Scribd
Upload
4 views

DS3-Lab5-v3

Uploaded by

Vijaylakshmi Bishnoi
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
4 views

DS3-Lab5-v3

Uploaded by

Vijaylakshmi Bishnoi
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Lab5: Data classification using Bayes Classifier with Gaussian Mixture Model (GMM);

Regression using Linear Regression and Polynomial Regression


Deadline for submission: 19th October 2021, 10:00 PM
PART-A:
You are given the Steel Plates Faults Data Set as a csv file (SteelPlateFaults-2class.csv) in
Assignment4 (Lab4). The dataset used for this assignment contains features extracted from the steel
plates of types A300 and A400 to predict whether the image of the surface of the steel plate contains
two types of faults such as Z_Scratch and K-Scratch. It consists of 1119 tuples each having 27 attributes
which are indicators representing the geometric shape of the fault. The last attribute (28th attribute) for
every tuple signifies the class label (0 for K_Scratch fault and 1 for Z_Scratch fault). It is a two-class
problem. Use the same train data file (SteelPlateFaults-train.csv) and test data file
(SteelPlateFaults-test.csv), used in Assignment4, in this assignment also.
1. Build a Bayes classifier with multi-modal Gaussian distribution (GMM) with Q Gaussian
components (modes) as class conditional density for each class on the training data
SteelPlateFaults-train.csv. Build a GMM with Q components for class1 and build
a GMM with Q components for class2. Classify every test tuple using the Bayes classifier with
GMM for the different values of Q=2, 4, 8, and 16. Perform the following analysis:
a. Find confusion matrix for each Q
b. Find the classification accuracy for each Q. Note the value of Q for which the accuracy
is high.
Note: Remove the attributes X_Minimum, Y_Minimum, TypeOfSteel_A300 and
TypeOfSteel_A400 from both training and test data set. Because correlation of X_Minimum &
X_Maximum is 1 and correlation of Y_Minimum & Y_Maximum is also 1. Also, correlation
between TypeOfSteel_A300 and TypeOfSteel_A400 is -1. This indicates the variance and
covariance of these pair of attributes are the same. This leads to the covariance matrix singular.
Due to this, the inverse of the covariance matrix cannot be computed, hence, the likelihood
cannot be computed. To avoid this issue, we need to remove the above-mentioned attributes.
Bayes classifier is now built using the data with only 23 attributes.
2. Tabulate and compare the best result of the KNN classifier, the best result of the KNN classifier
on normalized data, the result of the Bayes classifier using unimodal Gaussian density (all from
Assignment-4), and Bayes classifier using GMM.

Note:
Use the function “mixture.GaussianMixture” from scikit-learn to build GMM.

GMM = mixture.GaussianMixture(n_components=Q, covariance_type='full')


GMM.fit(x)

Compute the weighted log probabilities for each sample using


GMM.score_samples(x).

Compute accuracy using metrics.accuracy_score.


PART B:
You are given a data file abalone.csv. Abalones are marine snails. The dataset has been prepared
with the aim of making age predictions easier. Customarily, the age of abalone is determined by cutting
the shell through the cone, staining it, and counting the number of rings through a microscope. But it
is a tedious and time-consuming task. Therefore, other measurements, which are easier to obtain, are
used to predict age.
Attribute information:
Given is the attribute name, attribute type, the measurement unit, and a brief description. The number
of rings is the value to predict: either as a continuous value or as a classification problem.

Name / Data Type / Measurement Unit / Description

1. Length / continuous / mm / Longest shell measurement


2. Diameter / continuous / mm / Diameter of the shell calculated as perpendicular to length
3. Height / continuous / mm / Height of the shell with meat in shell
4. Whole weight / continuous / grams / Weight of whole abalone
5. Shucked weight / continuous / grams / Weight of meat
6. Viscera weight / continuous / grams / Gut-weight (after bleeding)
7. Shell weight / continuous / grams / Weight of the shell after being dried
8. Rings / integer / -- / Number of rings in a shell. (Adding 1.5 to the number of rings gives the
age of abalone in years)

Write a python program to split the data from abalone.csv into train data and test data. Train data
contain 70% of tuples and test data contain the remaining 30% of tuples. Save the train data as
abalone-train.csv and save the test data as abalone-test.csv.
Note: Use the command train_test_split from scikit-learn given below to split the data (keep
random_state=42 to get the same random values for every student).

1. Use the attribute which has the highest Pearson correlation coefficient with the target attribute
Rings as an input variable and build a simple linear (straight-line) regression model to predict
rings. (Prerequisite: calculate the Pearson correlation coefficient of every attribute with the target
attribute rings.)
a. Plot the best fit line on the training data where the x-axis represents the chosen attribute
value and the y-axis represents Rings.
b. Find the prediction accuracy on the training data using root mean squared error.
c. Find the prediction accuracy on the test data using root mean squared error.
d. Plot the scatter plot of actual Rings (x-axis) vs predicted Rings (y-axis) on the test data.
Draw inferences from the scatter plot.

2. Build a multivariate (multiple) linear regression model to predict Rings. All the attributes
other than the target attribute should be used as input to the model.
a. Find the prediction accuracy on the training data using root mean squared error.
b. Find the prediction accuracy on the test data using root mean squared error.
c. Plot the scatter plot of actual Rings (x-axis) vs predicted Rings (y-axis) on the test
data. Draw inferences from the scatter plot.
3. Use the attribute which has the highest Pearson correlation coefficient with the target attribute
Rings as input and build a simple nonlinear regression model using polynomial curve fitting
to predict Rings.
a. Find the prediction accuracy on the training data for the different values of degree of
the polynomial (p = 2, 3, 4, 5) using root mean squared error (RMSE). Plot the bar
graph of RMSE (y-axis) vs different values of degree of the polynomial (x-axis).
b. Find the prediction accuracy on the test data for the different values of degree of the
polynomial (p = 2, 3, 4, 5) using root mean squared error (RMSE). Plot the bar graph
of RMSE (y-axis) vs different values of degree of the polynomial (x-axis).
c. Plot the best fit curve using the best fit model on the training data where the x-axis
represents the chosen attribute value and the y-axis is Rings.
(Note: The best fit model is chosen based on the p-value for which the test RMSE is
minimum.)
d. Plot the scatter plot of the actual number of Rings (x-axis) vs the predicted number of
Rings (y-axis) on the test data for the best degree of the polynomial (p). Comment on
the scatter plot and compare it with that of in 1(d).

4. Build a multivariate nonlinear regression model using polynomial regression to predict Rings.
All the attributes other than the target attribute should be used as input to the model.
a. Find the prediction accuracy on the training data for the different values of degree of
the polynomial (p = 2, 3, 4, 5) using root mean squared error (RMSE). Plot the bar
graph of RMSE (y-axis) vs different values of degree of the polynomial (x-axis).
b. Find the prediction accuracy on the test data for the different values of degree of the
polynomial (p = 2, 3, 4, 5) using root mean squared error (RMSE). Plot the bar graph
of RMSE (y-axis) vs different values of degree of the polynomial (x-axis).
(Note: The best fit model is chosen based on the p-value for which the test RMSE is
minimum.)
c. Plot the scatter plot of the actual number of Rings (x-axis) vs the predicted number of
Rings (y-axis) on the test data for the best degree of the polynomial (p). Comment on
the scatter plot and compare it with that of in 1(d).

Hints and code snippets:


● For linear regression use:
from sklearn.linear_model import LinearRegression
reg = LinearRegression().fit(X, y)
#Input arguments: X: Input variable(s) of training data
and y: target values
y_pred = predict(X) #Predict using the linear model.
● For polynomial curve fitting and regression use:
from sklearn.preprocessing import PolynomialFeatures
poly_features = PolynomialFeatures(p) #p is the degree
x_poly = poly_features.fit_transform(X)
regressor = LinearRegression()
regressor.fit(x_poly, y)
#Input arguments: x_poly: Polynomial expansion of input
variable(s) of training data and y: target values
y_pred = regressor.predict(X)
Instructions:
● Your python program(s) should be well commented on. The comment section at
the beginning of the program(s) should include your name, registration number,
and mobile number.
● The python program(s) should be in the file extension .py
● The report should be strictly in PDF form. Write the report in word or latex form
and then convert it to PDF form. The template for the report (in word and latex) is
uploaded.
● The first page of your report must include your name, registration number, and
mobile number. Use the template of the report given in the assignment.
● Upload your program(s) and report in a single zip file. Give the name
as <roll_number>_Assignment5.zip. Example: b20001_Assignment5.zip
● Upload the zip file in the link corresponding to your group only.
In case the program is found to be copied from others, both the person who copied
and who helped to copy will get zero as a penalty.

You might also like