Classification

In [1]:

import numpy as np
import pandas as pd
import holmc as hc
import matplotlib.pyplot as plt
from sklearn.preprocessing import OneHotEncoder, LabelEncoder
from ucimlrepo import fetch_ucirepo
from sklearn.model_selection import train_test_split

import numpy as np import pandas as pd import holmc as hc import matplotlib.pyplot as plt from sklearn.preprocessing import OneHotEncoder, LabelEncoder from ucimlrepo import fetch_ucirepo from sklearn.model_selection import train_test_split

In [2]:

mushroom = fetch_ucirepo(id=73)
X_raw = mushroom.data.features
y_raw = mushroom.data.targets

# Step 1: Convert to pandas DataFrame / Series
X_df = pd.DataFrame(X_raw)
y_series = pd.Series(y_raw.values.ravel(), name="target")

# Step 2: Split the raw data (not encoded yet!)
X_train_df, X_test_df, y_train_series, y_test_series = train_test_split(
    X_df, y_series, test_size=0.30, random_state=42, stratify=y_series
)

# Step 3: Encode labels (target values)
label_encoder = LabelEncoder()
y_train = label_encoder.fit_transform(y_train_series)
y_test = label_encoder.transform(y_test_series) 

# Step 4: One-hot encode features
encoder = OneHotEncoder(sparse_output=False, handle_unknown="ignore")
X_train_encoded = encoder.fit_transform(X_train_df)
X_test_encoded = encoder.transform(X_test_df) 

# Step 5: Add intercept term (column of 1s)
X_train = np.hstack((np.ones((X_train_encoded.shape[0], 1)), X_train_encoded))
X_test = np.hstack((np.ones((X_test_encoded.shape[0], 1)), X_test_encoded))

mushroom = fetch_ucirepo(id=73) X_raw = mushroom.data.features y_raw = mushroom.data.targets # Step 1: Convert to pandas DataFrame / Series X_df = pd.DataFrame(X_raw) y_series = pd.Series(y_raw.values.ravel(), name="target") # Step 2: Split the raw data (not encoded yet!) X_train_df, X_test_df, y_train_series, y_test_series = train_test_split( X_df, y_series, test_size=0.30, random_state=42, stratify=y_series ) # Step 3: Encode labels (target values) label_encoder = LabelEncoder() y_train = label_encoder.fit_transform(y_train_series) y_test = label_encoder.transform(y_test_series) # Step 4: One-hot encode features encoder = OneHotEncoder(sparse_output=False, handle_unknown="ignore") X_train_encoded = encoder.fit_transform(X_train_df) X_test_encoded = encoder.transform(X_test_df) # Step 5: Add intercept term (column of 1s) X_train = np.hstack((np.ones((X_train_encoded.shape[0], 1)), X_train_encoded)) X_test = np.hstack((np.ones((X_test_encoded.shape[0], 1)), X_test_encoded))

In [3]:

seed = 101
np.random.seed(seed)
N = 150  # number of samples
lamb = 25

seed = 101 np.random.seed(seed) N = 150 # number of samples lamb = 25

In [4]:

etas = [0.00001, 0.0001, 0.0004, 0.0008, 0.003, 0.006, 0.009, 0.012]
gammas = [1, 5, 10, 20, 40, 80]
xis = [1, 5, 10, 20, 40, 80]

etas = [0.00001, 0.0001, 0.0004, 0.0008, 0.003, 0.006, 0.009, 0.012] gammas = [1, 5, 10, 20, 40, 80] xis = [1, 5, 10, 20, 40, 80]

In [5]:

search_o3 = hc.GridSearchClassification(
    gammas=gammas, etas=etas, xis=xis, N=N, seed=seed,show_progress=True
)
o3paramspace=search_o3.run(X=X_train, y=y_train, lamb=lamb)
o3best_gamma = o3paramspace.iloc[0,0]
o3best_eta = o3paramspace.iloc[0,1]
o3best_xi = o3paramspace.iloc[0,3]
o3p = hc.O3Params(gamma=o3best_gamma, eta=o3best_eta, xi=o3best_xi)
sampler3 = hc.HoLMCSamplerO3Classification(
    params=o3p, N=N, seed=seed,
    show_progress=True
)
sample3 = sampler3.sample(X=X_train, y=y_train, lamb=lamb)
metric = hc.AccuracyMeasure(X=X_test, y=y_test)
accuracies = {}
StDev = {}
accuracies['Order 3'] = metric.compute_accuracy(sample3)
StDev['Order 3'] = np.std(accuracies['Order 3'], axis=0)
# Sampling with order 4
search_o4 = hc.GridSearchClassification(
    gammas=gammas, etas=etas, N=N, seed=seed,show_progress=True
)
o4paramspace=search_o4.run(X=X_train, y=y_train, lamb=lamb)
o4best_gamma = o4paramspace.iloc[0,0]
o4best_eta = o4paramspace.iloc[0,1]
o4p = hc.O4Params(gamma=o4best_gamma, eta=o4best_eta)
sampler4 = hc.HoLMCSamplerO4Classification(
    params=o4p, N=N, seed=seed, show_progress=True
)
sample4 = sampler4.sample(X=X_train, y=y_train, lamb=lamb)
accuracies['Order 4'] = metric.compute_accuracy(sample4)
StDev['Order 4'] = np.std(accuracies['Order 4'], axis=0)

search_o3 = hc.GridSearchClassification( gammas=gammas, etas=etas, xis=xis, N=N, seed=seed,show_progress=True ) o3paramspace=search_o3.run(X=X_train, y=y_train, lamb=lamb) o3best_gamma = o3paramspace.iloc[0,0] o3best_eta = o3paramspace.iloc[0,1] o3best_xi = o3paramspace.iloc[0,3] o3p = hc.O3Params(gamma=o3best_gamma, eta=o3best_eta, xi=o3best_xi) sampler3 = hc.HoLMCSamplerO3Classification( params=o3p, N=N, seed=seed, show_progress=True ) sample3 = sampler3.sample(X=X_train, y=y_train, lamb=lamb) metric = hc.AccuracyMeasure(X=X_test, y=y_test) accuracies = {} StDev = {} accuracies['Order 3'] = metric.compute_accuracy(sample3) StDev['Order 3'] = np.std(accuracies['Order 3'], axis=0) # Sampling with order 4 search_o4 = hc.GridSearchClassification( gammas=gammas, etas=etas, N=N, seed=seed,show_progress=True ) o4paramspace=search_o4.run(X=X_train, y=y_train, lamb=lamb) o4best_gamma = o4paramspace.iloc[0,0] o4best_eta = o4paramspace.iloc[0,1] o4p = hc.O4Params(gamma=o4best_gamma, eta=o4best_eta) sampler4 = hc.HoLMCSamplerO4Classification( params=o4p, N=N, seed=seed, show_progress=True ) sample4 = sampler4.sample(X=X_train, y=y_train, lamb=lamb) accuracies['Order 4'] = metric.compute_accuracy(sample4) StDev['Order 4'] = np.std(accuracies['Order 4'], axis=0)

Eta: 100%|██████████| 8/8 [01:15<00:00,  9.49s/it]
100%|██████████| 150/150 [00:00<00:00, 1162.34it/s]
Eta: 100%|██████████| 8/8 [00:30<00:00,  3.76s/it]
100%|██████████| 150/150 [00:00<00:00, 300.10it/s]

In [6]:

# Plotting
figure, ax = plt.subplots(figsize=(8.8, 6.6))
index = list(range(N+1))
label_str3 = (
    "Order 3\n"
    f"$\\gamma$ = {o3best_gamma}\n"
    f"$\\eta$ = {o3best_eta}\n"
    f"$\\xi$ = {o3best_xi}"
)
label_str4 = (
    "Order 4\n"
    f"$\\gamma$ = {o4best_gamma}\n"
    f"$\\eta$ = {o4best_eta}\n"
)
plt.plot(index, accuracies['Order 3'], label=label_str3, color='blue', linewidth=3)
plt.fill_between(index, 
                 accuracies['Order 3'] - 0.5 * StDev['Order 3'], 
                 (accuracies['Order 3'] + 0.5 * StDev['Order 3']), 
                 color='blue', alpha=0.2)
plt.plot(index, accuracies['Order 4'], label=label_str4, color='green', linewidth=3)
plt.fill_between(index, 
                 accuracies['Order 4'] - 0.5 * StDev['Order 4'], 
                 accuracies['Order 4'] +0.5 * StDev['Order 4'], 
                 color='green', alpha=0.2)
plt.xlabel("Iteration", fontsize=30)
plt.ylabel("Accuracy", fontsize=30)
plt.legend(fontsize=30)
ax.tick_params(axis='both', which='major', labelsize=22) 
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
    label.set_fontname('Times New Roman')
plt.tight_layout()
plt.ylim(0.45, 1)
plt.savefig('../images/o3o4class.png')
plt.show()

# Plotting figure, ax = plt.subplots(figsize=(8.8, 6.6)) index = list(range(N+1)) label_str3 = ( "Order 3\n" f"$\\gamma$ = {o3best_gamma}\n" f"$\\eta$ = {o3best_eta}\n" f"$\\xi$ = {o3best_xi}" ) label_str4 = ( "Order 4\n" f"$\\gamma$ = {o4best_gamma}\n" f"$\\eta$ = {o4best_eta}\n" ) plt.plot(index, accuracies['Order 3'], label=label_str3, color='blue', linewidth=3) plt.fill_between(index, accuracies['Order 3'] - 0.5 * StDev['Order 3'], (accuracies['Order 3'] + 0.5 * StDev['Order 3']), color='blue', alpha=0.2) plt.plot(index, accuracies['Order 4'], label=label_str4, color='green', linewidth=3) plt.fill_between(index, accuracies['Order 4'] - 0.5 * StDev['Order 4'], accuracies['Order 4'] +0.5 * StDev['Order 4'], color='green', alpha=0.2) plt.xlabel("Iteration", fontsize=30) plt.ylabel("Accuracy", fontsize=30) plt.legend(fontsize=30) ax.tick_params(axis='both', which='major', labelsize=22) for label in (ax.get_xticklabels() + ax.get_yticklabels()): label.set_fontname('Times New Roman') plt.tight_layout() plt.ylim(0.45, 1) plt.savefig('../images/o3o4class.png') plt.show()

In [7]:

gamma_vals = [1, 5, 10, 20]
eta = o3best_eta
xi = o3best_xi 
AccForGammaVar_o3 = {}
for gamma in gamma_vals:
    o3p = hc.O3Params(gamma=gamma, eta=eta, xi=xi)
    sampler = hc.HoLMCSamplerO3Classification(
        params=o3p, N=N, seed=seed, show_progress=False
    )
    sample = sampler.sample(X=X_train, y=y_train, lamb=lamb)
    AccForGammaVar_o3[gamma] = metric.compute_accuracy(sample)
AccForGammaVar_o3df = pd.DataFrame(AccForGammaVar_o3)
figure, ax = plt.subplots(figsize=(8.8, 6.6))
for col, gamma in zip(AccForGammaVar_o3df, gamma_vals):
    plt.plot(
        AccForGammaVar_o3df.index,
        AccForGammaVar_o3df[col],
        label=f"$\\gamma = {gamma}$",
        linewidth=3
    )
plt.xlabel("Iteration", fontsize=30)
plt.ylabel("Accuracy", fontsize=30)
plt.legend(fontsize=14, loc="right")
ax.tick_params(axis='both', which='major', labelsize=22) 
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
    label.set_fontname('Times New Roman')
plt.tight_layout()
plt.ylim(0.9,1)
plt.savefig('../images/o3class_gamma_var.png')
plt.show()

gamma_vals = [1, 5, 10, 20] eta = o3best_eta xi = o3best_xi AccForGammaVar_o3 = {} for gamma in gamma_vals: o3p = hc.O3Params(gamma=gamma, eta=eta, xi=xi) sampler = hc.HoLMCSamplerO3Classification( params=o3p, N=N, seed=seed, show_progress=False ) sample = sampler.sample(X=X_train, y=y_train, lamb=lamb) AccForGammaVar_o3[gamma] = metric.compute_accuracy(sample) AccForGammaVar_o3df = pd.DataFrame(AccForGammaVar_o3) figure, ax = plt.subplots(figsize=(8.8, 6.6)) for col, gamma in zip(AccForGammaVar_o3df, gamma_vals): plt.plot( AccForGammaVar_o3df.index, AccForGammaVar_o3df[col], label=f"$\\gamma = {gamma}$", linewidth=3 ) plt.xlabel("Iteration", fontsize=30) plt.ylabel("Accuracy", fontsize=30) plt.legend(fontsize=14, loc="right") ax.tick_params(axis='both', which='major', labelsize=22) for label in (ax.get_xticklabels() + ax.get_yticklabels()): label.set_fontname('Times New Roman') plt.tight_layout() plt.ylim(0.9,1) plt.savefig('../images/o3class_gamma_var.png') plt.show()

In [8]:

gamma_vals = [1, 5, 10, 20]
eta = o4best_eta

AccForGammaVar_o4 = {}
for gamma in gamma_vals:
    o4p = hc.O4Params(gamma=gamma, eta=eta)
    sampler = hc.HoLMCSamplerO4Classification(
        params=o4p, N=N, seed=seed, show_progress=False
    )
    sample = sampler.sample(X=X_train, y=y_train, lamb=lamb)
    AccForGammaVar_o4[gamma] = metric.compute_accuracy(sample)
AccForGammaVar_o4df = pd.DataFrame(AccForGammaVar_o4)
figure, ax = plt.subplots(figsize=(8.8, 6.6))
for col, gamma in zip(AccForGammaVar_o4df, gamma_vals):
    plt.plot(
        AccForGammaVar_o4df.index,
        AccForGammaVar_o4df[col],
        label=f"$\\gamma = {gamma}$",
        linewidth=3
    )
plt.xlabel("Iteration", fontsize=30)
plt.ylabel("Accuracy", fontsize=30)
plt.legend(fontsize=14, loc="right")
ax.tick_params(axis='both', which='major', labelsize=22) 
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
    label.set_fontname('Times New Roman')
plt.tight_layout()
plt.ylim(0.9,1)
plt.savefig('../images/o4class_gamma_var.png')
plt.show()

gamma_vals = [1, 5, 10, 20] eta = o4best_eta AccForGammaVar_o4 = {} for gamma in gamma_vals: o4p = hc.O4Params(gamma=gamma, eta=eta) sampler = hc.HoLMCSamplerO4Classification( params=o4p, N=N, seed=seed, show_progress=False ) sample = sampler.sample(X=X_train, y=y_train, lamb=lamb) AccForGammaVar_o4[gamma] = metric.compute_accuracy(sample) AccForGammaVar_o4df = pd.DataFrame(AccForGammaVar_o4) figure, ax = plt.subplots(figsize=(8.8, 6.6)) for col, gamma in zip(AccForGammaVar_o4df, gamma_vals): plt.plot( AccForGammaVar_o4df.index, AccForGammaVar_o4df[col], label=f"$\\gamma = {gamma}$", linewidth=3 ) plt.xlabel("Iteration", fontsize=30) plt.ylabel("Accuracy", fontsize=30) plt.legend(fontsize=14, loc="right") ax.tick_params(axis='both', which='major', labelsize=22) for label in (ax.get_xticklabels() + ax.get_yticklabels()): label.set_fontname('Times New Roman') plt.tight_layout() plt.ylim(0.9,1) plt.savefig('../images/o4class_gamma_var.png') plt.show()

In [9]:

o3paramspace

o3paramspace

Out[9]:

	gamma	eta	MaxAcc	xi
190	5	0.00600	0.9942	40
220	1	0.00900	0.9940	40
181	1	0.00600	0.9940	5
191	5	0.00600	0.9940	80
221	1	0.00900	0.9940	80
...	...	...	...	...
19	20	0.00001	0.4736	5
18	20	0.00001	0.4736	1
17	10	0.00001	0.4736	80
16	10	0.00001	0.4736	40
0	1	0.00001	0.4736	1

288 rows × 4 columns

In [10]:

o4paramspace

o4paramspace

Out[10]:

	gamma	eta	MaxAcc
42	1	0.01200	0.9947
36	1	0.00900	0.9945
30	1	0.00600	0.9940
43	5	0.01200	0.9930
37	5	0.00900	0.9923
31	5	0.00600	0.9877
44	10	0.01200	0.9875
24	1	0.00300	0.9873
38	10	0.00900	0.9847
45	20	0.01200	0.9843
25	5	0.00300	0.9833
39	20	0.00900	0.9807
32	10	0.00600	0.9805
46	40	0.01200	0.9757
33	20	0.00600	0.9727
26	10	0.00300	0.9703
40	40	0.00900	0.9696
47	80	0.01200	0.9627
34	40	0.00600	0.9587
27	20	0.00300	0.9567
41	80	0.00900	0.9501
35	80	0.00600	0.9305
28	40	0.00300	0.9288
18	1	0.00080	0.9038
19	5	0.00080	0.9020
20	10	0.00080	0.8955
29	80	0.00300	0.8801
21	20	0.00080	0.8718
22	40	0.00080	0.7891
12	1	0.00040	0.7346
13	5	0.00040	0.7334
14	10	0.00040	0.7297
15	20	0.00040	0.7165
23	80	0.00080	0.6950
16	40	0.00040	0.6658
17	80	0.00040	0.5990
6	1	0.00010	0.4921
7	5	0.00010	0.4921
8	10	0.00010	0.4919
9	20	0.00010	0.4919
10	40	0.00010	0.4917
11	80	0.00010	0.4907
1	5	0.00001	0.4757
5	80	0.00001	0.4757
4	40	0.00001	0.4757
3	20	0.00001	0.4757
2	10	0.00001	0.4757
0	1	0.00001	0.4757

In [ ]: