Starting from:
$30
Feature Extraction Project
Feature Extraction Project
Part I
Implement several feature extraction methods that reduce the dimensionality of the data from m to 2.
Input: two files: data and labels. The data is a comma separated matrix of size n × m. Here the data
points are the rows, not the columns. The “labels” is an array of size n. The labels can only take the
values 1,2,3.
Output: • A comma separated file containing the n × 2 matrix of the reduced data.
• A comma separated file containing the two vectors v1, v2 as a matrix of size 2 × m.
Programs: Read the files for the matrices Xt and the labels y. Use these to calculate two vectors v1, v2,
each of size m. Compute the projections of Xt on v1, v2 and use them to create the matrix D of size
n × 2.
Visualization: For small data you can visualize your results using the program plot.py.
Here are the programs that you are asked to implement:
1. pca1.py : use PCA without subtracting the mean.
2. pca2.py : use PCA with mean subtraction.
3. scatter1.py : minimize the within-class scatter.
4. scatter2.py : maximize the between-class scatter.
5. scatter3.py : maximize the ratio of between-class scatter and within-class scatter.
Try your programs on the following two datasets: irises, wine.
Part II
Feature extraction can be guided by various criteria. In this part our goal is to perform dimensionality
reduction for nearest neighbor applications. Specifically, training is performed with the input being the
same as in Part I. Then, given a test vector x of length m, it is reduced into two dimensions vx, vy. The
program searches for the vector in the training data that is nearest to vx, vy and returns its index.
You are asked to implement such programs for two cases. In the first case all items in the training data
are candidates for the nearest neighbor. In the second case the nearest neighbor must be of a specific label.
Programs
1. reducedim1.py : reduced dimension for the first case.
2. reducedim2.py : reduced dimension for the second case.
3. nearest.py : find nearest neighbor.
The arguments for reducedim1.py and reducedim2.py are the same as in Part I. The arguments for
nearest.py are as follows:
nearest.py reduced.txt V.txt labels.txt label
Here reduced.txt and V.txt are the output files of the dimensionality reduction programs. labels.txt is
the same as the input to the dimensionality reduction program. If Label is 1/2/3/ compute nearest neighbor
of the appropriate label. Otherwise ignore the label in computing the nearest neighbor.
The output of the program is the row of the nearest neighbor in the reduced.txt file.
What you need to submit
• Source code and documentation for the python scripts.
You must be available to demonstrate your program to the TAs. Time slots and additional instructions will
be announced later.
Deadline:
TBA.
Part I
Implement several feature extraction methods that reduce the dimensionality of the data from m to 2.
Input: two files: data and labels. The data is a comma separated matrix of size n × m. Here the data
points are the rows, not the columns. The “labels” is an array of size n. The labels can only take the
values 1,2,3.
Output: • A comma separated file containing the n × 2 matrix of the reduced data.
• A comma separated file containing the two vectors v1, v2 as a matrix of size 2 × m.
Programs: Read the files for the matrices Xt and the labels y. Use these to calculate two vectors v1, v2,
each of size m. Compute the projections of Xt on v1, v2 and use them to create the matrix D of size
n × 2.
Visualization: For small data you can visualize your results using the program plot.py.
Here are the programs that you are asked to implement:
1. pca1.py : use PCA without subtracting the mean.
2. pca2.py : use PCA with mean subtraction.
3. scatter1.py : minimize the within-class scatter.
4. scatter2.py : maximize the between-class scatter.
5. scatter3.py : maximize the ratio of between-class scatter and within-class scatter.
Try your programs on the following two datasets: irises, wine.
Part II
Feature extraction can be guided by various criteria. In this part our goal is to perform dimensionality
reduction for nearest neighbor applications. Specifically, training is performed with the input being the
same as in Part I. Then, given a test vector x of length m, it is reduced into two dimensions vx, vy. The
program searches for the vector in the training data that is nearest to vx, vy and returns its index.
You are asked to implement such programs for two cases. In the first case all items in the training data
are candidates for the nearest neighbor. In the second case the nearest neighbor must be of a specific label.
Programs
1. reducedim1.py : reduced dimension for the first case.
2. reducedim2.py : reduced dimension for the second case.
3. nearest.py : find nearest neighbor.
The arguments for reducedim1.py and reducedim2.py are the same as in Part I. The arguments for
nearest.py are as follows:
nearest.py reduced.txt V.txt labels.txt label
Here reduced.txt and V.txt are the output files of the dimensionality reduction programs. labels.txt is
the same as the input to the dimensionality reduction program. If Label is 1/2/3/ compute nearest neighbor
of the appropriate label. Otherwise ignore the label in computing the nearest neighbor.
The output of the program is the row of the nearest neighbor in the reduced.txt file.
What you need to submit
• Source code and documentation for the python scripts.
You must be available to demonstrate your program to the TAs. Time slots and additional instructions will
be announced later.
Deadline:
TBA.
1 file (189.0KB)
