Deep-ML Practice

Daily machine learning problem solving focused on understanding fundamentals, not memorizing solutions.

Philosophy

"If you can't implement it from scratch, you don't truly understand it."

This repository tracks my journey through Deep-ML.com problems. Each solution prioritizes:

1. Understanding over speed - Break down the problem before coding
2. ML best practices - Use zip(), comprehensions, vectorized operations
3. Two implementations - Pure Python first, then NumPy for production

Problems Solved

Problem	Core Concept	Pattern
Matrix-Vector Multiply	Dot products	sum(r*v for r,v in zip(row, vec))
Transpose	Column extraction	zip(*matrix)
Reshape	Flatten then chunk	flat[i:i+cols] with range(0, len, step)
Row/Column Means	Aggregation	zip(*matrix) for columns
Scalar Multiply	Element-wise ops	Nested comprehension
Matrix Transform T⁻¹AS	Change of basis	Invertibility check + chain multiply
Eigenvalues	Spectral decomposition	np.linalg.eig()
2x2 Inverse	Linear algebra	Swap diagonal, negate off-diagonal, divide by det

Key Patterns

Pure Python

# Dot product
sum(a * b for a, b in zip(vec1, vec2))

# Matrix-vector multiply
[sum(r * v for r, v in zip(row, vector)) for row in matrix]

# Transpose
[list(col) for col in zip(*matrix)]

# Flatten
[elem for row in matrix for elem in row]

# Reshape (flatten then chunk)
[flat[i:i+cols] for i in range(0, len(flat), cols)]

NumPy Sandwich Pattern

def operation(A, ...):
    # 1. Convert to NumPy
    A_array = np.array(A, dtype=float)

    # 2. Do operations (fast, clean)
    result = A_array @ B_array  # or .T, .reshape(), etc.

    # 3. Convert back to list
    return result.tolist()

What I Learned

On zip(): Pairs elements position-by-position. With * unpacking, zip(*matrix) extracts columns.

On comprehensions: Read outer-to-inner. [f(x) for row in matrix for x in row] means "for each row, for each element in that row."

On reshape: Always flatten first, then chunk. The slice [i:i+width] cuts a ribbon of exactly width elements.

On invertibility: A matrix is invertible iff det != 0. If det = 0, the matrix "crushes" dimensions and information is lost.

Structure

deep-ml/
├── practice.ipynb    # All solutions with explanations
└── README.md         # This file

Running

# Activate Jupyter environment
source ~/jupyter-env/bin/activate

# Start JupyterLab
jupyter lab practice.ipynb

Resources

• Deep-ML.com - Daily ML problems
• Karpathy's Zero to Hero - Neural networks from scratch

Author

Felix Onyango (@Jaloch-glitch)