95-845, Applied Analytics: the Machine Learning Pipeline, will be taught in the Spring semester of 2018. Classes begin 1/18/18 and end 5/3/18. Spring break is observed the week of 3/12. Time: Fridays 1:30-4:20pm. Room: Hamburg Hall 1004
Jeremy C. Weiss, M.D./Ph.D., Assistant Professor of Health Informatics; jeremyweiss@cmu.edu
Office hours: Thursdays 9am, 2101F (E)
TA: Yoonjung Kim, PhD student in Information Systems; yoonjungkim@cmu.edu
Office hours: Tuesdays at 2pm, 2101E
Faculty assistant: Carole McCoy, HBH 2102
Machine learning is a highly valued set of analytics techniques, a confluence of ideas from computer science, statistics, economics, physics, and others. Machine learning is transforming fields with new capabilities, ways of understanding and visualizing data, and is becoming a key driver in decision making. However, knowing when (and how) to apply appropriate machine learning techniques requires understanding of data, machine learning, and the problem domain. This class seeks to teach students how to address the entire machine learning pipeline, starting from messy data and provisional questions and ending with actionable interpretations and insights.
The course will cover discovery, planning, analysis, and interpretation. Discovery involves understanding the data at hand, determining what is and is not answerable, and question generation. Planning involves contrasting the application of the desired machine learning method on ideal clean data with the messy data at hand. Dealing with representation, missing data, and designing appropriate machine learning machinery are all involved in planning. Analysis involves applying the machine learning method, checking model performance and assumptions in a principled and responsible manner. Interpretation involves the transformation of algorthm outputs into meaningful and actionable characterizations of the results. Each part of the pipeline is interconnected and students will learn to anticipate and address limitations through understanding of the pipeline as a whole.
Throughout the course we will focus on one vertical, health care, recognizing that the methods developed will generalize to others. We will work with real, messy, structured and unstructed data--including databases, text, and images. We will contrast machine learning methods against what is currently used in health care analytics, and describe the advantages and promise of each.
Students should have completed or be concurrently taking Data Mining, Machine Learning for Problem Solving, ML 17-601, ML 17-401 or the equivalent. Experience with R, Python or another programming language is required.
Grades will be based on:
All grades are tallied and at the end of the course they are scaled to meet the Heinz grading policy.
The project and that is submitted for grading is to be the work of the individual or team alone. Similarly, completed homework assignments is to be your work alone, although you are encouraged to discuss the problems with your classmates. Results that are identical or nearly identical across projects may be regarded as cheating. Penalties for cheating include lowering your grade including failing the course. In extreme cases, the instructors may recommend the termination of your enrollment at CMU.
Homework Policy: The lowest homework grade will be dropped. If the project grade is lower than any homework grade, all homeworks will be counted and the project grade will count for 10% less of the total grade.
Late Work Policy: You are expected to turn in all work on time (at the start of class on the due date). Assignments turned in within 48 hours of the deadline will be marked down 20% per day. Additional late assignments will not be accepted.
Wellness Policy: Take care of yourself and take care of others around you. There are resources to help you both in Heinz and around the University. The Counseling and Psychological Services (CaPS) help line is 412-268-2922. If the situation is life threatening, call the police.
Overview of machine learning
data wrangling and visualization
logistic regression
Bayesian networks
support vector machines
neural networks
partition-based methods
ensembling
dimensionality reduction;
prediction versus attribution
missing data
encoding domain expertise
observation versus intervention
algorithmic evaluation
bias-variance tradeoffs
causality
temporal modeling
relational learning
language modeling
There is not a required textbook. Readings will come from many sources and will be provided in Canvas and or in class. Useful references include Bishop's Pattern Recognition and Machine Learning, Murphy's Machine Learning: a Probabilistic Perspective, and James' et al's Introduction to Statistical Learning.
R, Rstudio, dplyr, purrr, ggplot, debug, Rmarkdown, Tensorflow; git; LaTeX