Data Science

You can write clean Python scripts, manipulate multidimensional arrays with NumPy, and clean tabular datasets using Pandas DataFrames.

1. Python syntax, data types, functions and OOP4 hrs
   • lists, dicts, and generator expressions
   • classes, dunder methods, and inheritance
   • exception handling with try/except blocks
2. NumPy: arrays, broadcasting and linear algebra4 hrs
   • ndarray creation and dtype optimization
   • vectorised operations and broadcasting rules
   • dot products, matrix inversion, and eigenvalues
3. Pandas: DataFrames, merging, groupby and reshaping6 hrs
   • DataFrame indexing, slicing, and loc/iloc
   • groupby aggregations and multi-index grouping
   • merge, join, and pivot_table operations
4. Matplotlib and Seaborn for exploratory visualization5 hrs
   • Figure and Axes object manipulation
   • distribution plots with histplot and kdeplot
   • correlation heatmaps using pairplot and heatmap
5. Jupyter Notebook and Google Colab workflow5 hrs
   • cell execution order and kernel management
   • magic commands like %timeit and %matplotlib
   • mounting Google Drive and importing external datasets

Hands-on project: Build an automated CSV cleaning script using Pandas and NumPy that handles missing values, normalises numeric columns, and exports a clean dataset.

You can calculate descriptive statistics, design hypothesis tests, and evaluate statistical significance using p-values.

1. Descriptive statistics: mean, median, variance, skewness5 hrs
   • measures of central tendency and dispersion
   • interquartile range and boxplot outlier detection
   • skewness and kurtosis interpretation
2. Probability theory and distributions5 hrs
   • conditional probability and Bayes theorem
   • binomial and Poisson discrete distributions
   • Gaussian normal distribution and standardisation
3. Hypothesis testing, p-values and confidence intervals5 hrs
   • null and alternative hypotheses formulation
   • Type I and Type II error trade-offs
   • student's t-test and chi-square test execution
4. Correlation, covariance and statistical significance4 hrs
   • Pearson vs Spearman correlation coefficients
   • covariance matrix calculation
   • degrees of freedom and critical values
5. Bayesian thinking and A/B testing fundamentals5 hrs
   • prior, likelihood, and posterior calculation
   • conjugate priors in update steps
   • sample size determination for A/B tests

Hands-on project: Perform an A/B test analysis on user conversion data, calculating confidence intervals, t-tests, and p-values in Python.

You can transform messy raw files into structured dataframes, engineer categorical features, and build interactive EDA dashboards.

1. Handling missing values and outliers5 hrs
   • MCAR, MAR, and MNAR missingness types
   • imputation using SimpleImputer and KNNImputer
   • Z-score and IQR clipping for outlier treatment
2. Data cleaning and transformation pipelines5 hrs
   • string cleaning and regex extraction in Pandas
   • datetime parsing and resampling time indices
   • memory reduction via categorical downcasting
3. Feature extraction and encoding categorical variables5 hrs
   • One-Hot Encoding and Label Encoding
   • target encoding for high-cardinality features
   • binning continuous variables into intervals
4. EDA on real-world datasets (housing, healthcare, finance)4 hrs
   • univariate and bivariate profiling
   • identifying collinearity and target leakage
   • summarising findings into structured executive reports
5. Interactive visualization with Plotly and Tableau5 hrs
   • Plotly Express scatter and sunburst charts
   • connecting Tableau Desktop to local CSV extracts
   • building calculated fields and dashboard filters

Hands-on project: Build an automated EDA pipeline that reads a raw housing dataset, imputes missing values, engineers square-footage features, and outputs an interactive Plotly dashboard.

You can write complex SQL queries with window functions and Common Table Expressions, and connect Python directly to relational databases.

1. SQL fundamentals: SELECT, JOIN, GROUP BY, subqueries5 hrs
   • INNER, LEFT, RIGHT, and FULL OUTER JOINs
   • GROUP BY aggregation with HAVING clauses
   • correlated and uncorrelated subqueries
2. Window functions and CTEs5 hrs
   • ROW_NUMBER(), RANK(), and DENSE_RANK()
   • LAG() and LEAD() for time-series row comparison
   • WITH clauses for recursive and modular queries
3. Working with PostgreSQL and MySQL5 hrs
   • creating schemas, tables, and foreign keys
   • indexing strategies for query performance
   • EXPLAIN ANALYZE execution plan reading
4. Connecting Python to databases (SQLAlchemy, psycopg2)4 hrs
   • connection pooling with psycopg2
   • declarative base models in SQLAlchemy
   • reading SQL queries directly into Pandas DataFrames
5. NoSQL basics: MongoDB overview for data scientists5 hrs
   • BSON documents and collection creation
   • PyMongo find and aggregate pipelines
   • unflattener patterns for nested JSON logs

Hands-on project: Write an analytical SQL script using PostgreSQL window functions to rank monthly customer transactions, and connect it to a Python script via SQLAlchemy.

You can train supervised and unsupervised machine learning models, handle the bias-variance trade-off, and evaluate models using ROC-AUC curves.

1. Supervised learning: linear regression, logistic regression7 hrs
   • Ordinary Least Squares gradient descent
   • cost functions and sigmoid activation
   • odds ratios and decision boundaries
2. Classification: decision trees, KNN and SVMs7 hrs
   • Gini impurity and information gain in decision trees
   • Euclidean distance metrics in K-Nearest Neighbors
   • kernel trick and margin maximisation in SVMs
3. Unsupervised learning: K-means, DBSCAN and PCA6 hrs
   • K-Means centroid initialization and elbow method
   • density-based clustering with DBSCAN
   • dimensionality reduction via Principal Component Analysis
4. Bias-variance trade-off, cross-validation and regularization6 hrs
   • overfitting and underfitting diagnosis
   • k-fold cross-validation strategies
   • L1 Lasso and L2 Ridge regularisation penalties
5. Model evaluation: accuracy, precision, recall, ROC-AUC6 hrs
   • confusion matrix interpretation
   • precision-recall tradeoffs at varying thresholds
   • ROC curve plotting and AUC calculation

Hands-on project: Train a logistic regression and a support vector classifier on customer churn data, tune regularisation parameters, and generate precision-recall evaluation curves.

You can build gradient boosted ensemble models, compute SHAP feature importances, and perform Bayesian hyperparameter tuning.

1. Ensemble methods: Random Forest, XGBoost, LightGBM, CatBoost6 hrs
   • bagging vs boosting architectural differences
   • gradient boosting tree construction
   • histogram-based acceleration in LightGBM
2. Feature importance and SHAP values5 hrs
   • permutation feature importance calculation
   • Shapley additive explanations theory
   • summary and waterfall plots for model debugging
3. Hyperparameter tuning: Grid Search and Bayesian optimization5 hrs
   • GridSearchCV and RandomizedSearchCV
   • Bayesian optimization loops with Optuna
   • preventing data leakage during hyperparameter search
4. Pipelines and feature engineering at scale4 hrs
   • Scikit-learn Pipeline and ColumnTransformer objects
   • custom transformer classes implementation
   • preserving transformer state across train-test splits
5. Kaggle competition strategy and winning approaches4 hrs
   • exploratory data analysis for tabular competitions
   • stacking and blending multiple model predictions
   • handling out-of-fold predictions safely

Hands-on project: Build an XGBoost and LightGBM ensemble pipeline for a Kaggle competition dataset, compute SHAP values for model explainability, and optimize hyperparameters with Optuna.

You can construct deep neural networks, train convolutional neural networks for image processing, and fine-tune transformer models.

1. Neural networks, backpropagation and activation functions5 hrs
   • forward propagation and matrix weight updates
   • ReLU, sigmoid, and softmax activation functions
   • vanishing gradients and weight initialization
2. CNNs for image data and computer vision tasks5 hrs
   • convolutional filters, padding, and stride
   • max pooling and spatial dimension reduction
   • PyTorch Dataset and DataLoader loaders for images
3. RNNs and LSTMs for sequential data5 hrs
   • recurrent neural network unrolling over time
   • LSTM forget, input, and output gating mechanisms
   • handling variable-length sequence padding
4. Transformers and attention: BERT and GPT for NLP5 hrs
   • scaled dot-product attention calculation
   • multi-head self-attention mechanisms
   • encoder-decoder transformer architecture
5. Transfer learning and fine-tuning pre-trained models4 hrs
   • freezing base model layers in PyTorch
   • attaching custom classification heads
   • learning rate warmup and cosine annealing schedulers

Hands-on project: Build and train a Convolutional Neural Network using PyTorch to classify image data, and implement a transfer learning pipeline using a pre-trained ResNet backbone.

You can preprocess text data, compute word embeddings, and build sentiment analysis pipelines using Hugging Face transformers.

1. Text preprocessing: tokenization, stemming, lemmatization5 hrs
   • regex cleaning and lowercasing strings
   • Porter stemming and WordNet lemmatization
   • subword tokenization with Byte-Pair Encoding
2. TF-IDF and word embeddings (Word2Vec, GloVe)5 hrs
   • Term Frequency-Inverse Document Frequency matrix creation
   • Word2Vec CBOW and Skip-gram architectures
   • cosine similarity calculation between vector embeddings
3. Sentiment analysis and text classification5 hrs
   • Naive Bayes baseline text classifiers
   • logistic regression on TF-IDF sparse matrices
   • evaluating classification metrics on unbalanced corpora
4. Named entity recognition and topic modelling5 hrs
   • Latent Dirichlet Allocation topic extraction
   • spaCy named entity recognition pipelines
   • extracting dates, organisations, and monetary amounts
5. Hugging Face pipelines for NLP tasks4 hrs
   • transformers AutoTokenizer and AutoModel classes
   • running zero-shot classification pipelines
   • saving and loading fine-tuned model weights

Hands-on project: Build an end-to-end customer review sentiment classifier using Hugging Face pipelines, BERT tokenizers, and PyTorch fine-tuning.

You can decompose time-series trends, fit ARIMA and Prophet forecasting models, and build sequential business forecasts.

1. Time-series decomposition: trend, seasonality and noise5 hrs
   • additive and multiplicative decomposition models
   • rolling averages and exponential smoothing
   • stationarity testing with Augmented Dickey-Fuller test
2. ARIMA, SARIMA and exponential smoothing5 hrs
   • Autoregressive and Moving Average components
   • identifying p, d, q parameters via ACF and PACF plots
   • seasonal ARIMA modelling with exogenous variables
3. LSTM-based forecasting5 hrs
   • windowing time-series arrays for neural inputs
   • building multivariate LSTM forecasting models
   • inverse scaling predicted values back to original units
4. Prophet for business forecasting5 hrs
   • specifying trend changepoints in Prophet
   • adding custom holiday and seasonality effects
   • generating uncertainty intervals for forecasts
5. Financial and IoT time-series projects4 hrs
   • stock price volatility analysis
   • sensor telemetry anomaly detection
   • backtesting rolling window forecast accuracy

Hands-on project: Build a retail sales forecasting pipeline using ARIMA and Facebook Prophet, evaluating root mean squared error on monthly financial data.

You can process large-scale datasets using PySpark, orchestrate workflows with Airflow, and store data in cloud object storage.

1. Big data concepts: Hadoop ecosystem overview5 hrs
   • HDFS distributed block storage architecture
   • MapReduce paradigm limitations
   • YARN resource management allocation
2. Apache Spark with PySpark for large-scale data processing5 hrs
   • SparkSession and Resilient Distributed Datasets
   • PySpark DataFrame transformations and actions
   • Catalyst optimizer and RDD lazy evaluation
3. Data pipelines and workflow orchestration with Airflow5 hrs
   • DAG definition and task dependency operators
   • Airflow scheduler execution and backfilling
   • XCom parameter passing between tasks
4. Cloud storage and compute: AWS S3, EC2 and GCP BigQuery5 hrs
   • AWS S3 bucket configuration and IAM policies
   • provisioning and accessing AWS EC2 instances
   • running SQL queries on massive datasets in BigQuery
5. Real-time streaming fundamentals with Kafka4 hrs
   • Kafka topics, producers, and consumers
   • consumer group offset management
   • streaming data ingestion into Pandas dataframes

Hands-on project: Build an Apache Spark PySpark data processing job that reads gigabytes of transaction logs from AWS S3, aggregates metrics, and runs on a cloud cluster.

You can build FastAPI prediction endpoints, containerize machine learning workflows with Docker, and track experiments using MLflow.

1. Building REST APIs for ML models with Flask and FastAPI5 hrs
   • FastAPI Pydantic request and response schemas
   • loading serialized pickle/joblib model artifacts
   • handling HTTP exceptions and status codes
2. Containerizing data science workflows with Docker5 hrs
   • writing multi-stage Dockerfiles for Python apps
   • managing environment variables and dependencies
   • building and running container images locally
3. CI/CD for machine learning pipelines5 hrs
   • GitHub Actions workflow configuration
   • automated linting with flake8 and pytest
   • automated container image building and pushing
4. Model monitoring, drift detection and retraining5 hrs
   • data drift and concept drift definitions
   • Population Stability Index calculation
   • triggering automated model retraining scripts
5. MLflow for experiment tracking and model registry4 hrs
   • logging hyperparameters, metrics, and artifacts
   • comparing run evaluations in MLflow UI
   • registering and staging model versions

Hands-on project: Containerize a trained machine learning model inside a multi-stage Dockerfile, expose it via a FastAPI inference endpoint, and log experiment runs to MLflow.

You can design an experiment that supports a causal claim, and know when you only have correlation.

1. Designing a valid experiment90 min
   • Randomisation unit
   • Power and sample size
   • Guardrail metrics
2. Analysing the result90 min
   • Effect size and intervals
   • Peeking and stopping rules
   • Multiple comparisons
3. When you cannot randomise90 min
   • Confounding
   • Difference-in-differences
   • Propensity matching
4. Reporting to decision makers90 min
   • Stating uncertainty
   • Practical significance
   • What would change the call

Hands-on project: Design an A/B test with a power calculation for a stated minimum detectable effect, analyse a supplied result set reporting the effect size and confidence interval, then estimate the same effect from observational data and explain which you would trust.

You can design, build, containerize, and deploy an end-to-end data science system to the cloud while clearing technical interview loops.

1. End-to-end data science capstone, deployed live6 hrs
   • integrating data ingestion, model training, and API serving
   • deploying containerized services to cloud VMs
   • smoke testing live endpoints and logging requests
2. Kaggle competitions and leaderboard practice5 hrs
   • submitting predictions to live Kaggle leaderboards
   • iterative feature engineering and cross-validation
   • analysing model failure cases and residuals
3. Portfolio, GitHub and Kaggle profile building4 hrs
   • writing professional README documentation
   • structuring repositories with clean directory layouts
   • showcasing data visualisations and metric notebooks
4. Mock interviews, statistics refresher and HR rounds5 hrs
   • answering foundational statistics and probability interview questions
   • explaining machine learning algorithms to non-technical stakeholders
   • behavioral interview preparation and salary negotiation
5. Tableau / Power BI dashboards for stakeholder presentation4 hrs
   • building executive summary slide decks
   • presenting model business impact and ROI metrics
   • designing interactive executive KPI dashboards

Hands-on project: Deploy a complete machine learning web service to a cloud virtual machine, build a professional GitHub portfolio repository, and complete technical mock interviews.