Data Science & Machine Learning(Theory+Projects)A-Z 90 HOURS

Explore turning problem solving into a universal solution by defining an algorithm. Translate steps into pseudocode and implement them in Python.

Compare flowcharts and pseudocode to build general solutions, then convert to Python code. Use the make tea problem to illustrate loops, input variations (sugar, milk), and iterative testing of conditions.

Discover why Python is the default data science language, with readability and open source libraries. Follow a beginner-friendly path from zero to advanced Python use for data science.

Explore why Python and Jupyter notebooks are favored for data science. Learn how Jupyter combines code, testing, and visualization in an interactive, web-based environment with markdown and LaTeX support.

Learn how to install Python using the Anaconda distribution, set up the 64-bit Windows version, launch the Anaconda Prompt, and run Jupyter Notebook to start Python coding.

Explore how to use the IPython shell and the Jupyter notebook within Anaconda, write Python code like a calculator, and introduce variables for storing and reusing results.

Use Python if statements to compare two user-provided numbers and print the larger one. Explore the role of boolean comparisons and the basic structure of if blocks.

Explore nested if conditions in python, where an if body contains another if. Understand how indentation defines blocks and how top-level and nested else parts shape outputs.

Practice Python control flow with if statements and comments, solving a real-number task by extracting the integer portion before the decimal and checking if it is even or odd.

Learn how to use the Python for loop to populate lists, iterate with range, and compute squares, while applying indexing and append operations.

Learn how Python functions encapsulate tasks with def and descriptive names. Call the function to run its body, and see this workflow demonstrated in a Jupyter Notebook.

Explore defining and calling Python functions with multiple arguments, and learn how argument order matters and how using parameter names makes calls order-free.

Learn how default values in Python functions are assigned at definition time and remain fixed, while mutable types like lists can share memory and influence subsequent calls.

Practice building and using Python functions to sort a list, including find minimum and its index, swap values, and composing functions in a module for clearer code.

Master Python string indexing and slicing, including zero-based and negative indices, end not included, and steps, while noting strings are immutable and reversing with [::-1].

Explore Python data structures in practice: list append and plus usage, tuple immutability and concatenation, set add and update, and dictionary insertion and deletion with del in Jupyter.

Explore lists, tuples, sets, and dictionaries via tab completion and help, and learn list methods like append, clear, pop, reverse, plus set and dictionary operations such as update and union.

Explore numpy, a fast numerical Python package, and learn how arrays store homogeneous numeric data efficiently. Create arrays from lists or tuples, inspect dtype, and use numeric types.

Explore how numpy shape property and ndim describe dimensions using a 3d array example, and learn about size, nbytes, and universal (vectorized) functions.

Explore core NumPy array creation and manipulation, including zeros and ones, arange and range behaviors, random permutation and rand int, and the reshape operation, with hands-on examples.

Discover how numpy slicing accesses a memory view versus creating a copy. Explore indexing in 1d and 2d arrays, and learn practical examples of reversing and extracting subarrays.

Learn to manipulate covid-19 data in pandas by grouping by country and date, summing confirmed, deaths, and recovered, and filtering records.

Explore seaborn's joint plot and pair plot to visualize two-dimensional distributions and pairwise relationships, using kde options, axis styles, and color-coded classifications for multi-attribute data.

Learn how numpy universal functions, including add, sum, and plus operators, enable fast numerical data processing, handle nan values, and manage shapes with new axis and squeeze.

Explore NumPy universal functions and operators for efficient, element-wise data processing, including subtract, multiply, power, and remainder, and understand how aggregates like mean and std reduce dimensions.

Practice numpy ufuncs, comparisons, and logical operators by building a 50-element array multiplied by 10, removing decimals, filtering values greater than five, and sorting ascending and descending.

Explore NumPy for numerical data processing by creating arrays with np.random.rand, removing decimals via floor, filtering values greater than five, and sorting arrays in ascending and descending order.

Explore numpy's ability to store heterogeneous data using structured arrays, defining fields like name, id, marks, and GPA; see how numpy structures underpin pandas and enable field-based indexing.

Learn to build a NumPy structured array with four fields: employee type (permanent or non-permanent), age (integer), birth year (string), and salary (integer), and populate three entries.

Learn to create and manipulate pandas data frames from series or dictionaries, access and modify columns, transpose, and apply masking while preparing for missing value handling.

Build a two-column pandas data frame with name (string) and vaccine (boolean), populate five entries, then create a new data frame with names where the vaccine status is true.

Explore how pandas loc and iloc differentiate explicit versus implicit indexing in series and dataframes, with practical examples for selecting, slicing, and reversing rows using numpy-like indexing.

Extend the previous pandas data frame by adding a country column with entries Pakistan and China. Group the data by country and calculate the sum to practice grouping in pandas.

Demonstrates creating a pandas data frame and applying a three-value rolling window, summing each window, stepping by three, and excluding NaN values.

Learn to create a pandas data frame and apply rolling with a three-value window to sum values, while excluding NaN entries and validating the results.

Practice using pandas clip to bound a data frame between 10 and 30, replacing values below 10 with 10 and above 30 with 30 using the where method.

Learn to merge multiple data frames in pandas to consolidate student records using the pandas merge function, with practice through a merge quiz.

Master the basics of data visualization with Matplotlib as the foundation. Explore Seaborn and Bokeh, import pyplot as plt, and try styles like classic and Seaborn White Grid in notebooks.

Generate data with NumPy and visualize it with Matplotlib, plotting sine and cosine curves and multiple plots on a figure. Learn to switch styles, including Seaborn, using plt and ax.

Master how to set matplotlib axis limits using x from 1 to 14 and y from -1 to 1.2, then reverse the y axis by swapping y limits.

Annotate figures with labels, titles, and axis descriptions in Matplotlib to make plots descriptive; learn legends for color-coded curves, and place them precisely for clearer data visualization.

Explore how Matplotlib's hist function visualizes data distributions through histograms and density plots, comparing multiple datasets with customizable bins, alpha blending, and step-filled histograms, using iris data as examples.

Practice creating a 3d scatter plot using Matplotlib, with specific x, y, z data, green dot marker for the first plot and yellow a-plus marker for the second, following guidelines.

Explore creating and customizing 3d surface plots with matplotlib's mplot3d, using plot_surface on X, Y, Z data, color maps, alpha, and meshgrid examples; and preview seaborn for high level plots.

Explore seaborn relplot fundamentals through a hands-on iris dataset quiz, creating a relationship plot with x and y variables and species as the style.

Explore Seaborn relplot facets to visualize the relationship between size and total bill, dividing by smoker status and using hue and style for gender.

Explore Seaborn catplot to visualize tips data by day, revealing how to tune jitter, hue by gender, and compare box, violin, swarm, and bar styles.

Explore 3d interactive plotting with Plotly, building a 3d scatter plot in Jupyter notebook, and customize traces, layout, markers, and color scales.

Learn to build a Plotly 3D interactive surface plot using x values from 20 to 60 with 40 samples, paired with y and z, and the ice fire color scale.

Discover how Pandas' built-in plotting module, built on Matplotlib, lets you create histograms, area plots, and more directly from data frames with df.plot, alongside seaborn, bokeh, and plotly.

Master pandas for plotting through a concise bonus video, reinforcing practical data skills and encouraging thoughtful reviews to help future learners.

Examine the definition of a set and determine if a set can include heterogeneous objects, using integers and strings as examples to analyze set theory basics.

Define the concept of a set as a collection with distinct, unordered elements, capable of containing heterogeneous objects in theory, though practical usage favors separate sets for different types.

Explore data science concepts of set, multisite (multiset), and topal, showing distinctness and unorderedness: sets have no duplicates and are unordered; multisite allows duplicates but is unordered; topal has neither.

Explore sets and elements, membership notation, and the empty set; distinguish finite and infinite sets by cardinality, and contrast countable versus uncountable sets with examples.

Explore how Venn diagrams visualize set operations—union, intersection, difference, and complement—within a universal set, and see links to probability, subsets, supersets, and partitions.

Learn to construct and enumerate the sample space for three rolls of a four-sided die and a coin toss, and determine that the sample space contains 128 elements.

Explore the terminology of experiment, outcomes, sample space, sample point, and event, then attempt homework on a four-sided die, finite versus infinite sample spaces, and even-sum events.

Design sample spaces and assign non-negative probability laws to events, defining a probability model that quantifies likelihoods and enables unambiguous predictions.

Show that an empty event cannot have non-zero probability, based on the probability axioms. Use P(sample space)=1 and P(A∪B)=P(A)+P(B) for disjoint events, and decompose S as S∪∅ to conclude P(∅)=0.

Explore discrete probability models with two dice, define sample spaces and probability laws, and compute events such as even sums and at least one die showing four.

Apply probability axioms and set theory to compute the probability of neither malaria nor typhoid using complements, union and intersection, yielding 0.1 and introducing a continuous probability model.

Explore conditional probability using a six-face die example. Derive the rule P(A|B)=P(A∩B)/P(B) and note independence when A's likelihood does not change with B, normalizing by P(B) as the sample space.

Explore how conditional probability underpins machine learning, modeling distributions of random variables for tasks like face recognition, activity recognition, and text-to-speech, linking statistics to classifiers and regressors.

Explain the concept of independence in probability, including statistical independence, conditional probability, and A given B, A intersection B, and independence across multiple events and subsets.

Explore how a zero probability for a specific value of a random variable relates to empty or impossible events, using practical examples.

Implement a Bernoulli trial in Python using NumPy to simulate coin tosses, estimating the probability of success by counting heads and tails across many trials.

Assess whether the next ball outcome can be modeled as a Bernoulli trial, a type of random variable, with six as the success leading to team A's win.

Learn to implement a binomial trial in Python with a vectorized numpy approach. Plot histograms to visualize how n and p shape the Bernoulli/binomial distribution and its mean.

Explore the exponential distribution as a continuous random variable with density lambda e to the minus lambda x for x greater than zero, where lambda is a parameter shaping curve.

Explore how the Gaussian distribution reacts to changes in sigma, comparing large versus small sigma, in this exercise on continuous random variables for data science and machine learning.

Explore the law of large numbers with iid data, showing how the sample mean converges to the population mean as sample size grows, and review key distributions and their means.

learn how transforming a random variable affects its expected value and how to compute moments and variance using the original probability mass or density function.

Compute the expectation of the product for two independent discrete random variables X and Y using their joint pmf, showing that E[XY] = E[X] E[Y].

Explore the general classification problem, predicting the finite discrete class Y from data X by a probability model, and compare generative modeling with discriminative modeling.

Understand how the curse of dimensionality makes joint probability estimation difficult as the number of random variables grows, demanding more data and tools like principal component analysis for dimensionality reduction.

Learn how parametric distributions are described by parameters like mu, sigma, and lambda, estimate them from data, and compare with non parametric approaches using kernel density estimates.

Explore logistic regression as a powerful binary classifier grounded in maximum likelihood on Bernoulli variables, using the logistic function to model Y given X and minimize binary cross-entropy loss.

Learn ridge regression, a powerful regression model that adds a ridge regularizer to ordinary least squares, improving generalization and connecting regression to probabilistic methods.

Explore data science and machine learning concepts through approachable techniques and methods, with emphasis on practical learning for new learners and workplace applications.

Explore machine learning fundamentals without heavy math, using intuitive explanations, Python code, and visualizations to understand driving forces behind models and apply to real-world projects like face recognition.

Learn the fundamentals of machine learning through theory and extensive Python practicals, with live coding on real and synthetic datasets. Explore features, regression, classification, clustering, and from-scratch models.

Convert text attributes to numeric features with one hot encoding and drop uninformative fields. Prepare data for machine learning pipelines with text and numeric data, managing the curse of dimensionality.

Discover why data standardization improves convergence and stability, and learn scaling techniques like zero-to-one ranges and the standard scalar with scikit-learn examples.

Explore how a model functions as a map from feature space to outcomes, define dimensions, and understand parameters, hyperparameters, and training from data.

Explore how model flexibility drives overfitting and harms generalization in regression, using polynomial features on noisy data to contrast training versus unseen data.

Measure generalization by holding out a validation set from training data, using an 80/20 split, and evaluating unseen data performance to detect overfitting.

Explore how convolutional neural networks enable learning for image data by automatically extracting salient features. Observe how convolutional and pooling layers build features fed to a classifier, unlike hand-engineered features.

Explore what overfitting means in data science and machine learning, and examine how dimensionality reduction relates to overfitting, including whether the two are linked or independent.

Explain why feature selection lowers dimensionality, improves model performance and generalization, and preserves original feature identities for better interpretation and data acquisition.

Explore how feature selection searches the subset space to pick the best features, using filter and wrapper methods, and greedy strategies like forward selection and backward elimination.

Examine why exhaustive subset search fails for wrapper feature selection and how greedy search offers a general approach, and relate simulated annealing to subset selection and filter methods.

Join an activity to determine whether forward selection can be learned using recursive feature elimination, backward elimination, and recursive feature elimination with cross-validation.

Explore linear independence by checking whether a vector can be formed from linear combinations of other vectors. Learn about scalars, as well as linearly independent sets in vector spaces.

Explore eigenvalues, eigenvectors, and eigen space, including how eigen decomposition preserves direction, forms spaces for each eigenvalue, and their role in optimization and data science.

Explore the properties of symmetric and positive semidefinite matrices, including real eigenvalues, orthonormal eigenvectors, and orthogonal diagonalization, with applications to principal component analysis and singular value decomposition.

This activity provides a mathematical treatment of linear algebra for data science, linking Strang’s book, MIT OCW videos, and Ali Ghodsi lectures on dimensionality reduction.

Master principal component analysis to perform unsupervised dimensionality reduction and feature extraction by identifying the best subspace and projecting data orthogonally to preserve information.

PCA identifies a subspace that best represents the data, balancing minimum reconstruction error and maximum variance after projection, to retain information.

Kernel PCA extends PCA with SVD-based insights to enable nonlinear dimensionality reduction on centered data using the eigenvectors of x^t x and the d and v matrices.

Learn kernel pca and isomap as nonlinear dimensionality reduction techniques that use kernel matrices to preserve pairwise similarities and geodesic geometry, via eigen-decomposition and graph-based distances.

Explore how one hot encoding converts categorical features, like neighborhoods, into a numeric matrix with sklearn's dict vectorizer. It expands features and yields sparse, memory-efficient matrices.

Explore text features for information retrieval by counting term frequencies and applying tf-idf weighting to rank documents, converting text to feature vectors with count vectorizer and tf-idf.

Explore the scalability limits of traditional feature selection and extraction for big data, and how activity-feature scaling motivates new, more efficient dimensionality reduction algorithms.

Learn the fundamentals of machine learning and the common concepts across subfields, including neural networks, framed as predictive modeling that builds models to predict future outcomes.

Explore how classification uses training data, input vectors, and labels to learn from data, with feature extraction and dimensionality considerations, and how validation and test data assess generalization.

The video shows solving a machine learning exercise by substituting x1, x2, x3 and parameters, performing arithmetic to produce a prediction of 11, while noting the ground truth may differ.

Explore machine learning model types, distinguishing linear from nonlinear models, and learn how training data guides the selection of function forms like polynomial, sinusoidal, and exponential.

Identify how a model remains linear in parameters despite a constant addition; the video explains that such functions are still treated as linear models in machine learning, called fine functions.

Identify real life problems that require multi target modeling, where inputs and outputs are represented by many numbers, and think through real life examples that fit this approach.

Explore machine learning literature to define overfitting and brainstorm practical ways to prevent it, as a prelude to the upcoming solution video.

Explore hyperparameter tuning for regularization, sampling values from 0 to infinity to minimize validation loss, with coarse-to-fine refinement and cross-validation guiding iterative parameter optimization.

Build a three-layer neural network with two computational layers and one output; initialize weights, perform the forward step via matrix multiplication, and discuss activation functions.

Explore why nonlinearity in activation functions enables deep networks to learn diverse features, compare sigmoid, relu, and softmax, and highlight differentiability and computational ease for backpropagation.

Explain binary cross-entropy loss for classification. Identify true labels as 0 or 1 and show predicted probabilities from a sigmoid producing zero loss when they match, otherwise large loss.

Shows why the negative gradient direction minimizes the loss via linear approximation, and weighs small learning rates against larger steps for faster gradient descent convergence.

Demonstrates gradient descent on a simple sigmoid unit with a fixed learning rate to minimize loss across iterations, showing forward pass, backward pass, and weight updates.

Discover how learning rate controls step size in deep neural networks, balancing overshoot risk and convergence speed. Explore schedulers, decay strategies, and validation-based tuning across epochs.

Learn how batch normalization stabilizes mini-batch gradient descent, mitigating covariate shift between training and test distributions, and offers regularization, with decisions on when to normalize and how many layers.

Early stopping uses a validation set to monitor training and validation loss, stopping when the validation loss stops decreasing and using a patience parameter to avoid overfitting.

Depth enables deep neural networks to model complex functions with fewer neurons than a single layer, despite the universal approximation theorem, using layered arrangements and tunable hyperparameters.

Explore why the bias term matters in deep neural networks, offsetting hyperplanes and enabling outputs as class probabilities or regression targets, within fully connected architectures and layer conventions.

Compare gradient descent variants, showing momentum and RMSProp accelerate convergence toward a global minimum, outperforming plain SGD; practical guidance: many batches, batch normalization, accelerated algorithms.

Explore advanced pandas conditional selections using and, or, and parentheses, and apply them to Titanic data with read_csv, head, describe, and null checks to handle missing values.

Learn data cleaning and preprocessing for machine learning by handling missing values with median, encoding categoricals with one-hot or label encoding, and preparing a numeric Titanic dataset ready for modeling.

Explore how a grayscale image is stored as a matrix of unsigned integers, with 0 for black and 255 for white. Eight-bit images offer 256 gray levels across pixels.

Learn to read and display images in Python using NumPy and Matplotlib, inspect and manipulate RGB channels, view subregions, discard the fourth channel, and experiment with channel-specific color adjustments.

Learn how to convert a color image to grayscale in Python by using weighted channel contributions (0.2989, 0.5870, 0.1140) versus uniform averaging, with matplotlib and OpenCV.

Explore image filtering as the foundation of convolution, using sliding 3x3 masks to detect features and edges, and distinguish convolution from image filtering and cross correlation.