def run_sfs_glass(filename, target_class):
# Setup data
glass_obj = glass.Glass()
glass_data = glass_obj.setup_data_glass(filename=filename,
target_class=target_class)
# Split the data set into 2/3 and 1/3
glass_data_train = glass_data.sample(frac=.667)
glass_data_test = glass_data.drop(glass_data_train.index)
# Create Naive Bayes for later comparison (Added in last, so that is why it is not greatly structured in the code)
naive = nb.NaiveBayes()
# Train classifier
trainer_classifier = naive.naive_bayes_train(
glass_data_train.iloc[:, 0:9], glass_data_train[target_class])
# Test classifier
test_classifier = naive.naive_bayes_test(glass_data_test.iloc[:, 0:9],
trainer_classifier)
# Get success rate back
nb_perf = naive.compare_prediction(predict_classier=test_classifier,
data=glass_data_test[target_class])
# Setup SFS object
stepwise_forward = sfs.SFS()
# Setup SFS features to go through
feature_set_columns = list()
for feature in glass_data.columns:
feature_set_columns.append(feature)
feature_set_columns.remove(target_class)
# Separate data for taining/testing purposes
glass_data_train_target_class = glass_data_train[target_class]
glass_data_train_features = glass_data_train.iloc[:, 0:9]
glass_data_test_features = glass_data_test.iloc[:, 0:9]
# Run SFS
best_features = stepwise_forward.perform_sfs(
feature_set_columns=feature_set_columns,
d_train=glass_data_train_features,
d_valid=glass_data_test_features,
predicted=glass_data_train_target_class)
# Return dataset with best features only
best_data = glass_data[best_features]
return best_features, best_data, glass_data, nb_perf
def nb_iris(iris_data, target_class):
# Split the data set into 2/3 and 1/3
iris_data_train = iris_data.sample(frac=.667)
iris_data_test = iris_data.drop(iris_data_train.index)
# Create Naive Bayes for later comparison (Added in last, so that is why it is not greatly structured in the code)
naive = nb.NaiveBayes()
# Train classifier
trainer_classifier = naive.naive_bayes_train(iris_data_train.iloc[:, 0:4],
iris_data_train[target_class])
# Test classifier
test_classifier = naive.naive_bayes_test(iris_data_test.iloc[:, 0:4],
trainer_classifier)
# Get success rate back
nb_perf = naive.compare_prediction(predict_classier=test_classifier,
data=iris_data_test[target_class])
return nb_perf
def main():
# Print all output to file
# Comment out for printing in console
sys.stdout = open("./Assignment2Output.txt", "w")
##### Iris #####
iris_target_class = "class"
sfs_iris, sfs_best_data_iris, iris_data, iris_nb = run_sfs_iris(
filename="data/iris.data", target_class=iris_target_class)
print("Best Feature Selection of Iris: ")
print(sfs_iris)
print(sfs_best_data_iris)
print()
# Run k means with k=2 since my data setup normalizes data to be either 0 or 1 for the class we intend to have
kmeans_iris = km.KMeans(num_classes=2, data=sfs_best_data_iris)
iris_centroids, iris_clusters = kmeans_iris.kmeans_alg()
print("Centroids for Iris: ")
print(iris_centroids)
print()
print("Clusters for Iris: ")
print(iris_clusters)
print()
# Get silhouette coefficient
silhouette_coefficient_iris = kmeans_iris.silhouette_coefficient(
iris_clusters)
print("Silhouette coefficient for Iris: " +
str(silhouette_coefficient_iris))
print()
# Get success rate for comparison
compare_iris = nb.NaiveBayes()
success_rate_iris = compare_iris.compare_prediction(
iris_clusters, iris_data[iris_target_class])
print("Success rate for Iris: " + str(success_rate_iris) + "%")
print()
print("Success rate for Iris Naive Bayes: " + str(iris_nb) + "%")
print('\n' * 3)
##### Glass #####
glass_target_class = "Type of glass"
sfs_glass, sfs_best_data_glass, glass_data, glass_nb = run_sfs_glass(
filename="data/glass.data", target_class=glass_target_class)
print("Best Feature Selection of Glass: ")
print(sfs_glass)
print(sfs_best_data_glass)
print()
# Run k means with k=2 since my data setup normalizes data to be either 0 or 1 for the class we intend to have
kmeans_glass = km.KMeans(num_classes=2, data=sfs_best_data_glass)
glass_centroids, glass_clusters = kmeans_glass.kmeans_alg()
print("Centroids for Glass: ")
print(glass_centroids)
print()
print("Clusters for Glass: ")
print(glass_clusters)
print()
# Get silhouette coefficient
silhouette_coefficient_glass = kmeans_glass.silhouette_coefficient(
glass_clusters)
print("Silhouette coefficient for Glass: " +
str(silhouette_coefficient_glass))
print()
# Get success rate for comparison
compare_glass = nb.NaiveBayes()
success_rate_glass = compare_glass.compare_prediction(
glass_clusters, glass_data[glass_target_class])
print("Success rate for Glass: " + str(success_rate_glass) + "%")
print()
print("Success rate for Glass Naive Bayes: " + str(glass_nb) + "%")
print('\n' * 3)
##### Spambase #####
spambase_target_class = "57"
sfs_spambase, sfs_best_data_spambase, spambase_data, spambase_nb = run_sfs_spambase(
filename="data/spambase.data", target_class=spambase_target_class)
print("Best Feature Selection of Spambase: ")
print(sfs_spambase)
print(sfs_best_data_spambase)
print()
# Run k means with k=2 since my data setup normalizes data to be either 0 or 1 for the class we intend to have
kmeans_spambase = km.KMeans(num_classes=2, data=sfs_best_data_spambase)
spambase_centroids, spambase_clusters = kmeans_spambase.kmeans_alg()
print("Centroids for Spambase: ")
print(spambase_centroids)
print()
print("Clusters for Spambase: ")
print(spambase_clusters)
print()
# Get silhouette coefficient
silhouette_coefficient_spambase = kmeans_spambase.silhouette_coefficient(
spambase_clusters)
print("Silhouette coefficient for Spambase: " +
str(silhouette_coefficient_spambase))
print()
# Get success rate for comparison
compare_spambase = nb.NaiveBayes()
success_rate_spambase = compare_spambase.compare_prediction(
spambase_clusters, spambase_data[spambase_target_class])
print("Success rate for Spambase: " + str(success_rate_spambase) + "%")
print()
print("Success rate for Spambase Naive Bayes: " + str(spambase_nb) + "%")
print('\n' * 3)
def __init__(self):
self.nb = nb.NaiveBayes()