This document provides instructions for hands-on exercises in data mining using two datasets: the mushroom dataset and MNIST dataset. For each dataset, participants are instructed to explore the data through visualization and understanding of features, and perform classification using various methods like Naive Bayes, decision trees, nearest neighbors, logistic regression, and Bayesian classification. Results are to be documented in a report with screenshots and observations.
This document provides instructions for hands-on exercises in data mining using two datasets: the mushroom dataset and MNIST dataset. For each dataset, participants are instructed to explore the data through visualization and understanding of features, and perform classification using various methods like Naive Bayes, decision trees, nearest neighbors, logistic regression, and Bayesian classification. Results are to be documented in a report with screenshots and observations.
This document provides instructions for hands-on exercises in data mining using two datasets: the mushroom dataset and MNIST dataset. For each dataset, participants are instructed to explore the data through visualization and understanding of features, and perform classification using various methods like Naive Bayes, decision trees, nearest neighbors, logistic regression, and Bayesian classification. Results are to be documented in a report with screenshots and observations.
This document provides instructions for hands-on exercises in data mining using two datasets: the mushroom dataset and MNIST dataset. For each dataset, participants are instructed to explore the data through visualization and understanding of features, and perform classification using various methods like Naive Bayes, decision trees, nearest neighbors, logistic regression, and Bayesian classification. Results are to be documented in a report with screenshots and observations.
Instructions Use ANY Software you are comfortable with R, MATLAB, FreeMat, SAS, etc.
We will focus on TWO datasets:
Mushroom dataset MNIST dataset
We will focus on TWO areas
EXPLORATION Data visualization and understanding CLASSIFICATION With various classification methods
What you need to submit:
Create a document where you can just put pictures and tables and brief observations of what you learnt from each of those. Each Slide explains step by step what to try, what to observe, and what to report in the final report. Mention the slide title in each section of your report.
Mushroom Exploration Some of the values are missing. Substitute the most common value of that feature for missing.
Summarize the following for Mushroom data:
Number of FEATURES: Number of DATA POINTS: Number of CLASSES: PRIOR probability of each class: Draw the histogram of for each feature:
Information Gain of each FEATURE
Compute the information gain of each feature List all the features in descending sort order of this value.
Mushroom Naive Bayes classifier
Partition the Mushroom dataset into 40% testing and 60% training data. Make sure you get proportional points from each class while sampling (or just do random sampling). Build a Nave Bayes classifier using the training data Evaluate the Nave Bayes classifier on test data. Now build the NB classifier with top k features based on the information gain (k = 5, k = 10, k = all) Report the test accuracy on each of these classifiers.
Mushroom Decision Tree classifier
Using the same training test split as before. Learn a Decision Tree Classifier using any of the tools. Build a HIGH complexity and a LOW complexity decision tree. Complexity could be in terms of maximum depth or number of leaf nodes, etc. (depending on the parameters allowed in your tool). Report accuracy on the high and low complexity decision trees. Also define how you created the high and low complexity DTs.
Mushroom Nearest Neighbor
Define similarity between two data points as the fraction of features that match. Using the training test split built above, report the test accuracy with k = 1, 3, 5, and 7 nearest neighbor classifiers.
MNIST Explore Fisher Projection of the entire data: For the entire data, compute the Fisher projection. Sample 50 points of class 3, 5, and 8. Plot those in the top two Fisher projections. Color code the points with their class label. Repeat above exercise for classes 1, 7, and 9.
PCA projection of the entire data:
Compute the top two PCA projections for the data. Using the same 50 points of class 3, 5, 8 as above. Plot these points in top two PCA projections. Color code the points by their class labels. Repeat the above exercise for classes 1, 7, and 9.
MNIST Logistic Regression
Partition each class data into 60% training and 40% testing Build the following classifiers for each PAIR of 10 classes Logistic Regression model with all the 784 features. Logistic Regression model with Top 9 PCA features. Logistic Regression model with Top 9 Fisher features.
Take the average of the 45 classifiers in each case.
Report average accuracy w.r.t. different way of projection.
MNIST K-Nearest Neighbor
Sample 50 examples in each class as TRAINING data. Sample 50 examples in each class as TEST data. Build the k-Nearest neighbor classifier for k = 1, 3, 5, 7. Do this in Original Space, PCA(9) space, and Fisher(9) space. Compare the performance in the table with k values on one side and transformation on the other (no transformation, PCA(9) and Fisher(9)). NOTE: PCA(9) means top 9 PCA components.
MNIST Bayesian Classifier
Partition each class into 40% testing and 60% training. Build a Bayesian classifier for each pair of classes Assume single full covariance matrix Gaussian for each class. Do this over 784 dimensional full data. Do this over PCA(9) dimensions. Do this over Fisher(9) dimensions.
Take average of 45 classifiers for each method.
Report accuracies for the three feature projection methods.
Full Download (Ebook) Independent Random Sampling Methods by Luca Martino, David Luengo, Joaquín Míguez ISBN 9783319726335, 9783319726342, 3319726331, 331972634X PDF DOCX
Full Download (Ebook) Independent Random Sampling Methods by Luca Martino, David Luengo, Joaquín Míguez ISBN 9783319726335, 9783319726342, 3319726331, 331972634X PDF DOCX
