6.6 KiB
6.6 KiB
| fecha |
|---|
| 2022-06-12 |
Hoy he hecho unos fideos en la paella intermedia y han quedado buenisimos. Eli ha hecho un apaño cutre pero perfectamente funcional con unos cartones para pararme el viento y he cocinado de perlas. Me ha servido para quitarme la espina que tenía atascada en la cabeza de "menuda cagada has hecho comprando este armatoste".
Voy a mandarle un email al contacto de Esade que me ha pasado Tresa. Cosillas que me viene a la mente que debería preguntarle:
- Hay syllabus ya cerrado? Me gustaría verlo.
- Los materiales están preparados/cerrados, o se espera que el profesor los prepare?
- Hay un formato ya cerrado sobre cómo se evalúa a los alumnos, o hay que prepararlo?
- Qué cantidad de alumnos se matriculan?
- Cómo son las clases, solo lectures? Lectures + sesiones prácticas?
- Las clases son presenciales, en remoto o híbridas? Si hay parte presencial, donde se realizan?
- La persona a la que buscais se encargaria del 100% de la carga de la asignatura, o teneis en mente algun formato donde diferentes personas cubren diferentes areas de la asignatura?
Vamos a dejarnos querer...
Tengo que preparar el examen. Voy a pensar en qué cosas quiero evaluar: que conocimientos quiero que tengan. Abajo va mi primer intento a plasmarlo en un árbol.
- Simulation
- Knowledge
- What is simulation modeling?
- What are the advantages of a simulation?
- What are the drawbacks or limitations of a simulation?
- When does it make sense to use simulation to solve problems?
- Why is simulation useful when studying complex systems?
- What is the difference between a continuous and a discrete simulation?
- Imagine you have to model an uncertain, real world phenomenon like demand for a product, but you don't have historical data that you can rely on. How would you decide how to simulate it?
- What tools can be used to create a simulation?
- What is simulation based optimization. How is it different from other types of optimization?
- We discussed how random search is tipically a poor choice to run simulation based optimization. But still, it is very useful for one thing. What is it?
- What are heuristics and meta heuristics? How are they different? When would you use each?
- Explain roughly how could you apply a genetic algorithm to
- Applied
- You are working for an airport. The airport is designing a new terminal. The management needs to decide how many lines to put in the security control area. Management is concerned about the purchase and operation costs of the rather expensive machines that need to be bought and maintained for each line that gets placed. But at the same time, they must make sure that the waiting times for passengers are bearable, avoiding long queues that make passengers unhappy.
- Describe how simulation can help management in this situation
- What is the decision that needs to be made What are the goals? Is there a trade-off?
- What data would you ask for to design your simulation? How would you represent it?
- Describe (high level, no code) a heuristic to come to a good decision
- Knowledge
- Optimization
- Knowledge
- What is optimization?
- What are the advantages of optimization?
- What are the drawbacks or limitations of optimization?
- What are the different parts of an optimization problem? What is their role?
- How are optimization problems solved?
- What would you do if your optimization problem does not have a feasible solution?
- What is the issue with trade-off situations in optimization? Provide an example of how can trade-offs can be managed
- Why do we generally prefer to simplify problems into linear programs?
- Why are piece-wise functions useful in optimization?
- What is a solver? Describe three criteria you would use to pick one for solving a specific optimization problem.
- What is sensitivity analysis? Why is it useful?
- Applied
- You work for a manufacturing plant. The plant develops three families of chemical products in three different production lines: cleaning agents, fertilizer and battery electrolites. All three lines require sulfuric acid in different proportions to work and manufacture their chemicals.
- Each unit at each line produces the following revenue when sold:
- Cleaning agents: 10$/unit
- Fertilizers: 5$/unit
- Battery electrolites: 30$/unit
- Each unit at each line requires the following amount of sulfuric acid
- Cleaning agents: 0.4L/unit
- Fertilizers: 2L/unit
- Battery electrolites: 0.05L/unit
- During the next year, the plant can acquire up to 300L of sulfuric acid at 2$/L. Beyond that, the company can acquire up to 2000L at 20$/L.
- The company must produce at least 250 units of fertilizer per year due to national regulation of the industry.
- The plant manager wants to you design a model to make the largest possible profit for the plant.
- Design an optimization problem to solve this. You need to include your target function, decision variable and constraints. Please, use clear notation that allows to understand the model.
- Knowledge
- Machine Learning
- Knowledge
- How are supervised and unsupervised ML different?
- Why is data necessary for supervised ML?
- What is the difference between classification and regression problems?
- Why do we split data between train and test sets?
- What defines a good split when building a decision tree?
- Is letting a decision tree grow fully a good idea or not? Why?
- Why is accuracy typically not enough to measure performance in classification problems?
- How can you know if a ML model is overfitting?
- What would you do if your ML is underfitting?
- What is an ensemble method? Give two positive and two negative characteristics of ensemble methods compared to simple models.
- What is hyperparameter tuning?
- Applied
- You work for the Marketing department of a car manufacturing company. The company is interested in building a ML model to classify tweets as negative or not negative towards the brand. A negative tweet is a tweet where the customer expresses being unhappy about the companies services or products.
- You can assume that the company can obtain all the tweets where the brand is mentioned.
- Propose a basic methodology to build the data to solve the companies request. This should, at least, include:
- What data would you propose using.
- Propose at least 5 features that you would build out of the data to prepare the training and testing dataset.
- Decide on one performance metric and motivate your choice.
- Propose a baseline algorithm for the problem.
- You are pondering whether to build a simple decision tree classifier or go for a random forest classifier. Explain two advantages of each model type.
- Knowledge