Description
Description
Background
You now work for a prominent winery that has hired you to predict the quality of the wine they produce, based on alreadycollected data. The winery collects two main sets of data: one on the white wines they produce (winequality-white.csv,
n=4898), and one on the red wines they produce (winequality-red.csv, n=1599).
Data Description
Data are in two files: winequality-white.csv (4898 rows x 12 columns) and winequality-red.csv (1599 rows x 12 columns).
Input Variables
These input variables are based on physiochemical tests that occur regularly.
1. fixed acidity Range: 3.8 to 15.9
2. volatile acidity Range: 0.08 to 1.58
3. citric acid Range: 0 to 1.66
4. residual sugar Range: 0.9 to 65.8
5. chlorides Range: 0.009 to 0.611
6. free sulfur dioxide Range: 1 to 289
7. total sulfur dioxide Range: 6 to 440
8. density Range: 0.98711 to 1.03898
9. pH Range: 2.72 to 4.01
10. sulphates Range: 0.22 to 2.0
11. alcohol Range: 8 to 14.9
Output Variable
12. quality Range: 0 to 10
Part 1 – Unsupervised Learning
Coding and Analysis Requirements
Create a file called project4_clustering.py. Write a program that does the following:
1. Read winequality-white.csv and winequality-red.csv into two separate Pandas data frames.
a. Reference: https://pandas.pydata.org/docs/reference/api/pandas.read_csv.html#pandas.read_csv
2. Create a target_white data frame and a target_red data frame by selecting the data’s last column (the ‘quality’
column) and storing it there. For example: target_red = data_red[‘quality’]. Be sure to use the drop function
after you have copied it to remove it from the original data.
CSCI 5260 – Artificial Intelligence Page 2 | 4
a. See
https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.drop.html?highlight=drop#pandas.Dat
aFrame.drop
3. Using sklearn.cluster.KMeans, run the k-means clustering algorithm on the white wines and the red wines. You
should use 11 clusters because we know there are 11 quality metrics (labeled 0-10).
a. See https://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html.
b. Note that the result of the fit function returns a data structure containing the following:
i. cluster_centers_ndarray of shape (n_clusters, n_features)
1. Coordinates of cluster centers. If the algorithm stops before fully converging
(see tol and max_iter), these will not be consistent with labels_.
ii. labels_ndarray of shape (n_samples,)
1. Labels of each point
iii. inertia_float
1. Sum of squared distances of samples to their closest cluster center.
iv. n_iter_int
1. Number of iterations run.
4. Analyze the results for the white wine and the red wine examples. Add a discussion to the Project4.docx writeup
document. Remember that the cluster labels ARE NOT predictions of quality. The label is simply the grouping to
which an example belongs. To analyze this you should:
a. Write a procedure that determines the quality for each cluster by averaging the qualities of all items in that
cluster.
b. This is OPEN-ENDED but you should use this information to plot the quality values for each cluster. Include
these plots in your Project4.docx writeup.
c. Does the data indicate 11 clearly-defined quality metrics? Explain why it does or does not.
Part 2 – Supervised Learning
Coding and Analysis Requirements
Create a file called project4_ml.py. Using the same data set as above, do the following.
1. Combine the data sets into a single data set.
a. To do this, add a column called “type” to each data frame.
b. Set red wine as type 0 and white wine as type 1.
2. Split the data into train and test sets.
3. Train and Test two of the following learning algorithms from the scikit-learn library. Be sure to use the same train
and test data for each.
a. Decision Tree Classifier – https://scikit-learn.org/stable/modules/tree.html#classification
b. Linear Regression Classifier – https://scikit-learn.org/stable/modules/linear_model.html#generalizedlinear-regression
c. Gaussian Naïve Bayes Classifier – https://scikit-learn.org/stable/modules/naive_bayes.html#gaussiannaive-bayes
d. Nearest Neighbor Classifier – https://scikit-learn.org/stable/modules/neighbors.html#nearest-neighborsclassification
e. Support Vector Machine – https://scikit-learn.org/stable/modules/svm.html#classification
4. Analyze the results by showing the following (add your analysis to Project4.docx):
a. Which classification method performed better?
i. You should measure the number of true negatives, true positives, false negatives, and false
positives. If you want to drill down, it might be helpfult to track this by class.
b. Based on the results, what could you do to improve performance?
i. Keep in mind the ideas of feature engineering and feature scaling as you respond to this.
Part 3 – Deep Learning
Coding and Analysis Requirements
Create a file called project4_nn.py.
CSCI 5260 – Artificial Intelligence Page 3 | 4
1. Use the combined data set from Part 2, and the same train and test sets.
2. Train and test a Multilayer Perceptron Neural Network Classifier (MLPClassifer).
a. https://scikit-learn.org/stable/modules/neural_networks_supervised.html#classification
3. Analyze the results (recording the analysis in Project4.docx) by:
a. Showing true negatives, true positives, false negatives, and false positives. If you want to drill down, you
might track this by class to better analyze results.
b. Comparing the results to the models trained above.
Submission and Due Date
Interim Submission
An interim submission is required for this assignment to the Project 4 Interim dropbox. You should briefly detail the
progress you’ve made and note any problems or questions you have. Submit a screenshot of one portion of the code that
you have completed.
Failure to submit the interim submission ON TIME will result in the loss of 10% from the final Project 4 grade. Until you
have feedback on this dropbox, you will not be allowed to submit the final solution. This is to encourage you to work on
this early!
Project 4 interim submission is due to the D2L dropbox at or before Monday, April 12, 2021 at 11:59 PM
Final Submission
Submit all code and documentation, zipped into an archive: Surname_Project4.zip. The folder should be self-contained in
a way that allows the code to run. You can assume that I have all necessary libraries installed.
Your archive should contain the following files:
1. project4_clustering.py
2. project4_ml.py
3. project4_nn.py
4. winequality-red.csv
5. winequality-white.csv
6. The Project4.docx Word Document containing your analysis.
Project 4 is due to the D2L dropbox at or before Monday, April 19, 2021 at 11:59 PM
The rubric appears on the following page.
CSCI 5260 – Artificial Intelligence Page 4 | 4
Rubric
A letter grade will be assigned to each of the following, and will translate to a numeric grade based on the scale in the
syllabus, and averaged into an overall percentage. As a reminder, anything below a C translates to an F by University
Graduate School policy. It is provided here to appropriately reflect each level.
For source code, please add comments so I can understand what is going on. Believe it or not, some student code is
difficult to read. 😀
A B C D F Zero
A A- B+ B B- C+ C C- D+ D F 0
Part 1 – Unsupervised Learning
Q1/2 – Data Frames and
Targets
Q2 –
sklearn.cluster.KMeans
Q4a – Cluster Quality
Q4b – Plots of each
cluster’s quality values
Q4c – Analysis of quality
Part 2 – Supervised Learning
Q1 – Combined Data
Sets
Q2 – Test and Train Split
Q3-1 – First Classifier
Q3-2 – Second Classifier
Q4a – Performance
Analysis
Q4b – Improvement
Suggestions
Part 3 – Deep Learning
Q2 – MLP Classifier
Q3a – Analysis of test
set
Q3b – Comparison to
classifers from Part 2.
