Overview

This repository collects my TensorTonic solutions for core machine learning and NumPy-based exercises.

The goal is not just to store answers. It is to build a compact learning archive where each problem includes:

• a working solution.py
• a problem-focused README.md
• a short explanation.md
• quick-reference notes
• test case examples for revision

If you are learning ML fundamentals from first principles, this repo is designed to feel more like a study lab than a plain answer dump.

---

Learning Flow

Math Basics -> NumPy Operations -> Optimization -> Sequence Models -> Transformer Foundations

Current solved sequence in this repo:

1. TT-001 Sigmoid
2. TT-002 Mean Squared Error
3. TT-003 Softmax
4. TT-004 Cross Entropy
5. TT-005 Gradient Descent
6. TT-006 Adam
7. TT-007 Matrix Transposition
8. TT-008 Dot Product
9. TT-009 Linear Regression Closed Form
10. TT-010 Batch Normalization Forward
11. TT-011 RNN Step Forward
12. TT-012 Euclidean Distance
13. TT-013 Tokenization
14. TT-014 Embedding
15. TT-015 Positional Encoding
16. TT-016 Scaled Dot Product Attention
17. TT-017 Multi-Head Attention
18. TT-018 FeedForward Network
19. TT-019 Layer Normalization
20. TT-020 Encoder Block
21. TT-021 Logistic Regression
22. TT-022 KNN Distance
23. TT-023 K-Means Clustering
24. TT-024 PCA Projection
25. TT-025 Naive Bayes Log Likelihood (Bernoulli)

---

Repository Map

Tensortonic Solutions/
|
+-- 00-math-and-numpy-basics/
|   |
|   +-- TT-001-sigmoid-numpy/
|   +-- TT-007-Matrix_Transposition/
|   +-- TT-008-Dot-Product/
|   +-- TT-012-Euclidean-Distance/
|
+-- 01-core-ml/
|   +-- TT-009-Linear-Regression-Closed-Form/
|   +-- TT-021-Logistic-Regression/
|   +-- TT-022-KNN-Distance/
|   +-- TT-023-K-Means-Clustering/
|   +-- TT-024-pca-projection/
|   +-- TT-025-Naive-Bayes-Log-Likelihood-(Bernoulli)/
|
+-- 02-optimization/
|   +-- TT-005-Gradien-Descent/
|   +-- TT-006-Adam/
|
+-- 03-losses-and-activations/
|   +-- TT-002-MSE-numpy/
|   +-- TT-003-Softmax/
|   +-- TT-004-Cross-Entropy/
|
+-- 04-neural-networks/
|   +-- TT-010-Batch-Normalization/
|
+-- 05-sequence-models/
|   +-- TT-011-RNN-Step-Forward/
|
+-- 06-attention-and-transformers/
|   +-- TT-013-Tokenization/
|   +-- TT-014-Embedding/
|   +-- TT-015-Positional-Encoding/
|   +-- TT-016-Scaled-Dot-Product/
|   +-- TT-017-MultiHead-Attention/
|   +-- TT-018-FeedForward-Network/
|   +-- TT-019-LayerNorm/
|   +-- TT-020-Encoder-Block/
|
+-- 07-computer-vision/
+-- 08-generative-models/
+-- 09-research-paper-implementations/
|
+-- templates/
+-- roadmap.md

---

Solved Problems

ID	Problem	Concepts	Folder
TT-001	Sigmoid	activation function, NumPy basics	00-math-and-numpy-basics/TT-001-sigmoid-numpy
TT-002	MSE	regression loss, averaging, arrays	03-losses-and-activations/TT-002-MSE-numpy
TT-003	Softmax	normalization, stability, exponentials	03-losses-and-activations/TT-003-Softmax
TT-004	Cross Entropy	classification loss, log probabilities	03-losses-and-activations/TT-004-Cross-Entropy
TT-005	Gradient Descent	derivatives, updates, optimization	02-optimization/TT-005-Gradien-Descent
TT-006	Adam	momentum, adaptive learning rates, bias correction	02-optimization/TT-006-Adam
TT-007	Matrix Transposition	matrix operations, axes, NumPy	00-math-and-numpy-basics/TT-007-Matrix_Transposition
TT-008	Dot Product	linear algebra, vector multiplication	00-math-and-numpy-basics/TT-008-Dot-Product
TT-009	Linear Regression Closed Form	normal equation, regression, matrix inversion	01-core-ml/TT-009-Linear-Regression-Closed-Form
TT-010	Batch Normalization Forward	normalization, feature scaling, neural nets	04-neural-networks/TT-010-Batch-Normalization
TT-011	RNN Step Forward	recurrent networks, hidden state update	05-sequence-models/TT-011-RNN-Step-Forward
TT-012	Euclidean Distance	vector distance, L2 norm	00-math-and-numpy-basics/TT-012-Euclidean-Distance
TT-013	Tokenization	NLP basics, vocabulary mapping, encoding	06-attention-and-transformers/TT-013-Tokenization
TT-014	Embedding	token vectors, PyTorch, transformer basics	06-attention-and-transformers/TT-014-Embedding
TT-015	Positional Encoding	sinusoidal positions, sequence order, transformers	06-attention-and-transformers/TT-015-Positional-Encoding
TT-016	Scaled Dot Product Attention	attention scores, softmax, weighted values	06-attention-and-transformers/TT-016-Scaled-Dot-Product
TT-017	Multi-Head Attention	attention heads, projections, sequence mixing	06-attention-and-transformers/TT-017-MultiHead-Attention
TT-018	FeedForward Network	MLP, ReLU, transformer sublayers	06-attention-and-transformers/TT-018-FeedForward-Network
TT-019	LayerNorm	feature normalization, gamma, beta, stability	06-attention-and-transformers/TT-019-LayerNorm
TT-020	Encoder Block	self-attention, residuals, layer norm, FFN	06-attention-and-transformers/TT-020-Encoder-Block
TT-021	Logistic Regression	binary classification, sigmoid, gradient descent	01-core-ml/TT-021-Logistic-Regression
TT-022	KNN Distance	nearest neighbors, Euclidean distance, indexing	01-core-ml/TT-022-KNN-Distance
TT-023	K-Means Clustering	clustering, centroid assignment, squared distance	01-core-ml/TT-023-K-Means-Clustering
TT-024	PCA Projection	dimensionality reduction, covariance, eigenvectors	01-core-ml/TT-024-pca-projection
TT-025	Naive Bayes Log Likelihood (Bernoulli)	binary features, Laplace smoothing, log probabilities	01-core-ml/TT-025-Naive-Bayes-Log-Likelihood-(Bernoulli)

---

What Makes This Repo Useful

Readable Solutions Problems are solved with focused Python implementations instead of bloated notebooks or framework-heavy code.

Revision Friendly Each folder is structured so a concept can be reviewed quickly before interviews, classes, or practice sessions.

Concept First The repository emphasizes intuition: what the formula means, why the code works, and where the result is used in ML.

---

Example Folder Structure

Inside a typical problem directory, you will usually find:

TT-xxx-problem-name/
|
+-- solution.py
+-- README.md
+-- explanation.md
+-- note.md / notes.md
+-- test_cases.md

This makes each exercise self-contained and easy to revisit later.

---

Tech Stack

{
    "language": "Python",
    "numerical_computing": ["NumPy", "PyTorch"],
    "focus": [
        "Machine Learning fundamentals",
        "Mathematical intuition",
        "Optimization basics",
        "Vectorized computation",
        "Sequence modeling foundations",
        "Transformer building blocks"
    ]
}

---

Why TensorTonic

TensorTonic is a great environment for practicing the mechanics behind machine learning systems.

Instead of hiding everything behind high-level libraries, these exercises push you to implement core ideas yourself:

• activation functions
• loss functions
• probability transformations
• optimization steps
• array-based reasoning
• sequence modeling basics
• transformer input representations

That is where a lot of real intuition gets built.

---

Roadmap

The broader learning direction for this repo lives in roadmap.md.

The intention is to keep expanding this repository in a clean progression:

• stronger math and NumPy foundations
• classical ML building blocks
• deeper optimization concepts
• neural network components
• more complete end-to-end implementations

---

Quick Start

git clone https://github.com/mutassimalzeem/TensorTonic-Solutions.git
cd "Tensortonic Solutions"

Then open any exercise folder and run or inspect solution.py.

---

Final Note

This project is part practice set, part knowledge base, and part progress tracker.

If you are also learning machine learning from scratch, I hope this repo gives you a clean and motivating example of how to document your journey while improving your implementation skills.