Strumenti Utente

Strumenti Sito


dm:start

Data Mining A.A. 2021/22

DM1 - Data Mining: Foundations (6 CFU)

Instructors:

Teaching Assistant

DM2 - Data Mining: Advanced Topics and Applications (6 CFU)

Instructors:

Teaching Assistant

News

  • [24.02.2022] Project Groups link
  • [28.04.2022] The exams are held “in person”, compatibly with the availability of adequate spaces. For particular categories of students (students with disabilities and international or Erasmus students), the exams ensured remotely upon request presented by the student when completing the registration form for the exam or, after the closing of the registration deadline, by filling in the appropriate form available at the following link: http://su.unipi.it/DichiarazioneEsameRemoto.
  • [05.05.2022] Rules for DM2 exam available here.
  • [26.05.2022] It is now possible to book slots for oral exams here. The exams will be in Aula X1.

Learning Goals

  • DM1
    • Fundamental concepts of data knowledge and discovery.
    • Data understanding
    • Data preparation
    • Clustering
    • Classification
    • Pattern Mining and Association Rules
    • Clustering
  • DM2
    • Outlier Detection
    • Dimensionality Reduction
    • Regression
    • Advanced Classification
    • Time Series Analysis
    • Sequential Pattern Mining
    • Advanced Clustering
    • Transactional Clustering
    • Ethical Issues

Hours and Rooms

DM1

Classes

Day of Week Hour Room
Monday 11:00 - 13:00 Aula C / MS Teams
Thursday 11:00 - 13:00 Aula A1 / MS Teams

Office hours - Ricevimento:

  • Prof. Pedreschi: Monday 16:00 - 18:00, Online
  • Prof. Nanni: appointment by email, Online

DM 2

Classes

Day of Week Hour Room
Monday 11:00 - 13:00 MS Teams
Thursday 11:00 - 13:00 MS Teams

Office Hours - Ricevimento:

  • Room 268 Dept. of Computer Science
  • Tuesday: 15-17, Room: MS Teams
  • Appointment by email

Learning Material -- Materiale didattico

Textbook -- Libro di Testo

  • Pang-Ning Tan, Michael Steinbach, Vipin Kumar. Introduction to Data Mining. Addison Wesley, ISBN 0-321-32136-7, 2006
  • Berthold, M.R., Borgelt, C., Höppner, F., Klawonn, F. GUIDE TO INTELLIGENT DATA ANALYSIS. Springer Verlag, 1st Edition., 2010. ISBN 978-1-84882-259-7
  • Laura Igual et al. Introduction to Data Science: A Python Approach to Concepts, Techniques and Applications. 1st ed. 2017 Edition.

Slides

Software

  • Python - Anaconda (3.7 version!!!): Anaconda is the leading open data science platform powered by Python. Download page (the following libraries are already included)
  • Scikit-learn: python library with tools for data mining and data analysis Documentation page
  • Pandas: pandas is an open source, BSD-licensed library providing high-performance, easy-to-use data structures and data analysis tools for the Python programming language. Documentation page
  • KNIME The Konstanz Information Miner. Download page
  • WEKA Data Mining Software in JAVA. University of Waikato, New Zealand Download page
  • Didactic Data Mining DDM

Class Calendar (2021/2022)

First Semester (DM1 - Data Mining: Foundations)

Day Room Topic Learning material Recording Instructor
1. 16.09.2021 11:00-12:45 Aula Fib A1 Introduction. Introducing DM1 Project-work guidelines (updated 22.11.2021) Lecture 1 Pedreschi
2. 20.09.2021 11:00-12:45 Aula Fib C Course overview Overview of contents Lecture 2 Pedreschi
3. 23.09.2021 11:00-12:45 Aula Fib A1 Data Understanding Slides Lecture 3 Pedreschi
4. 27.09.2021 11:00-12:45 Aula Fib C Data Preparation Slides Lecture 4 Pedreschi
5. 30.09.2021 11:00-12:45 Aula Fib A1 Lab: Data Understanding & Preparation – Python Python Introduction Dataset: Iris Hands-On Python (Iris) Lecture 5 Citraro
6. 04.10.2021 11:00-12:45 Aula Fib C Lab: Data Understanding & Preparation – Python (cont.) & KNIME Dataset: Titanic Hands-On Python (Titanic), Titanic DU+DP (complete) KMIME: Intro, KNIME DU+DP Lecture 6 Citraro
7. 07.10.2021 11:00-12:45 Aula Fib A1 Clustering: Intro & K-means Clustering intro and k-means [revised version] Lecture 7 Nanni
11.10.2021 11:00-12:45 Aula Fib C
8. 14.10.2021 11:00-12:45 Aula Fib A1 Clustering: k-means Lecture 8 Nanni
18.10.2021 11:00-12:45 Aula Fib C
9. 21.10.2021 11:00-12:45 Aula Fib A1 Clustering: Hierarchical methods Clustering: Hierarchical Methods Lecture 9 Nanni
10. 25.10.2021 11:00-12:45 Aula Fib C Clustering: density-base methods & exercises Clustering: Density-based methods Lecture 10 Nanni
11. 28.10.2021 11:00-12:45 Aula Fib A1 Lab: Clustering Python Hands-On Clust. (Iris) Python Titanic Knime Lecture 11 Citraro
12. 04.11.2021 11:00-12:45 Aula Fib A1 Classification: intro and decision trees Classification and decision trees (updated 11.11.2021) Lecture 12 Nanni
13. 08.11.2021 11:00-12:45 Aula Fib C Classification: decision trees/2 Lecture 13 Nanni
14. 11.11.2021 11:00-12:45 Aula Fib A1 Classification: decision trees/3 Lecture 14 Nanni
15. 15.11.2021 11:00-12:45 Aula Fib C Classification: decision trees/4 Lecture 15 Nanni
16. 18.11.2021 11:00-12:45 Aula Fib A1 Classification: decision trees exercises Exercise Lecture 16 Nanni
17. 22.11.2021 11:00-12:45 Aula Fib C Lab:Classification knime_classification Hands_on_Python_Titanic Python_Iris Online: TBD Lecture 17 (offline) Citraro
18. 25.11.2021 11:00-12:45 Aula A1 Pattern Mining - 1 Slides Lecture 18 Pedreschi
19. 29.11.2021 11:00-12:45 Aula C Pattern Mining - 2 Lecture 19 Pedreschi
20. 02.12.2021 11:00-12:45 Aula A1 Lab: Pattern Mining Apriori Exercise Hands_on_Python_Titanic KNIME Lecture 20 Citraro

Second Semester (DM2 - Data Mining: Advanced Topics and Applications)

Day Room Teams Topic Learning material Instructor Recordings
01. 14.02.2022 11:00–13:00 C Introduction, CRIPS, Evaluation, KNN Intro, CRISP, Eval, KNN, Notebbok_KNN_Eval Guidotti link
02. 17.02.2022 11:00–13:00 A1 Imbalanced Learning, Evaluation ImbLearn Eval, ImbLearn Guidotti link
03. 21.02.2022 11:00–13:00 C Dimensionality Reduction DimRed, Notebook_DimRed Guidotti link
04. 24.02.2022 11:00–13:00 A1 Outlier Detection (part 1) Outlier Detection, Notebook_OutlierDetection Guidotti link
05. 28.02.2022 11:00–13:00 C Outlier Detection (part 2) Outlier Detection, Notebook_OutlierDetection Guidotti link
06. 03.03.2022 11:00–13:00 A1 Outlier Detection (part 3) Outlier Detection, Notebook_OutlierDetection Guidotti link
07. 07.03.2022 11:00–13:00 C Naive Bayes Classifier, Linear Regression NBC , Notebook_NBC, LinReg Guidotti link
08. 10.03.2022 11:00–13:00 A1 Linear Regression, Gradient Descent, Maximum Likelihood Estimation, Odds LinReg, GradDes, MLE, Odds Guidotti link
09. 14.03.2022 11:00–13:00 C Logistic Regression, Support Vector Machines LogReg, SVM, Notebook_LR, Notebook_SVM Guidotti link
10. 17.03.2022 11:00–13:00 A1 Linear and Logistic Perceptron Perceptron Guidotti link1, link2
11. 21.03.2022 11:00–13:00 C Neural Networks NeuralNetwork, Notebook_NN, Notebook_NN_impl Guidotti link
12. 24.03.2022 11:00–13:00 A1 Ensemble Classifiers, Bagging, Random Forest EnsembleClassifiers, Notebook_ENS Guidotti link
13. 28.03.2022 11:00–13:00 C Boosting, Gradient Boost GBM Guidotti link
14. 31.03.2022 11:00–13:00 A1 XGBoost, LightGBM GBM, Notebook_GBM Guidotti link
15. 04.04.2022 11:00–13:00 C Time Series Introduction, Distance Functions TS_Intro_Distances, Notebook_TS_Sim, Notebook_TS_DTW_Impl, Notebook_TS_DTW_Constr_Impl Guidotti link
16. 07.04.2022 11:00–13:00 A1 Time Series Approximations, Clustering TS_Approx_Clustering, Notebook_TS_ApproxClus Guidotti link
17. 11.04.2022 11:00–13:00 C Time Series Motifs, Discord, Matrix Profile TS_MatrixProfile, TS_MatrixProfile Guidotti link
18. 14.04.2022 11:00–13:00 A1 Time Series Classification TS_Classification Notebook_TSC, Notebook_TSC_SoA Guidotti link
19. 21.04.2022 11:00–13:00 A1 Sequential Pattern Mining SPM Guidotti link
20. 28.04.2022 11:00–13:00 A1 Sequential Pattern Mining SPM, Notebook_SPM Guidotti link
21. 02.05.2022 11:00–13:00 C Advanced Clustering Approaches Advanced_Clustering, Notebook_AC Guidotti link
22. 05.05.2022 11:00–13:00 A1 Transactional Clustering Transactional Clustering, Notebook_TC Guidotti link
23. 09.05.2022 11:00–13:00 C Explainable Artificial Intelligence Explainability, Notebook_XAI Guidotti link
24. 12.05.2022 11:00–13:00 A1 Explainable Artificial Intelligence Explainability, Notebook_XAI Guidotti link

Exams

Exam DM1

The exam is composed of two parts:

  • An oral exam , that includes: (1) discussing the project report; (2) discussing topics presented during the classes, including the theory and practical exercises.
  • A project, that consists in exercises requiring the use of data mining tools for analysis of data. Exercises include: data understanding, clustering analysis, frequent pattern mining, and classification (guidelines will be provided for more details). The project has to be performed by min 3, max 4 people. It has to be performed by using Knime, Python or a combination of them. The results of the different tasks must be reported in a unique paper. The total length of this paper must be max 20 pages of text including figures. The paper must be emailed to datamining [dot] unipi [at] gmail [dot] com. Please, use “[DM1 2021-2022] Project” in the subject.

Project 1

  1. Assigned: 30/09/2021
  2. MidTerm Deadline: 21/11/2021 (half project required, i.e., Data understanding & Preparation and at least 2 clustering algorithms)
  3. Final Deadline: 14/01/2022 (complete project required)
  4. Data: choose between Glasgow Norms, Seismic Bumps

Project 2

  1. Assigned: After Project 1 Final Deadline
  2. Deadline: one week before the oral exam

Exam DM part II (DMA)

Exam Rules

  • Rules for DM2 exam available here.

Exam Booking Periods

  • 3rd Appello: 08/05/2022 00:00 - 05/06/2022 23:59
  • 4th Appello: 29/05/2022 00:00 - 26/06/2022 23:59
  • 5th Appello: 19/06/2022 00:00 - 17/07/2022 23:59

Exam Booking Agenda

  • Agenda Link: here
  • 3rd Appello: starts 07/06/2022
  • 4th Appello: starts 28/06/2022
  • 5th Appello: starts 19/07/2022
  • Important! if you book in the agenda in data in days between 07/06/2022 and 27/06/2022 you MUST be registered for the 3rd appello, if you book in the agenda in data in days between 28/06/2022 and 18/07/2022 you must be registered for the 4th appello, if you book in the agenda in data in days after 19/07/2022 you must be registered for the 5th appello.

For online exams the camera must remain open and you must be able to share your screen. For the online exams could be required the usage of the Miro platform (https://miro.com/app/dashboard/).

The exam is composed of two parts:

  • A project, that consists in employing the methods and algorithms presented during the classes for solving exercises on a given dataset. The project has to be realized by max 3 people. The results of the different tasks must be reported in a unique paper. The total length of this paper must be max 30 pages (suggested 25) of text including figures + 1 cover page (minimum font 11, minimum interline 1). The project must be delivered at least 7 days before the oral exam. The project must be delivered to riccardo [dot] guidotti [at] unipi [dot] it AND francesco [dot] spinnato [at] sns [dot] it with subject “[DM2 Project]”
  • An oral exam, that includes: (1) discussing topics presented during the classes, including the theory of the parts already covered by the written exam; (2) resolving simple exercises using the Miro platform; (3) discussing the project report with a group presentation;
  • Dataset: the data is about Human Activity Recognition
    • Data can be downloaded here
    • Submission Draft 1: 20/04/2022 23:59 Italian Time (we expect Modules 1 and 2)
    • Submission Draft 2: 20/05/2022 23:59 Italian Time (we expect Modules 1, 2 and 3)
    • Final Submission: one week before the oral exam.

Project Guidelines

  • Module 1 - Imbalanced Learning, Dimensionality Reduction, Anomaly Detection
    1. Explore and prepare the dataset. You are allowed to take inspiration from existing notebooks you can find online and figure out your personal research perspective (from choosing a subset of variables to the class to predict…). You are welcome in creating new variables and performing all the pre-processing steps the dataset needs.
    2. Define one or more (simple) classification tasks and solve them with Decision Tree and KNN.
    3. Identify the top 1% outliers: adopt at least three different methods from different families (e.g., density-based, angle-based… ) and compare the results. Deal with the outliers by removing them from the dataset or by treating the anomalous variables as missing values and employing replacement techniques. In this second case, you should check that the outliers are not outliers anymore. Justify your choices in every step.
    4. Analyze the value distribution of the class to predict with respect to point 2; if it is unbalanced leave it as it is, otherwise turns the dataset into an imbalanced version (e.g., 96% - 4%, for binary classification). Then solve the classification task using the Decision Tree or the KNN by adopting various techniques of imbalanced learning.
    5. Exploit and tests different dimensionality reduction techniques for (i) visualization in two dimensions, (ii) improve classification performance, (iii) improve outlier detection.
    6. Draw your conclusions about the techniques adopted in this analysis.
  • Module 2 - Advanced Classification Methods
    1. Solve the classification task defined in Module 1 (or define new ones) with the other classification methods analyzed during the course: Naive Bayes Classifier, Logistic Regression, Support Vector Machines, Neural Networks, Ensemble Methods, Gradient Boosting Machines and evaluate each classifier with the techniques presented in DM1 (accuracy, precision, recall, F1-score, ROC curve). Perform hyper-parameter tuning phases and justify your choices.
    2. Besides the numerical evaluation draw your conclusions about the various classifiers, e.g. for Neural Networks: what are the parameter sets or the convergence criteria which avoid overfitting? For Ensemble classifiers how the number of base models impact the classification performance? For any classifier which is the minimum amount of data required to guarantee an acceptable level of performance? Is this level the same for any classifier? What is revealing the feature importance of Random Forests?
    3. Select two continuous attributes, define a simple linear univariate regression problem and try to solve it using different techniques reporting various evaluation measures. Plot the two-dimensional dataset. Then generalize to multiple linear regression and observe how the performance varies. Solve it using linear regressions, regularized linear regressions (such as Lasso and Ridge) but also machine learning approaches such as Gradient Boosting Machines.
  • Module 3 - Time Series Analysis
    1. Prepare a dataset on which you can run time series clustering; motif/anomaly discovery and classification.
    2. On the dataset created, compute classification with KNN based on Euclidean/Manhattan and DTW distances and compare the results.
    3. To perform the clustering you can choose among different distance functions and clustering algorithms. Remember that you can reduce the dimensionality through time series approximation. Analyze the clusters and highlight similarities and differences.
    4. Analyze the dataset for finding motifs and/or anomalies. Visualize and discuss them and their relationship with other features.
    5. Solve the classification task on the time series dataset(s) and evaluate each result. In particular, you should use shapelet-based classifiers and structural-based classifiers. Analyze the shapelets retrieved and discuss if there are any similarities/differences with motifs and/or shapelets.
  • Module 4 - Sequential Patterns and Advanced Clustering
    1. Sequential Pattern Mining: Convert the time series into a discrete format (e.g., by using SAX) and extract the most frequent sequential patterns (of at least length 3/4) using different values of support, then discuss the most interesting sequences.
    2. Advanced Clustering: On a dataset already prepared for one of the previous tasks in Module 1 or Module 2, run at least one clustering algorithm presented in the advanced clustering lectures (e.g. X-Means, Bisecting K-Means, OPTICS). Discuss the results that you find analyzing the clusters and reporting external validation measures (e.g SSE, silhouette).
  • Module 5 - Explainability
    1. Try to use one or more explanation methods (e.g., TREPAN, LIME, LORE, SHAP, Counterfactual Explainers, etc.) to illustrate the reasons for the classification in one of the steps of the previous tasks.

N.B. When “solving the classification task”, remember, (i) to test, when needed, different criteria for the parameter estimation of the algorithms, and (ii) to evaluate the classifiers (e.g., Accuracy, F1, Lift Chart) in order to compare the results obtained with an imbalanced technique against those obtained from using the “original” dataset.

Exam Dates

Exam Sessions

Session Date Time Room Notes Marks
1.11.01.2022 14:00 - 18:00 MS Teams Please, use the system for registration: https://esami.unipi.it/
3.07.06.2022 Please, use the system for registration: https://esami.unipi.it/
4.28.06.2022 Please, use the system for registration: https://esami.unipi.it/
5.19.07.2022 Please, use the system for registration: https://esami.unipi.it/

Past Exams

  • Past exams texts can be found in old pages of the course. Please do not consider these exercises as a unique way of testing your knowledge. Exercises can be changed and updated every year and will be published together with the slides of the lectures.

Reading About the "Data Scientist" Job

… a new kind of professional has emerged, the data scientist, who combines the skills of software programmer, statistician and storyteller/artist to extract the nuggets of gold hidden under mountains of data. Hal Varian, Google’s chief economist, predicts that the job of statistician will become the “sexiest” around. Data, he explains, are widely available; what is scarce is the ability to extract wisdom from them.

Data, data everywhere. The Economist, Special Report on Big Data, Feb. 2010.

  • Data, data everywhere. The Economist, Feb. 2010 download
  • Data scientist: The hot new gig in tech, CNN & Fortune, Sept. 2011 link
  • Welcome to the yotta world. The Economist, Sept. 2011 download
  • Data Scientist: The Sexiest Job of the 21st Century. Harvard Business Review, Sept 2012 link
  • Il futuro è già scritto in Big Data. Il SOle 24 Ore, Sept 2012 link
  • Special issue of Crossroads - The ACM Magazine for Students - on Big Data Analytics download
  • Peter Sondergaard, Gartner, Says Big Data Creates Big Jobs. Oct 22, 2012: YouTube video
  • Towards Effective Decision-Making Through Data Visualization: Six World-Class Enterprises Show The Way. White paper at FusionCharts.com. download

Previous years

dm/start.txt · Ultima modifica: 26/05/2022 alle 12:09 (5 settimane fa) da Riccardo Guidotti