Contents

Jupyter and QuantEcon

Peifan Wu (Postdoc @ Vancouver School of Economics, Intructor of ECON323)

About QuantEcon

QuantEcon is a nonprofit organization dedicated to improving economic modeling by enhancing computational tools for economists. In particular, it provides three sets of lectures:

  • QuantEcon with Python

    • Basic: Introductory

    • Advanced: Tools in (macro)economic Research

  • QuantEcon with Julia

  • QuantEcon DataScience

QuantEcon DataScience

QuantEcon DataScience presents lectures on (Python) programming, data science, and economics.

  • Consists of different parts

    • Python Fundamentals: Basic operations, collections, control flow, functions, etc.

    • Scientific Computing: Brief intro to Numpy and Scipy; Optimizations; Basic Applications in Economics

    • Pandas: Playing around with Pandas DataFrames

    • Applications: Case studies. More advanced techniques in visualization; More statistical methods and machine learning in Economics

  • Taught in various places

    • Utilize all the course contents: UBC, Huazhong University of Science and Technology

    • Partly used: 5+ schools around the world (NYU, PennState, Peking-HSBC Business School, NUS)

  • Good Resource for Self-study

Jupyter and QuantEcon DataScience Lectures

Running the Lectures with Jupyter

We offer several options to run the lectures:

  • Local Installation (not recommended) running the notebooks offline and locally

  • Running on the cloud, with NBGitPuller(https://github.com/jupyterhub/nbgitpuller) fetching the latest version of the notebooks

(Demo: Navigation Bar)

Advantages:

  • A uniform environment saves teaching time from software installation

  • In-class demonstrations can be easily shared and fully replicated by the students

Writing/Updating Lectures

We write the lectures with reStructuredText(rst) file format. For webpage formatting, we include custom environments (warning, exercise, etc).

Then the Jupinx tool converts rst files into a website via Jupyter Notebooks. The name of Jupinx comes from Jupyter + Sphinx.

  • Jupinx generates several versions of notebooks: original vs executed, website vs local ones

  • WYSIWYG

  • The source files are synchronized from Github repositories, and compilations are on AWS clusters.

For web-based data, we cache the data we need on AWS bucket.

  • Fetching data online depends on the stability of the API, and the Internet. Not ideal for teaching.

  • Some data are relatively large and takes time for the Jupyter notebook to download.

  • Easy and consistent to maintain.

Teaching

Instructor’s Experience

  • Code and execute in class, immediate feedback from the students

  • Students vary in programming background. Skip the hardest steps…

  • Explain in detail the meaning of the codes, and do variations

  • Remote teaching through Canvas is slick!

Students’ Feedback

(Mean 4.5/5.0, IM 4.7/5.0 – not bad!) The students love the comprehensive selection of the topics and the inter-discipline course material. The setting of this course is perfect for online teaching during COVID-19 as well.

Feedback and comments:

  • It touches too many topic without spending sufficient time on each of them

    • We aim at an introductory course that covers a wide range of topics

    • We don’t require previous knowledge on machine learning

  • Focus on the machine learning portion

    • This course is more about economics

    • Maybe machine learning in data science programs?

  • This class would greatly benefit from a flipped class format

    • Instead of going through the lectures one by one, we can put more course material for reading and give interactive coding sessions

    • More interactive options?

Final Project Showcase

The students are supposed to work on a final project at the end of the semester and we showcase their projects here with their permissions.

  • Resume building, advertising, cross-reference…

Effective pedagogies for teaching data science Interactive exploration with widgets and dashboards