Learn Artificial Neural Network From Scratch in Python

download and install the PyCharm community edition on Windows, create a demo project with a virtual environment, configure a Python 3.8 interpreter, and install packages like pygame to run code.

Install Visual Studio Code on Windows, add the Python extension, and run Python code with the built-in terminal and run button; note optional editors like PyCharm.

Start by downloading and installing the PyCharm community edition on Linux, set up a Python project and interpreter, and install libraries like pandas and Beautiful Soup.

Learn to use the Python official documentation to explore modules, including the math module, and apply documented functions with help from the Python community.

Learn to create and assign variables in python using the assignment operator, storing integers, floats, strings, and booleans, with naming rules and keyword restrictions.

Explore Python data types—strings, sets, and numbers—and how dynamic typing allocates memory at runtime. Grasp string immutability, indexing and slicing, and numeric types: integer, float, and complex.

Explore Python data types by learning lists, dictionaries, and tuples, including creation, indexing, slicing, and accessing elements to build complex structures.

Master Python operators and operands, including true and floor division, addition, subtraction, multiplication, exponentiation, and multiple assignment, while using operator and math modules for sqrt and trig.

Explore boolean operators in Python, mastering and, or, not, truth tables, and operator precedence with parentheses and arithmetic rules to build logical conditions.

Learn how comments explain code with single-line and multi-line styles, and how to take user input with prompts using Python's input function, including type casting between strings and integers.

Discover how built-in and custom modules work in Python, including importing, aliasing, and using math and random modules. Learn to create and import your own modules for clean, reusable code.

Learn Python lists as dynamic, mutable data structures and master operations like append, extend, insert, slice, and index. Explore sort, reverse, count, and copy.

Explore Python dictionaries as fast key-value structures. Learn to create, access, and modify items with curly braces or dict(), using keys, values, and items.

Master Python indentation and whitespace to define blocks for if statements, loops, functions, and classes, improving readability and preventing errors.

Explore Python conditionals such as if, elif, and else to control program flow based on boolean conditions, with practical examples like weather checks and input handling.

Explore how while loops in Python control repetition using conditions and iterations, with break and continue to manage flow and avoid infinite loops.

Master Python for loops by iterating over lists, strings, ranges, and dictionaries. Learn break and continue, nested loops, and using enumerate to access index and item pairs.

Learn how to create user defined functions in Python using def, keyword arguments, pass parameters, call and return values, and pack arguments with *args and **kwargs for reusable, readable code.

Explore how default and optional arguments work in Python functions, learn about non-default arguments following defaults, and handle lists as arguments with function calls.

Learn Python object oriented programming by creating classes and objects, exploring inheritance, polymorphism, attributes and methods, constructors, and the role of self, with class, static, and bound methods.

Explore Python inheritance by deriving classes from a base, using super for initialization, reusing rectangle behavior in a square, and monkey patching to add methods post-definition (area, perimeter).

Explore how Python supports multiple inheritance by combining attributes and methods from two base classes, illustrated with a Foobar class and the super method for calling parent implementations.

Explore the __name__ == __main__ pattern in Python, showing how the interpreter sets special variables for main programs versus imported modules.

Explore data types in machine learning, including numerical data, continuous and discrete, categorical data, time series data, and textual data, and learn why encoding converts strings to numerical input.

learn data preprocessing for neural networks by importing pandas, numpy, and matplotlib, loading and cleaning a dataset, handling missing values, and preparing input features and labels.

Learn how to handle missing values with min and mean imputation, encode categorical data with label encoding, visualize distributions with matplotlib, and normalize features for neural network-ready data.

Split the data into training and testing sets, using an 80/20 ratio with a random_state to ensure repeatable results, train the model, and evaluate on unseen data.

Learn how to install and import numpy as np, create and manipulate arrays, perform arithmetic and advanced math (log, sqrt, exp), and reshape to build multi-dimensional arrays for machine learning.

Explore the pandas library for data preprocessing and manipulation in machine learning, including creating series and data frames, loading datasets, inspecting data, renaming columns, and removing duplicates.

Split data into training and testing sets to train a model on training data and evaluate its accuracy on unseen testing data using a defined test size and random state.

Learn how encoding converts categorical text to numerical values for machine learning models, using one-hot encoding, dummy variables, and level (label) encoding with salary categories as examples.

Learn how dimensionality affects machine learning, mastering underfitting and overfitting, choosing informative features, and using cross-validation and dimensionality reduction to build robust models.

Explore how logistic regression uses the sigmoid function and cross-entropy loss to train classifiers, addressing convexity, local versus global minima, and preparing for gradient descent.

Learn the basics of artificial neural networks in artificial intelligence, covering data representation, modeling, thinking, perception and action to map inputs to outputs via regression and classification concepts.

Explains how a neuron computes a weighted sum of inputs with weights and a bias, applies an activation function, and yields an activation value for 28×28 grayscale images.

Explore the history and applications of artificial neural networks, from early models to handwriting, face, and speech recognition, self-driving car. Learn training with forward and backward propagation in supervised learning.

Explore a simple neural network example that maps inputs to outputs by forward propagation and doubling, follow the seven steps, and minimize loss with the Adam optimizer toward global minimum.

Explore how gradient descent updates weights via partial derivatives to reduce loss. See how forward and backward propagation drive learning with mean squared error and cross-entropy losses in neural networks.

Explore partial differentiation fundamentals essential for back propagation algorithm in neural networks, including differentiating expressions, handling constants, and applying breakdown rules to compute derivatives.

Explore how activation functions shape neural network outputs, including sigmoid and relu, their domains, their role in backpropagation, and implications for binary and multiclass classification.

Learn how bias affects neural networks by adding an intercept to the input-weight sum, increasing model flexibility and reducing error, enabling activations to fire at different points.

Discover how gradient descent minimizes a cost function via iterative updates guided by the gradient and learning rate alpha. It covers partial derivatives and single or multi-input cases.

Explore stochastic gradient descent and how it updates weights one sample at a time to reduce computation. Learn how momentum and Adam optimizer smooth the path to a global minimum.

Explore how mini-batch stochastic gradient descent blends speed and bias reduction by updating weights with batches of data, balancing full gradient descent and single-sample methods for a neural network.

Derive the sigmoid function's derivative and apply it to backpropagation and gradient descent in a from-scratch neural network implemented in Python, using the result as sigmoid(x) times (1 minus sigmoid(x)).

Explore back propagation and feed-forward in a two-input neural network, using sigmoid activation and its derivative y*(1-y). Learn how weights and biases connect input, hidden, and output layers.

Explore the forward propagation in an artificial neural network, calculating net inputs, applying weights and bias, and using sigmoid activation to produce layer outputs, and a preview of backward propagation.

The lecture shows a two-neuron network comparing model outputs to targets, computing error with mean squared error, and updating weights and biases via backpropagation and gradient descent to reduce error.

Explore backpropagation in artificial neural networks by updating weights through gradient descent, using loss, derivatives, sigmoid activation, and a learning rate to minimize error.

Set up a neural network environment with conda, then build a single neuron from scratch in a Jupyter notebook, multiplying inputs by weights, adding a bias, and computing the output.

Code a simple neuron layer with four inputs, three neurons, and associated weights and biases, demonstrate weighted sums, and explain why lack of activation yields linear regression.

Use dot product to compute a neuron layer's output by multiplying inputs with weights and adding biases, implemented in Python from scratch for a neural network.

Code a dense layer from scratch using object oriented programming, implementing forward propagation with dot products, weights, and biases across multiple inputs and neurons.

Explore how activation functions bound net input via weights and bias, with examples like sigmoid, enabling non-linear learning in artificial neurons and deep neural networks.

Learn how activation functions introduce non-linearity, focusing on step and sigmoid functions, their thresholding behavior, net input with weights and bias, and neuron firing, binary classification, and vanishing gradient.

Explore hyperbolic tangent and ReLU activation functions, comparing domains from minus one to one versus zero to one, and explain how sigmoid, backpropagation, and vanishing gradients motivate using ReLU.

Create a custom dataset from scratch and implement a multilayer perceptron classifier with skateland to solve multiclass classification in an artificial neural network.

Explore visualizing imbalanced multiclass data, balance it with oversampling using imbalance-learn, and implement a scikit-learn mlp classifier with a 300-neuron hidden layer and adam optimizer.

Evaluate the neural network with 10-fold cross-validation, train-test splits, and a confusion matrix, then use classification metrics to tune hyperparameters for higher accuracy.

Experiment with hyperparameters in a neural network to improve training performance. Explore learning rate strategies and activation functions such as relu and sigmoid, plus hidden layer and neuron counts.

Explore forward and backward propagation in neural networks using a computational graph, including multiplicative and additive gates, loss, and backpropagation with local and upstream gradients.