def tune_part5networks():
# TODO
neural_params = {
"hidden_layer_sizes": [(64, 64), (128, 128), (64, 64, 64), (128, 128, 128)],
"learning_rate_init": [0.0001, 0.001, 0.01, 0.1],
}
X, y = load_intention()
kmeans = KMeans(2)
em = GaussianMixture(2)
Xtrain, Xtest, ytrain, ytest = train_test_split(
X, y, train_size=0.8, random_state=1
)
kmeans.fit(Xtrain)
em.fit(Xtrain)
cluster1, cluster2 = kmeans.cluster_centers_
distance_from_cluster1 = (np.sqrt((X - cluster1) ** 2)).mean(axis=1)
distance_from_cluster2 = (np.sqrt((X - cluster2) ** 2)).mean(axis=1)
point_probabilities = em.predict_proba(X)
X_minimal = pd.DataFrame(
{
"dist1": distance_from_cluster1,
"dist2": distance_from_cluster2,
"prob1": point_probabilities[:, 0],
"prob2": point_probabilities[:, 1],
}
)
X_augmented = X.copy()
X_augmented[["dist1", "dist2", "prob1", "prob2"]] = X_minimal.copy()
Xminimaltrain, Xminimaltest = train_test_split(
X_minimal, train_size=0.8, random_state=1
)
Xaugtrain, Xaugtest = train_test_split(X_augmented, train_size=0.8, random_state=1)
mlp = MLPClassifier(max_iter=2000, early_stopping=True)
print("Tuning neural network on minimal dataset")
clf_minimal = GridSearchCV(mlp, neural_params)
clf_minimal.fit(X_minimal, y)
print("Tuning neural network on augmented dataset")
clf_augmented = GridSearchCV(
MLPClassifier(max_iter=2000, early_stopping=True), neural_params
)
clf_augmented.fit(X_augmented, y)
minimal_outpath = os.path.join("output", "part5", "clf_minimal.pkl")
augmented_outpath = os.path.join("output", "part5", "clf_augmented.pkl")
with open(minimal_outpath, "wb") as f:
pickle.dump(clf_minimal, f)
with open(augmented_outpath, "wb") as f:
pickle.dump(clf_augmented, f)
def generate_cluster_dfs():
X, y = load_intention()
kmeans = KMeans(2)
em = GaussianMixture(2)
Xtrain, Xtest, ytrain, ytest = train_test_split(
X, y, train_size=0.8, random_state=1
)
kmeans.fit(Xtrain)
em.fit(Xtrain)
cluster1, cluster2 = kmeans.cluster_centers_
distance_from_cluster1 = (np.sqrt((X - cluster1) ** 2)).mean(axis=1)
distance_from_cluster2 = (np.sqrt((X - cluster2) ** 2)).mean(axis=1)
point_probabilities = em.predict_proba(X)
X_minimal = pd.DataFrame(
{
"dist1": distance_from_cluster1,
"dist2": distance_from_cluster2,
"prob1": point_probabilities[:, 0],
"prob2": point_probabilities[:, 1],
}
)
X_augmented = X.copy()
X_augmented[["dist1", "dist2", "prob1", "prob2"]] = X_minimal.copy()
return X_minimal, X_augmented, y
def plot_pca_points(dataset):
if dataset == "pulsar":
_, y = load_pulsar()
else:
_, y = load_intention()
pca_data = get_transformed_data(dataset, "PCA", pca_dir)
fig = plt.figure()
ax = fig.add_subplot(111, projection="3d")
colors = ["red", "blue"]
cmap = mpl.colors.ListedColormap(colors)
ax.scatter(pca_data[:, 0],
pca_data[:, 1],
zs=pca_data[:, 2],
c=y,
cmap=cmap,
alpha=0.2)
ax.set_xlabel("PCA Dimension 1")
ax.set_ylabel("PCA Dimension 2")
ax.set_zlabel("PCA Dimension 3")
title = "PCA-Transformed Data Points by Label"
ax.set_title(title)
plt.savefig(os.path.join(part2_plot_dir, f"PCAPlot{dataset}.png"))
plt.close()
def plot_principal_axes(dataset):
if dataset == "pulsar":
X, y = load_pulsar()
cols_to_use = X.columns # Use all
else:
X, y = load_intention()
cols_to_use = [
"Administrative",
"Administrative_Duration",
"Informational",
"Informational_Duration",
"ProductRelated",
"ProductRelated_Duration",
"BounceRates",
"ExitRates",
"PageValues",
]
index = [f"Component {i+1}" for i in range(X.shape[1])]
pca_transformer_name = f"{dataset}_PCA_transformer.pkl"
pca_data = get_transformed_data(dataset, "PCA", pca_dir)
pca = get_transformer(pca_transformer_name, pca_dir)
components = pd.DataFrame(pca.components_, columns=X.columns, index=index)
components = components[cols_to_use]
first_two_components = components.iloc[:2]
ax = first_two_components.plot(kind="bar")
ax.set_ylabel("Value")
ax.set_title(
"Principal Axes in Feature Space resulting from PCA Decomposition")
plt.savefig(os.path.join(part2_plot_dir, f"PrincipalAxes{dataset}.png"))
plt.close()
def plot_ICA_components(dataset):
if dataset == "intention":
X, y = load_intention()
cols_to_use = [
"Administrative",
"Administrative_Duration",
"Informational",
"Informational_Duration",
"ProductRelated",
"ProductRelated_Duration",
"BounceRates",
"ExitRates",
"PageValues",
]
else:
X, y = load_pulsar()
cols_to_use = X.columns
n_components = ica_components_map[dataset]
transformer_name = f"{dataset}_ICA{n_components}_transformer.pkl"
folder = os.path.join("output", "part2", "ICA")
transformer = get_transformer(transformer_name, folder)
components = pd.DataFrame(transformer.components_, columns=X.columns)
components = components[cols_to_use]
rows = int(np.sqrt(n_components))
fig = plt.figure(figsize=(10, 10))
plt.rcParams.update({"font.size": 5})
for i in range(n_components):
ax = fig.add_subplot(int(f"{rows}{rows}{i+1}"))
ax.set_title(f"Component {i+1}")
components.loc[i, :].plot(kind="bar", ax=ax)
# Turn off ticks
#
if n_components - i > rows:
ax.set_xticklabels([])
for tic in ax.xaxis.get_major_ticks():
tic.tick1On = tic.tick2On = False
plt.subplots_adjust(wspace=1)
plt.rcParams.update({"font.size": 14})
fig.suptitle("ICA Components")
plot_dir = os.path.join("plots", "part2", f"ica_components{dataset}.png")
plt.savefig(plot_dir)
plt.close()
def run_experiment_1():
X1, y1 = load_pulsar()
X2, y2 = load_intention()
print("Running Exp1 on Pulsar Dataset")
pulsar_results = run_clustering_experiment(X1, y1)
pulsar_filename = os.path.join(OUTPUT_DIR, "exp1_pulsar_data.json")
with open(pulsar_filename, "w") as f:
json.dump(pulsar_results, f)
print("Running Exp1 on Intention Dataset")
intention_results = run_clustering_experiment(X2, y2)
intention_filename = os.path.join(OUTPUT_DIR, "exp1_intention_data.json")
with open(intention_filename, "w") as f:
json.dump(intention_results, f)
def part2(intention=True, pulsar=True):
part2_dir = os.path.join(OUTPUT_DIR, "part2")
if intention:
X, y = load_intention()
print("Collecting Exp2 on Intention")
save_all_transformations(X, "intention", output_dir=part2_dir)
if pulsar:
X, y = load_pulsar()
print("Collecting Exp2 Data on Pulsar")
save_all_transformations(
X,
"pulsar",
output_dir=part2_dir,
ica_max_component_ratio=1.0,
rp_component_ratio=1.0,
lle_max_component_ratio=1.0,
)
def evaluate_part4network(transformer):
clf = load_part4_network(transformer)
if transformer == "None":
X, y = load_intention()
elif transformer == "PCA":
X, y = load_intention_PCA_reduced()
elif transformer == "ICA":
X, y = load_intention_ICA_reduced()
elif transformer == "RP":
X, y = load_intention_RP_reduced()
elif transformer == "LLE":
X, y = load_intention_LLE_reduced()
# Should be consistent due to random seed
Xtrain, Xtest, ytrain, ytest = train_test_split(
X, y, train_size=0.8, random_state=1
)
train_preds = clf.predict(Xtrain)
test_preds = clf.predict(Xtest)
train_acc = accuracy_score(ytrain, train_preds)
test_acc = accuracy_score(ytest, test_preds)
train_f1 = f1_score(ytrain, train_preds)
test_f1 = f1_score(ytest, test_preds)
best_classifier_index = np.argmin(clf.cv_results_["rank_test_score"])
train_time = clf.cv_results_["mean_fit_time"][best_classifier_index]
return {
"train_accuracy": train_acc,
"train_f1": train_f1,
"test_accuracy": test_acc,
"test_f1": test_f1,
"train_time": train_time,
}
def plot_cluster_means(
data_loader,
transformer_path,
dataset,
output_dir,
file_prefix,
kmeans_clusters=2,
em_clusters=2,
):
if dataset == "intention":
X_untransformed = load_intention()
else:
X_untransformed = load_pulsar()
Xtransformed, y = data_loader()
with open(transformer_path, "rb") as f:
transformer = pickle.load(f)
X = get_inverse_transform(transformer, Xtransformed)
X = pd.DataFrame(X, columns=X_untransformed.columns)
if data_loader is load_intention:
X_plot = X[
[
"Administrative",
"Administrative_Duration",
"Informational",
"Informational_Duration",
"ProductRelated",
"ProductRelated_Duration",
"BounceRates",
"ExitRates",
"PageValues",
]
]
else:
X_plot = X
fig, (ax1, ax2) = plt.subplots(1, 2)
kmeans = KMeans(kmeans_clusters, random_state=1)
em = GaussianMixture(n_components=em_clusters, random_state=1)
kmeans.fit(X)
em.fit(X)
kmeans_df = pd.DataFrame(kmeans.cluster_centers_, columns=X.columns)
kmeans_df = kmeans_df[X_plot.columns]
em_df = pd.DataFrame(em.means_, columns=X.columns)
em_df = em_df[X_plot.columns]
kmeans_df.plot(kind="bar", ax=ax1)
em_df.plot(kind="bar", ax=ax2)
ax1.set_ylabel("Mean Value")
ax2.set_ylabel("Mean Value")
ax1.set_xlabel("Cluster")
ax2.set_xlabel("Cluster")
ax1.set_title("K-Means Cluster Centers")
ax2.set_title("EM Cluster Centers")
ax1.get_legend().remove()
plt.savefig(os.path.join(output_dir, f"{file_prefix}_clusterprojections.png"))
plt.close()
output_dir="output/part2",
ica_max_component_ratio=0.3,
rp_component_ratio=0.3,
lle_max_component_ratio=0.3):
pca_dir = os.path.join(output_dir, "PCA")
ica_dir = os.path.join(output_dir, "ICA")
rp_dir = os.path.join(output_dir, "RP")
lle_dir = os.path.join(output_dir, "LLE")
print(f"Running PCA Collection on {file_prefix}")
# run_PCA_collection(data, file_prefix, pca_dir)
print(f"Running ICA Collection on {file_prefix}")
# run_ICA_collection(data, file_prefix, ica_dir, max_component_ratio=ica_max_component_ratio)
print(f"Running RP Collection on {file_prefix}")
run_RP_collection(data,
file_prefix,
rp_dir,
max_component_ratio=rp_component_ratio)
print(f"Running LLE Collection on {file_prefix}")
# run_LLE_collection(data, file_prefix, lle_dir, max_component_ratio=lle_max_component_ratio)
if __name__ == "__main__":
X, y = load_intention()
X2, y2 = load_pulsar()
print("Collecting Exp2 on Intention")
save_all_transformations(X, "intention")
print("Collecting Exp2 Data on Pulsar")
save_all_transformations(X2, "pulsar")
