Summary: Resit 2017

Read the summary and the most important questions on Resit 2017

• 1 Introduction

• (4 pt) Provide the definition of the quantified self that has been discussed during the lecture.

  ”The quantified self is any individual engaged in the self-tracking of any kind of biological, physical, behavioral, or environmental information. The self- tracking is driven by a certain goal of the individual with a desire to act upon the collected information.”

• Steve is a student with serious health problems. He is obese due to a lack of movement and regularly has mental health problems (depression, anxiety). He is pretty addicted to his mobile phone which is equipped with all imaginable sensors.(5 pt) Identify a supervised machine learning task and an unsupervised machine learning task that could be useful for the case of Steve.

  An example of a supervised learning task could be the prediction of a state of depression or anxiety based on the sensor values measured. An unsupervised task could be to find clusters of locations where Steve suffers from anxiety attacks.

• Steve is a student with serious health problems. He is obese due to a lack of movement and regularly has mental health problems (depression, anxiety). He is pretty addicted to his mobile phone which is equipped with all imaginable sensors.(3 pt) List three measurements that could be collected for the case of Steve and argue their relevance based on one or both tasks you have identified above.

  Examples of measurements could be accelerometer data to measure how active Steve is (could help with prediction the state of depression), the heart rate (could predict an anxiety attack), or the GPS location (from clustering the locations for an anxiety attack)

• (3 pt) Provide the definition for reinforcement learning that has been treated during the lecture.

  ”Reinforcement learning tries to find optimal actions in a given situation so as to maximize a numerical reward that does not immediately come with the action but later in time.”

• 2 Feature Engineering

  This is a preview. There are 1 more flashcards available for chapter 2
  Show more cards here

• (4 pt) Apply a transformation in the time domain for the attribute Activity levelusing a mean aggregation function. Use a window size λ = 1. Provide the values for the newly created attribute.

  See table.

• (4 pt) Next to the time domain, which other domain is available for aggregation of temporal features? Discuss how features are created in that domain.

  The frequency domain. We apply a Fourier transformation to a selected historical window in our data (for each data point) and decompose the data within that window into frequencies with an accompanying amplitude. These form the basis to compute the features such as the power spectral entropy.

• (6 pt) Sometimes we have unstructured data. Describe the pipeline you can use to identify features from natural language.

  The pipeline contains four steps: (1) tokenization (identify sentences and words); (2) lower case; (3) stemming (map words to their stam), and (4) stop word removal.

• 3 Clustering

  This is a preview. There are 1 more flashcards available for chapter 3
  Show more cards here

• (3 pt) We are going to cluster on a person level using this data. Explain what a feature-based distance metric is on this person level.

  The feature-based distance metric tries to extract features from a set of measurements to summarize the measurements (e.g. the mean value). Distance between persons is defined by the difference in values for those features.

• (5 pt) Compute the distance using the Euclidean distance function on a person level (an example of a so-called raw-based person level distance metric). Show your calculations.

  We just pair the instances up and compute the Euclidean distance. For this we square the distance per instance, sum them over all instances, and
  
  take the square root. This results in \sqrt{0^2 + 20^2 + 0^2 + 20^2 + 0^2} = \sqrt{800} ≈ 28.28.

• (8 pt) Let us now use dynamic time warping as a distance metric. Fill in Table 4 by using the dynamic time warping algorithm. Use the absolute difference between the values as distance metric. Show the steps you used in the calculations.

  The filled in table that results is shown below (note that the order of the time points has been flipped for patient 2 to make it easier to apply the algorithm). Each position is calculated by considering the distance between the matched points and the cheapest path to get there. Note that you can only move up, to the right, or diagonal to the upper right.