This repository has been archived on 2023-06-18. You can view files and clone it, but cannot push or open issues or pull requests.

7.0 KiB

Raw Blame History

author	title	geometry
Claudio Maggioni	Information Modelling & Analysis -- Project 2	margin=2cm,bottom=3cm

Code Repository

The code and result files, part of this submission, can be found at

Repository: https://github.com/infoMA2023/project-02-bug-prediction-maggicl
Commit ID: TBD

Data Pre-Processing

I use the sources of the [Closure]{.smallcaps} repository that were already downloaded using the command:

defects4j checkout -p Closure -v 1f -w ./resources

and used the code in the following subfolder for the project:

./resources/defects4j-checkout-closure-1f/src/com/google/javascript/jscomp/

relative to the root folder of the repository. The resulting csv of extracted, labelled feature vectors can be found in the repository at the following path:

./metrics/feature_vectors_labeled.csv

relative to the root folder of the repository.

Feature Vector Extraction

I extracted 291 feature vectors in total. Aggregate metrics about the extracted feature vectors, i.e. the distribution of the values of each code metric, can be found in Table 1{reference-type="ref" reference="tab:metrics"}.

::: {#tab:metrics}

Metric	Average	Max
BCM	13.4124	221
CPX	5.8247	96
DCM	4.8652	176.2
EX	0.1134	2
FLD	6.5773	167
INT	0.6667	3
MTH	11.6529	209
NML	13.5622	28
RET	3.6735	86
RFC	107.2710	882
SZ	18.9966	347
WRD	314.4740	3133

: Distribution of values for each extracted code metric. :::

Feature Vector Labelling

After feature vectors are labeled, I determine that the dataset contains 75 buggy classes and 216 non-buggy classes.

Classifiers

In every subsection below, describe in a concise way which different hyperparameters you tried for the corresponding classifier, and report the corresponding precision, recall and F1 values (for example in a table or an [itemize]{.smallcaps}-environment). Furthermore, for every type of classifiers, explicitly mention which hyperparameter configuration you chose (based on above reported results) to be used in further steps, and (in one or two sentences), explain why these hyperparameters may outperform the other ones you tested..

Decision Tree (DT)

Naive Bayes (NB)

Support Vector Machine (SVP)

Multi-Layer Perceptron (MLP)

Random Forest (RF)

Evaluation

Output Distributions

Add a boxplot showing mean and standard deviation for Precision values on all 6 classifiers (5 trained + 1 biased)
Add a boxplot showing mean and standard deviation for Recall values on all 6 classifiers (5 trained + 1 biased)
Add a boxplot showing mean and standard deviation for F1 values on all 6 classifiers (5 trained + 1 biased)

{#fig:boxplot}

Comparison and Significance

For every combination of two classifiers and every performance metric (precision, recall, f1) compare which algorithm performs better, by how much, and report the corresponding p-value in the following subsubsections:

::: {#tab:precision}

	DecisionTreeClassifier	GaussianNB	MLPClassifier	RandomForestClassifier
BiasedClassifier	0	0	0	0
DecisionTreeClassifier	--	0.0893	0.4012	0
GaussianNB	--	--	0.0348	0
MLPClassifier	--	--	--	0
RandomForestClassifier	--	--	--	--

: Pairwise Wilcoxon test for precision for each combination of classifiers. :::

::: {#tab:recall}

	DecisionTreeClassifier	GaussianNB	MLPClassifier	RandomForestClassifier	SVC
BiasedClassifier	0	0	0	0	0
DecisionTreeClassifier	--	0	0.0118	0.3276	0
GaussianNB	--	--	0	0	0
MLPClassifier	--	--	--	0.0001	0
RandomForestClassifier	--	--	--	--	0
: Pairwise Wilcoxon test for recall for each combination of classifiers.
:::

::: {#tab:f1}

	DecisionTreeClassifier	GaussianNB	MLPClassifier	RandomForestClassifier	SVC
BiasedClassifier	0	0	0	0	0
DecisionTreeClassifier	--	0	0.4711	0	0
GaussianNB	--	--	0	0	0
MLPClassifier	--	--	--	0	0
RandomForestClassifier	--	--	--	--	0
: Pairwise Wilcoxon test for the F1 score metric for each combination of classifiers.
:::

F1 Values

Precision

(same as for F1 above)

Recall