Data science foundations 10
- Data Science Lifecycle:
The iterative loop of Ask, Collect, Clean, Explore, Model, Communicate and Decide; each stage feeds back into earlier ones. Lec 1
- Data Types:
Quantitative (continuous or discrete) versus categorical (nominal or ordinal) variables; the type decides which summaries and charts make sense. Lec 1
- Unstructured Data:
Data that does not sit in rows and columns, such as text, images and audio; roughly 80 percent of business data. Also in PS 06. Lec 1
- Four Types of Analytics:
Descriptive (what happened), diagnostic (why), predictive (what will happen) and prescriptive (what to do). Lec 1
- Correlation vs Causation:
Two variables moving together does not mean one causes the other; a hidden third variable can drive both. Seen in: Machine Learning. Lec 1
- Confounding Variable:
A hidden third variable that drives both observed variables and creates a misleading association between them. Lec 1
- Survivorship Bias:
Looking only at the successes and ignoring the failures, which distorts conclusions (Wald's WW2 aircraft armour example). Lec 1
- Data Ethics:
Consent, purpose limitation, data minimisation, deletion rights and explainability when collecting and using personal data. Lec 1
- Google Colab:
Free cloud-hosted Jupyter notebook that runs Python in the browser with no local setup. Lec 1
- Garbage In, Garbage Out:
Bad input data gives bad results whatever the model; cleaning can take up to 80 percent of project time. Lec 1
Data handling and exploratory analysis 13
- Exploratory Data Analysis:
First-pass investigation of a dataset with summaries and charts to find structure, problems and hypotheses before modelling. Seen in: Machine Learning. Also in PS 01. Lec 2
- Pandas:
Python library for tabular data, built around the DataFrame. Seen in: Machine Learning. Lec 2
- DataFrame:
Pandas table of labelled rows and columns, the standard container for data in this course. Seen in: Machine Learning. Also in PS 01. Lec 2
- Summary Statistics:
Mean, median, standard deviation, quartiles and counts that describe a variable in a few numbers (df.describe()). Seen in: Machine Learning. Lec 2
- Distributions:
The shape of how a variable's values are spread: skew, spread, modes and tails. Seen in: Machine Learning. Lec 2
- Groupby:
Split a DataFrame by category, compute a statistic per group, and combine the results. Seen in: Machine Learning. Also in PS 01. Lec 2
- Missing Values:
Empty (NaN) cells; detect them, then drop rows or impute with the mean, median or mode. Seen in: Machine Learning. Also in PS 01. Lec 2
- Sentinel Value:
A placeholder such as -1 or 999 that stands for missing data but is not NaN, so it must be found and replaced by hand. Also in PS 02. Lec 2
- Outliers:
Values far from the bulk of the data; check whether they are errors or real before removing them. Seen in: Machine Learning. Also in PS 01. Lec 2
- Histograms:
Bar chart of binned counts showing the distribution of one numeric variable. Seen in: Machine Learning. Also in PS 01. Lec 2
- Boxplots:
Chart of median, quartiles and whiskers that compares spread across groups and flags outliers. Seen in: Machine Learning. Also in PS 01. Lec 2
- Matplotlib:
Core Python plotting library that Seaborn builds on. Seen in: Machine Learning. Lec 2
- Seaborn:
Statistical plotting library on top of Matplotlib with cleaner defaults for grouped charts. Seen in: Machine Learning. Lec 2
Classification 11
- Classification:
Predicting a discrete label (a category) from features; binary with two classes, multiclass with three or more. Also in PS 01. Lec 3
- Supervised Learning:
Learning a mapping from features to a known label using labelled examples. Seen in: Machine Learning. Also in PS 01. Lec 3
- Decision Tree:
Flowchart classifier that splits the data with if/else tests on features until the leaves are mostly one class. Seen in: Machine Learning. Also in PS 01. Lec 3
- Gini Impurity:
Node purity measure for tree splits: 0 when a node holds one class, higher when classes are mixed. Seen in: Machine Learning. Lec 3
- Logistic Regression:
Probabilistic binary classifier: a linear score passed through the sigmoid function, then thresholded. Seen in: Machine Learning. Also in PS 01. Lec 3
- Sigmoid Function:
Maps any real number into (0, 1) so it can be read as a probability. Seen in: Machine Learning. Lec 3
- KNN:
K-nearest neighbours: classify a point by the majority label among its K closest training points. Seen in: Machine Learning. Also in PS 01. Lec 3
- Random Forest:
Ensemble of decision trees trained on bootstrap samples with random feature subsets, voting for the answer. Seen in: Machine Learning. Also in PS 01. Lec 3
- Ensemble Methods:
Combining several models so their errors average out, giving lower variance than any single model. Seen in: Machine Learning. Lec 3
- One-Hot Encoding:
Turning a categorical variable into one binary column per category. Seen in: Machine Learning. Also in PS 01, PS 02. Lec 3
- Feature Scaling:
Putting features on a comparable scale (StandardScaler or MinMaxScaler); essential for distance-based and penalised models. Seen in: Machine Learning. Also in PS 01, PS 02. Lec 3
Evaluation metrics and validation 9
- Train Test Split:
Holding out part of the data as a test set so performance is measured on unseen examples. Seen in: Machine Learning. Also in PS 01. Lec 3
- Confusion Matrix:
Table of true and false positives and negatives that shows exactly where a classifier goes wrong. Seen in: Machine Learning. Also in PS 01. Lec 3
- Precision and Recall:
Precision is the share of predicted positives that are right; recall is the share of real positives found; both matter on imbalanced classes. Seen in: Machine Learning. Lec 3
- Cross-Validation:
K-fold evaluation: train and test on K different splits and average, for a more robust score than one split. Seen in: Machine Learning. Also in PS 02. Lec 4
- Data Leakage:
Information from the test set or target reaching the training step, which inflates scores that will not hold on new data. Seen in: Machine Learning. Also in PS 02. Lec 4
- Scikit-Learn Pipeline:
Chains preprocessing and a model into one object so scaling is fitted only on training data, preventing leakage. Seen in: Machine Learning. Also in PS 02. Lec 4
- RMSE:
Root mean squared error: typical prediction error in the units of the target, with large errors penalised more. Seen in: Machine Learning. Also in PS 04. Lec 4
- MAE:
Mean absolute error: average size of the prediction errors in target units, less sensitive to outliers than RMSE. Also in PS 02. Lec 4
- R-Squared:
Share of the target's variance the model explains, between 0 and 1 on the data it is scored on. Seen in: Machine Learning. Also in PS 04. Lec 4
Regression and regularisation 10
- Linear Regression:
Model that predicts a continuous target as a weighted sum of the features plus an intercept. Seen in: Machine Learning. Also in PS 04. Lec 4
- OLS:
Ordinary least squares: picks the coefficients that minimise the sum of squared residuals. Seen in: Machine Learning. Lec 4
- Multicollinearity:
Features that are strongly correlated with each other, making individual coefficients unstable and hard to interpret. Seen in: Intro Econometrics. Also in PS 04. Lec 4
- Dummy-Variable Trap:
Perfect multicollinearity from keeping every dummy level plus an intercept; drop one level per category. Seen in: Intro Econometrics. Also in PS 02. Lec 4
- Polynomial Features:
Squared and cross-product columns (interaction terms) that let a linear model fit curves and joint effects. Also in PS 02. Lec 4
- Overfitting:
A model memorises training noise: training error keeps falling while test or validation error rises. Seen in: Machine Learning. Also in PS 02, PS 07. Lec 4
- Bias-Variance Tradeoff:
Simple models underfit (high bias); flexible models overfit (high variance); the aim is the balance with the lowest test error. Seen in: Machine Learning. Also in PS 02. Lec 4
- L1 and L2 Penalty:
Regularisation terms added to the loss: L1 adds the sum of absolute coefficients, L2 the sum of squared coefficients. Also in PS 02. Lec 4
- Ridge Regression:
L2-regularised linear regression that shrinks all coefficients towards zero without removing any. Seen in: Machine Learning. Also in PS 02, PS 04. Lec 4
- Lasso Regression:
L1-regularised regression that can shrink coefficients exactly to zero, so it also selects features. Seen in: Machine Learning. Also in PS 02, PS 04. Lec 4
Clustering and dimensionality reduction 14
- Unsupervised Learning:
Finding structure in unlabelled data, with no target to predict. Seen in: Machine Learning. Also in PS 03, PS 05. Lec 5
- K-Means:
Centroid-based clustering: assign points to the nearest centroid, move centroids to the cluster means, repeat. Seen in: Machine Learning. Also in PS 03, PS 05. Lec 5
- Centroid:
The mean point of a cluster, which K-Means moves on every update step. Also in PS 03. Lec 5
- n_init:
K-Means setting for the number of random restarts; scikit-learn keeps the run with the lowest inertia. Also in PS 03. Lec 5
- Euclidean Distance:
Straight-line distance between two points, the default metric for K-Means and KNN. Seen in: Machine Learning. Lec 5
- Inertia:
Sum of squared distances from each point to its own centroid (within-cluster sum of squares); K-Means minimises it. Also in PS 03. Lec 5
- Elbow Method:
Plot inertia against K and choose the K where the curve bends and gains flatten. Seen in: Machine Learning. Also in PS 03, PS 05. Lec 5
- Silhouette Score:
Per-point measure of how close a point is to its own cluster versus the next nearest, from -1 to +1. Seen in: Machine Learning. Also in PS 03, PS 05. Lec 5
- Internal vs External Cluster Validation:
Internal checks use only the geometry (silhouette, inertia); external checks compare clusters with ground-truth labels. Also in PS 03. Lec 5
- Customer Segmentation:
Business use of clustering to group customers with similar behaviour for targeted action. Seen in: Machine Learning. Also in PS 03. Lec 5
- PCA:
Principal component analysis: linear dimensionality reduction onto the directions of greatest variance. Seen in: Machine Learning. Also in PS 03, PS 05. Lec 5
- Principal Component:
A direction of maximum remaining variance found by PCA; components are uncorrelated and ordered by variance. Also in PS 03. Lec 5
- Explained Variance:
Share of the total variance captured by each principal component, used to decide how many to keep. Seen in: Machine Learning. Also in PS 03. Lec 5
- Curse of Dimensionality:
In many dimensions distances lose meaning and data becomes sparse, hurting KNN and K-Means. Seen in: Machine Learning. Lec 5
Recommendation systems 18
- Recommendation System:
Predicts how much a user would like an unseen item and suggests the top-ranked ones. Lec 6
- User-Item Matrix:
Users as rows, items as columns, ratings as values, mostly empty. Lec 6
- Sparsity:
The share of user-item cells with no rating; real matrices are mostly empty. Lec 6
- Explicit Feedback:
Direct signals of preference such as star ratings. Lec 6
- Implicit Feedback:
Preference inferred from behaviour such as watch time or clicks. Lec 6
- Long Tail:
Many niche items each with little popularity; recommenders help users reach them. Lec 6
- Content-Based Filtering:
Recommends items similar to ones the user liked, using item features. Lec 6
- Cosine Similarity:
Angle-based similarity between two vectors, from -1 to +1, ignoring their length. Also in PS 06. Lec 6
- User Profile:
Average feature vector of the items a user rated highly, matched against candidate items. Lec 6
- Collaborative Filtering:
Recommends from the rating patterns of similar users or items, with no item features needed. Lec 6
- User-Based CF:
Predicts a rating from what similar users gave the item. Lec 6
- Item-Based CF:
Predicts a rating from the user's own ratings of similar items. Lec 6
- Cold Start Problem:
Too little data about a new user or item to make good recommendations. Lec 6
- Hybrid Recommender:
Weighted blend of content-based and collaborative approaches to cover each other's weaknesses. Lec 6
- Matrix Factorization:
Splits the user-item matrix into low-rank user and item latent-factor matrices whose product fills the gaps. Lec 6
- Precision at K:
Share of the top K recommendations that are relevant. Lec 6
- Recall at K:
Share of all relevant items that appear in the top K. Lec 6
- Filter Bubble:
Recommenders narrowing exposure to familiar content only. Lec 6
Natural language processing 9
- Natural Language Processing:
Making computers understand and generate human language. Also in PS 06. Lec 7
- Text Preprocessing:
Cleaning text before modelling: lowercase, strip punctuation, tokenize, drop stopwords, stem or lemmatize. Also in PS 06. Lec 7
- Tokenization:
Splitting text into tokens, usually words. Also in PS 06. Lec 7
- Stopwords:
Very common, low-meaning words (the, is, and) removed before analysis. Also in PS 06. Lec 7
- Stemming:
Chopping suffixes to a rough root, which may not be a real word. Also in PS 06. Lec 7
- Lemmatization:
Reducing a word to its dictionary form using grammar, so results are real words. Also in PS 06. Lec 7
- Bag of Words:
Represents a document as word counts and ignores word order. Also in PS 06. Lec 7
- Document-Term Matrix:
Matrix with one row per document and one column per term, filled with counts. Also in PS 06. Lec 7
- TF-IDF:
Word count weighted up for rare words and down for words common to every document. Also in PS 06. Lec 7
Image analysis 6
- Image Analysis:
Extracting information from visual data (photos, video, scans), also called computer vision; images are already grids of pixel numbers. Also in PS 06. Lec 7
- Pixel:
Smallest image unit, holding a brightness or colour value. Lec 7
- Filter:
Small kernel grid of weights slid across an image to detect a pattern such as edges. Also in PS 06. Lec 7
- Convolution:
Slide a filter over an image, multiply with the pixels beneath, and sum to one output value. Also in PS 06. Lec 7
- Feature Map:
The output grid produced by applying one filter across an image. Lec 7
- Convolutional Neural Network:
Deep network whose filters are learned from data instead of hand-designed. Lec 7
Deep learning 17
- Deep Learning:
Machine learning with many layers that learn useful features automatically. Lec 8
- Neural Network:
Layers of connected units that transform inputs into outputs through learned weights. Also in PS 07. Lec 8
- Perceptron:
One unit: a weighted sum of its inputs plus a bias, passed through an activation. Lec 8
- Weight:
Number controlling how much one input matters to a neuron; learned in training. Lec 8
- Bias:
Constant added before the activation; shifts the decision boundary. Lec 8
- Activation Function:
Non-linear function applied after the weighted sum, which is what lets networks learn curves. Also in PS 07. Lec 8
- ReLU:
Activation max(0, z): passes positive values, zeroes negatives. Lec 8
- Multi-Layer Perceptron:
Network of fully connected layers: input, one or more hidden layers, output. Also in PS 07. Lec 8
- Hidden Layer:
Middle layer whose neurons learn increasingly abstract features. Lec 8
- Forward Pass:
Data flowing from input to output through the layers to produce a prediction. Lec 8
- Width vs Depth:
Neurons per layer (width) versus number of hidden layers (depth). Lec 8
- Loss Function:
Measures how wrong a prediction is; training tries to minimise it. Lec 8
- Gradient Descent:
Repeatedly step the weights downhill on the loss surface. Seen in: Machine Learning. Lec 8
- Learning Rate:
Size of each gradient descent step: too high overshoots, too low crawls. Lec 8
- Epoch:
One full pass through the training set. Lec 8
- Backpropagation:
Tracing the error backward through the network to work out how to adjust each weight. Lec 8
- Early Stopping:
Stop training when validation loss stops improving, to avoid overfitting. Also in PS 07. Lec 8