Trjtdtk

I'm unsure of how to ask a question without making it seem like a code review question. At what point does one question whether they've actually implemented the algorithm and-or model correctly? Getting spot-on results is great and all, but seems highly suspect. Also, what checks can be done to ensure that the algorithm and-or model is being implemented correctly? The reason I'm asking is because I'm getting perfect classification and subsequently accuracy, precision, etc. w/ the implementation of SVM.

I am including the code, but feel free to ignore.

# Make a copy of the df
iris_df_copy = iris_df.copy()

# Create a new column, labeled 'T/F', whose value will be based on the value in the 'Class' column. If the value in the
# 'Class' column is 'Iris-setosa', then set the value of the 'T/F' column to 1. If the value in the 'Class' column is
# not 'Iris-setosa', then set the value of the 'T/F' column to 0.
iris_df_copy.loc[iris_df_copy.Class == 'Iris-setosa', 'T/F'] = 1
iris_df_copy.loc[iris_df_copy.Class != 'Iris-setosa', 'T/F'] = 0

X_svm = np.array(iris_df_copy[['Sepal_Length', 'Sepal_Width', 'Petal_Length', 'Petal_Width']])
y_svm = np.ravel(iris_df_copy[['T/F']])

# Split the samples into two subsets, use one for training and the other for testing
X_train_svm, X_test_svm, y_train_svm, y_test_svm = train_test_split(X_svm, y_svm, test_size=0.25, random_state=4)

# Instantiate the learning model - Linear SVM
linear_svm = svm.SVC(kernel='linear')

# Fit the model - Linear SVM
linear_svm.fit(X_train_svm, y_train_svm)

# Predict the response - Linear SVM
linear_svm_pred = linear_svm.predict(X_test_svm)

# Confusion matrix and quantitative metrics - Linear SVM
print("The confusion matrix is: " + np.str(confusion_matrix(y_test_svm, linear_svm_pred)))
print("The accuracy score is: " + np.str(accuracy_score(y_test_svm, linear_svm_pred)))
print("The precision is: " + np.str(precision_score(y_test_svm, linear_svm_pred, average="macro")))
print("The recall is: " + np.str(recall_score(y_test_svm, linear_svm_pred, average="macro")))

You need to know what the outcome should be of a given test on a dataset before you try to test a new method on them. Ask yourself, 'What do I expect from this?'

Linear SVM finds a plane to cut through the data to best represent the difference between two sets.

If you have a look at what you are separating (Iris_setosa from Iris_virginica and iris_versicolor), you'll find that the clumps themselves are perfectly separated. You can draw a line easily on each graph you care to use, and that is what I have done in the picture below. If the clumps are perfectly separated, then the SVM will return a perfectly separated result.

By Nicoguaro - Own work, CC BY 4.0, https://commons.wikimedia.org/w/index.php?curid=46257808

Test the SVM on separating virginica and versicolor to see how it does in a more difficult context. Or alternatively, just generate a dataset of your own from randomly placed gaussian points.