Imperial Data Science Intensive Course

Python for data science

Learn programming in Python, how to work with Jupyter Notebook and to use powerful Python libraries like Pandas and NumPy to explore and analyse big datasets. Collect data from various sources, including CSV files, SQL queries on relational databases, Google Big Query, APIs and webscraping.

Relational database and SQL

Learn how to formulate a good question and how to answer it by building the right SQL query. This module will cover schema architecture and then dive deep into the advanced manipulation of SELECT to extract useful information from a stand-alone database or using a SQL client software like DBeaver.

Data visualisation

Make your data analyses more visual and understandable by including data visualisations in your Notebook. Learn how to plot your data frames using Python libraries such as matplotlib and seaborn and transform your data into actionable insights.

Statistics, probability, linear algebra

Understand the underlying math behind all the libraries and models used in the course. Become comfortable with the basic concepts of statistics and probabilities (including mean, variance, random variable, Bayes’s Theorem, etc.) and with matrix computation, at the core of numerical operations in libraries like Pandas and NumPy.

Statistical inferences

You'll learn how to structure a Python repository with object-oriented programming in order to clean your code and make it re-usable, how to survive the data preparation phase of a vast dataset, and how to find and interpret meaningful statistical results based on multivariate regression models.

Communication

Data analysts are meant to communicate their findings to non-technical audiences: you will learn how to create impact by explaining your technical insights, and how to turn them into business decisions using cost/benefits analysis. You’ll be able to share your progress, present your results and compare your results to your teammates’.

Preprocessing and supervised learning

Learn how to explore, clean, and prepare your dataset through pre-processing techniques like vectorisation. Get familiar with the classic models of supervised learning – linear and logistic regressions. Learn how to solve prediction and classification tasks with the Python library scikit-learn using learning algorithms like KNN (k-nearest neighbors).

Generalisation and overfitting

Implement training and testing phases to make sure your model can be generalised to unseen data and deployed in production with predictable accuracy. Learn how to prevent overfitting using regularisation methods and how to choose the right loss function to improve your model's accuracy.

Performance metrics

Evaluate your model's performance by defining what to optimise and the right error metrics in order to assess your business impact. Improve your model's performance with validation methods such as cross validation or hyperparameter tuning. Finally, discover a powerful supervised learning method called SVM (Support Vector Machines).

Unsupervised learning and advanced methods

Move to unsupervised learning and implement methods like PCA for dimensionality reduction or clustering for discovering groups in a data set. Complete your toolbelt with ensemble method that combine other models to improve performance, such as Random Forest or Gradient Boosting.

Managing images and text data

Get comfortable with managing high-dimensional variables and transforming them into manageable input. Learn classic pre-processing techniques for images like normalisation, standardisation and whitening. Apply the right type of encodings to prepare your text data for different NLP tasks (Natural Language Processing).

Neural networks

Understand the architecture of neural networks (neurons, layers, stacks) and their parameters (activation functions, loss function, optimiser). Become autonomous to build your own networks like Convolutional NeuralNetworks (for images), Recurrent Neural Networks (for time-series) and Natural Language Processing networks (for text).

Deep learning with Keras

Discover a new library called Keras, which is a developer-friendly wrapper over tensorflow, a deep learning library created by Google. We'll teach you the fundamental techniques to build your first deep learning model with Keras.

Computer vision

Go further into computer vision with deep learning by building networks for object detection and recognition. Implement advanced techniques like data augmentation to augment your training set by computing image perturbations (random crops, intensity changes, etc) in order to improve your model's generalisation.

Machine learning pipeline

Move from Jupyter Notebook to a code editor and learn how to setup a machine learning project in the right way in order to quickly and confidently iterate. Learn how to convert a machine learning model into a model with a robust and scalable pipeline with sklearn-pipeline using encoders and transformers.

Machine learning workflow with MLflow

Building a machine learning model from start to finish requires a lot of data preparation, experimentation, iteration and tuning. We'll teach you how to do your feature engineering and hyperparameter tuning in order to build the best model. For this, we will leverage a library called MLflow.

Deploying to production with Google Cloud Platform

Finally, we'll show you how to deploy your code and model to production. Using Google Cloud AI Platform, you'll be able to train your model at scale, package it and make it available to the world. Cherry on top, you will use a Docker environment to deploy your own RESTful Flask API which could be plugged to any front-end interface.