and Accuracy Measures
6. 📈 Regression Methods
7. 🔍 Classification Methods
8. 🌂 Multiclass/Multilabel Classification and Boosting
9. 🧠 Neural Networks
10. 🚀 Deep Learning
11. 🖼️ Deep Learning Application: Image Classification
12. 🤖 Generative AI

Datasets

The following datasets are utilized in this repository:

• 🚗 Car Features and MSRP
• 🧠 Stroke Prediction
• 🏠 Ames Iowa Housing Data
• 🏨 Hotel Booking Demand
• 🥑 Avocado Prices

Assignments

Assignment 1: Introduction to Pandas 🐼📊

• Complete Kaggle mini tutorial on Pandas: https://www.kaggle.com/learn/pandas
• Choose a dataset for future assignments.

---

Assignment 2: Exploratory Analysis and Data Cleaning 🔍🧹

• Perform EDA, cleaning, and preprocessing on the Stroke Prediction dataset: https://www.kaggle.com/datasets/fedesoriano/stroke-prediction-dataset
• Complete Kaggle mini tutorial on Data Cleaning: https://www.kaggle.com/learn/data-cleaning

---

Assignment 3: Data Visualization + Web Scraping 📈🎨🕸️

• Data visualization on the Hotel Booking Demand dataset: https://www.kaggle.com/datasets/jessemostipak/hotel-booking-demand/data
• Kaggle mini tutorial on Data Visualization: https://www.kaggle.com/learn/data-visualization
• Web-scrape 50 Samand cars from bama.ir (price, mileage, color, year, transmission, description)

---

Assignment 4: Feature Engineering 🧱⚙️

• Create new features (ratio, binning, functional transforms, combining columns)
• Create date/time features
• Perform aggregation and counts
• Feature selection using Mutual Information
• PCA for dimensionality reduction
• Write explanation: When is feature engineering optional vs necessary?
• Optional: Kaggle mini tutorial on Feature Engineering

---

Assignment 5: Accuracy Measures 🎯📏

• Compute regression metrics: MSE, MAE, MAPE, R2
• Compute binary classification metrics: Precision, Recall, F1
• Compute multi-class metrics: per-class Precision and Recall, macro/micro/weighted F1
• Answer question about multi-label metrics (football example)
• Use models or random predictions for metric evaluation

---

Assignment 6: Regression Methods 📉🤖

• Train and evaluate:
  • Linear Regression
  • Kernel Regression
  • Logistic Regression
  • Ridge Regression
  • LASSO Regression
• Explain the kernel trick

---

Assignment 7: Binary Classification Methods ⚡📘

• Train and evaluate:
  • Logistic Regression
  • SVM (linear + kernel)
  • KNN (implement + tune K)
  • Decision Trees (implement + tune depth)
  • Random Forest
• Explain three regularization techniques for decision trees
• Bonus: Achieve F1 > 0.9 on diabetes dataset

---

Assignment 8: Multiclass Classification Methods 🎨🔢

• Implement and evaluate:
  • Multiclass SVM
  • Multiclass Logistic Regression (OVR, multinomial, log loss)
  • Multiclass KNN (implement + tune K)
  • Multiclass Decision Trees
  • Boosting methods: XGBoost, LightGBM, Adaboost/Catboost
• Perform grid search on one boosting method
• Explain how KNN and Decision Trees extend to multi-label classification
• Bonus: Achieve F1 > 0.6 on 12-class player-position dataset

---

Assignment 9: Neural Networks 🧠💡

• Train and evaluate:
  • Scikit-Learn MLP (classification + regression)
  • Keras 4-layer sequential network (classification + regression)
  • PyTorch 4-layer network (classification + regression)
  • Keras 4-layer non-sequential network
• Explain why neural networks are powerful and difficult to design
• Bonus: RNN (3-layer) if dataset includes time-series

---

Assignment 10: Deep Neural Network Tuning 🚀🔧

• Tune Keras network using at least 5 options:
  • Optimizer
  • Learning rate
  • Learning rate decay
  • Batch size
  • Activation functions
  • Weight initialization
  • Network depth and width
  • L1/L2 weight regularization
  • L1/L2 activity regularization
  • Dropout rate
• Explain why deeper networks are harder to train
• Use 4-fold cross validation for evaluation

---

Assignment 11: CNNs, Transfer Learning, and Data Augmentation 🖼️📦🤝

• Build CNN with at least two convolutional layers
• Tune:
  • Convolution kernel size
  • Convolution stride
  • Pooling size
  • Pooling stride
• Perform data augmentation using ImageGenerator
• Perform transfer learning using two pretrained models (VGG19, ResNet, EfficientNet)
• Explain impact of receptive field size on performance

---

Assignment 12: Autoencoders and Generative Models 🎛️✨🧬

• Build and train:
  • Dense autoencoder
  • Convolutional autoencoder
  • Denoising autoencoder
• Train a GAN on CIFAR-10 dataset
• Use OpenAI API to generate:
  • An image
  • An audio voice reading generated text
• Explain the adversarial learning process
• Bonus: Build a Variational Autoencoder on Fashion-MNIST

Practices

• Practice 1: Explore pandas features on Car Features and MSRP dataset.

Maybe I will practice more in the future :).

Repository Structure

• Notes → Summarized lecture notes & key concepts
• Practices → Hands-on coding exercises & projects
• Assignments → Solutions to course assignments

---

Project

• Graduate Admission Prediction
• CNN and NST from Scratch