Applied Optimization Techniques

Course goals

The goal of this course is to provide an introduction to simulation, optimization and machine learning techniques to students with a background in social sciences, with an approach biased towards practical work. The expected outcome is that students that have passed this course know a variety of modern and useful techniques that can be applied in real-life business contexts. With this knowledge and experience, the students understand what are the right techniques for different problems, which are the main steps and requirements to apply each of these techniques and how to judge the successful application of them.

Many of the techniques taught in this course are usually taught to engineering and technical profiles. This course does not aim to bring students to the same level of technical expertise as their engineering counterparts, but rather to provide enough background so that the students can successfully interact with such profiles. Having said that, this course can also be a first introduction for students that are willing to pursue a more thorough learning of the techniques discussed in the course, after or during itself.

With the knowledge and skills obtained in this course, students become fit for tasks such as:

Applying simulation, optimization and machine learning techniques to simple cases.
Planning and designing simulation, optimization and machine learning initiatives.
Leading simulation, optimization and machine learning projects from a managerial point of view.
Acting as a liaison between management and technical profiles in business contexts.

Pre-requisities

The course assumes the student has covered Mathematics I, II, III courses and the Probability & Statistics course. Passing this course is not impossible if that is not the case, but the student should expect a non-trivial challenge ahead.

Knowledge of the following topics will help students better leverage this course, but is not strictly required:

Basic programming, specially in data oriented languages such as Python or R.
Operations research

Teaching method and contents

The course will have lecture classes and practical seminars. Classes start on April 7th.

There will be 20 lecture classes and 6 practical seminars. Lecture classes will be used to present material to students as well as having discussions on the course contents. For the practical seminars, students will be divided into two groups with independent sessions to reduce the class size. The practical seminars will be used to deep-dive in the three mandatory case assignments that students will do throughout the course. The sesions will also be hands-on and students will work in the case together with the professor.

Students are expected to attend all the activities in the course. Beyond lectures and practical seminars, additional reading resources will be provided to students. For students that need to level up their Python skills, self-paced materials will be suggested.

Lectures will have the following contents:

Week	Classes	Student work
1	- L1: Introduction and motivation of the course - L2: Simulation, Optimization and Machine Learning in companies	- Python prep
2	- L3: Introduction to simulation: What is it, When do we use it, Types of simulation - L4: Simulation examples in Python. Introduction to case 1.	- Python prep - View Primer: Simulating a pandemic - Read Agent-based modeling: Methods and techniques for simulating human systems - Read case 1.
3	- L5: Simulation methodology. - L6: Simulation-based optimization I. Challenges and issues with simulation. Where to go from here - S1: Workshop for case 1	- Work on case 1 - Review HASH model market simulation - Review HASH warehouse simulation
4	- L7: Introduction to optimization - L8: Modeling optimization problems - S2: Workshop for case 1	- Work on case 1 - Read Gurobi's Modelling Basics - Read Neos taxonomy of optimization problems - View this video on the Simplex algorithm
5	- L9: Taxonomy of optimization techniques - L10: Simulation-based optimization II. Introduction to case 2	- Deliver case 1 - Read case 2 - Enjoy watching simulation-based race car training - Read how the 4th most popular database software in the world uses GAs to access data faster.
6	- L11: Challenges in real-world usage. Simulation vs Optimization - L12: Introduction to Machine Learning - S3: Workshop for case 2	- Work on case 2 - Read this review on simulation optimization techniques and softwares
7	- L13: Supervised Machine Learning (SML): NIPS - L14: Typical SML workflow. Introduction to case 3 - S4: Workshop for case 2	- Work on case 2 - Read case 3
8	- L15: Algorithm deep dive: Decision trees - L16: Feature Engineering and Model Evaluation - S5: Workshop for case 3	- Deliver case 2 - View this intro to neural networks and this intro to random forests
9	- L17: Deployment of Models - L18: Stories from the trenches: applying all of this in the real world - S6: Workshop for case 3	- Work on case 3 - View this video on why businesses fail at ML
10	- L19: Where to go from here: further learning and carreer advice - L20: Final Q&A, exam preparation	- Work on case 3
11	- Exam	- Deliver case 3

Lecture 1 INTRO
- Introduction to the course
  - Citizenship rules
    - Won't force you to come, but I advice you to.
    - I'll always try to start 5min late, finish 5min late, and stop for 5min.
    - You can come and go, just please be respectful.
  - Calendar
  - Contents
  - Expectations
  - The teacher
  - Evaluation
  - Contact
  - Questions?
  - The relevance of math and computers in management
    - Examples: pricing, logistics, staffing.
    - The skills and profiles required
    - The tools used
Lecture 2 INTRO
- The techniques we will see in the course
  - Simulation
  - Optimization
  - Supervised machine learning (aka "prediction")
- Why this stuff is important
Lecture 3 SIM
- A humbling example
- What is simulation and when do we use it
- Different types of simulations
Lecture 4 SIM
- Toy simulations in Python
- How to approach simulation in practical terms
- Tools in industry
Lecture 5 SIM
- Theoretical background on simulation
- Present case 1
Lecture 6 SIM
- Simulation-based optimization
- Where to go from here
Lecture 7 OPT
- What is optimization
- A trivial example
Lecture 8 OPT
- Different optimization techniques
- Present case 3
Lecture 9 OPT
- How to model optimization problems (target functions, decision variables and constraints)
Lecture 10 OPT
- Simulation-based optimization: Genetic algorithms
Lecture 11 OPT
- Real world challenges and optimization deployment
Lecture 12 ML
- Good news, you already know Machine Learning
- Different branches of Machine Learning
- Real world examples of applications
Lecture 13 ML
- How does Supervised Machine Learning work?
- Present case 2
Lecture 14 ML
- The Machine Learning workflow (EDA, Feature Engineering, Model Evaluation, Deployment)
Lecture 15 ML
- Feature Engineering
Lecture 16 ML
- Model evaluation
Lecture 17 ML
- Deployment and real world challenges
Lecture 18 Real life stories from the trenches
Lecture 19 Real life stories from the trenches
Lecture 20
- Q&A pre-exam
- Feedback on the course

Case details

Case 1

Title: Choosing stock policies under uncertainty
Description: Students role-play their participation as consultants in a project for Beanie Limited, a coffee beans roasting company. Elisa, the regional manager for the italian region, is not happy with their inventory policies for raw beans. The students are asked to analyse the problems posed by Elisa and apply simulation techniques, together with real data, to recommend a stock policy for the company's warehouse in the italian region. Python notebooks with some helpful prepared functions are provided to the students. The final delivery is a report with their recommendation to the client company, along with the used code.

Case 2 candidate

Title: ?
Description: ?
Sample idea: https://www.gurobi.com/resource/facility-location-problem/

Case 3

Title: Improving last-mile scheduling with Machine Learning
Description: Students role-play their participation as consultants in a project for Beanie Limited, a coffee beans roasting company. Pieter, the director of secondary transportation, has requested help from the student consultants. One of the key activities in Pieter's team is the daily scheduling, where the different trucks get assigned which deliveries and routes will perform. The students are asked to develop a machine-learning algorithm to predict the drop-time for each delivery (the drop-time is the time a driver takes in unloading the goods in a a client location. More informally, the time that passes since he removes the key from the truck until he starts the engine again). The goal is to provide more advanced information for Pieter's schedulers so they can better plan the routes of their drivers. The students are asked to build and deliver a Machine Learning algorithm that predicts this time. The students will be provided a labelled dataset. The final delivery is the working prediction model, along with a report explaining their methodology in building it, and answering some business questions to the client company.

Grading

The following items compose the final grade:

Case assignments: 50% of the grade. There will be three assignments, each with the same weight. The average grade of the assignments must be of 5 or more to pass the course.
Final exam: 50% of the grade. There will be a final exam at the end of the course. The grade must be of 5 or more to pass the course.

Students who fail the final exam will get the chance to sit a retake exam.

A final grade is calculated as:

if avg(case1_grade, case2_grade, case3_grade) < 5:
    passed_course = False
if final_exam_grade < 5: 
    passed course = False

passed_course = True
final_grade = (avg(case1_grade, case2_grade, case3_grade) + final_exam_grade) / 2

Bibliography

All compulsory and required materials will be provided during the course. These include lecture notes, required readings and description readings.

A good book that follows the approach of this course is "Guttag, John. Introduction to Computation and Programming Using Python: With Application to Understanding Data. 2nd ed. MIT Press, 2016. ISBN: 9780262529624", used in the homonymous course at MIT. It is not compulsory to use this book, but some students might find it helpful.

Additional specific readings will be provided throughout the course. Students will be requested to read some of these materials in advance of some sessions.

For students that want to dive deeper in the topics covered in the course, the following books are recommended:

On simulation: Louis G. Birta Gilbert Arbez, Modelling and Simulation. Springer 2019 ISBN: 978-3-030-18869-6 or Law A., Kelton D., Simulation and Modelling Analysis, Second Edition, McGraw-Hill, ISBN: 978-0071165372
On machine learning: Hastie T., Tibshirani R., Friedman J., The Elements Of Statistical Learning: Data Mining, Inference, And Prediction, Second Edition ISBN: 978-0387848570

Cool ideas & notes

Hold a Kaggle competition with the students. Winners come to spend a morning in Accenture.
Start every lecture with a fun fact.
Let them choose the challenge for one of the practical labs.
Should I have office hours?
Will the classes be recorded?
What's are the policies on:
- Late deliveries
- Not attending exam
- Re-takes
https://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-0002-introduction-to-computational-thinking-and-data-science-fall-2016

16 KiB Raw Blame History