Machine Learning Studies

This repository organizes my Machine Learning studies with a focus on practical applications. The goal is to learn how to solve real-world problems efficiently, prioritizing what produces the most impact.

Study Structure

📌 ML + DL Study Roadmap

1. Data Understanding & Preparation ✅

• Data exploration using pandas, matplotlib, and seaborn X
• Cleaning: handling missing values, duplicates, and outliers X
• Normalization and standardization X
• Encoding categorical variables (one-hot, label encoding)
• Train/validation/test split X

2. Core ML Models ✅

• Logistic Regression — simple classification X
• Random Forest — robust, good for tabular data X
• XGBoost / LightGBM — strong performance on structured data X
• 👉 Compare baselines vs ensembles, using cross-validation

3. Model Evaluation ✅

• Metrics: accuracy, precision, recall, F1-score, ROC-AUC X
• Choose metrics according to the problem type X

4. Avoiding Overfitting ✅

• Techniques: regularization, cross-validation, early stopping
• Focus: train well and generalize to unseen data

5. Hyperparameter Tuning

• Tools: GridSearchCV, RandomizedSearchCV, Optuna
• Apply to RandomForest / XGBoost / LightGBM ✅

---

6. Intro to Deep Learning (Concepts First)

• Neurons, layers, activations, loss functions, optimizers ✅
• Backpropagation (high-level understanding) ✅
• Overfitting/underfitting in neural nets ✅

---

7. TensorFlow & PyTorch 🚀

• TensorFlow (Keras API)
  • Building dense feedforward networks ✅
  • Using callbacks (early stopping, checkpointing) ✅
  • Training on structured/tabular data ✅
• PyTorch
  • Manual training loops vs high-level API (Lightning/FastAI)
  • Understanding tensors & autograd
  • Implementing a basic neural net from scratch
    👉 Project: train the same simple NN in both frameworks, compare coding style

---

8. Text & Basic NLP

• Vectorization: TF-IDF, word embeddings, BERT embeddings
• Classification tasks (reviews, spam detection)
• Try both scikit-learn + TF-IDF and PyTorch/TensorFlow embeddings

---

9. Practical ML/DL Cycle

1. Collect and prepare data
2. Split into train/validation/test sets
3. Train ML or DL models
4. Evaluate and adjust
5. Deploy (pickle/joblib for ML, SavedModel/torchscript for DL)
6. Serve via FastAPI/Flask

---

10. What to Avoid Initially

• Very deep neural nets (CNNs/RNNs) without tabular basics
• Reinforcement learning, GANs, diffusion models (too advanced for now)
• MLOps heavy tools (Kubeflow, Airflow)

---

📂 Next Repo Section:
/deep_learning_basics → notebooks comparing TensorFlow and PyTorch on the same problems.

Getting Started

Clone the repository and follow the notebooks and scripts in the /notebooks folder for exercises and mini-projects.