MULTILBAEL CLASSIFICATION USING PYTORCH
Our team is currently working on a project focused on Face Image Retrieval using state-of-the-art Deep Learning Models. The core of our approach involves leveraging Convolutional Neural Networks (CNN) to identify and extract facial features. The project centers around the comprehensive CelebA dataset, which contains a diverse collection of celebrity images.
| NAME | ID | CONTRIBUTE | ROLE |
|---|---|---|---|
| Trần Mai Chí Tâm | 21280047 | 25% | Leader |
| Ngô Ngọc Hà Khanh | 21280019 | 25% | |
| Nguyễn Đặng Anh Thư | 21280111 | 25% | |
| Y Cao Lâm | 21280027 | 25% |
import pandas as pd
import os, os.path
from os import listdir
import cv2
import numpy as np
from matplotlib import pyplot as plt
import warnings
import seaborn as sns
warnings.filterwarnings('ignore')
ROOT = r'Z:\ndm\note'
PATH = r'Z:\ndm\note\img_align_celeba'
from PIL import Image as PILImage
from IPython.display import Image as IPImage os.listdir()['anno', 'eval', 'img_align_celeba', 'README.txt']
anno_dir = 'anno'
eval_dir = 'eval'
print('anno:', os.listdir(anno_dir))
print('eval:', os.listdir(eval_dir))anno: ['identity_CelebA.txt', 'list_attr_celeba.txt', 'list_bbox_celeba.txt', 'list_landmarks_align_celeba.txt', 'list_landmarks_celeba.txt']
eval: ['list_eval_partition.txt']
f = open(os.path.join(anno_dir, 'list_attr_celeba.txt'))
file = f.readlines()
sample = int(file[0])
classes = file[1].split(' ')
classes.pop(-1)
attr = []
for i in file[2:]:
list_ = i.split()
list_.pop(0)
list_ = list(map(int, list_))
attr.append(list_)convert the attribute to binary
for li in attr:
for ind in range(len(li)):
if(li[ind] == -1):
li[ind] = 0f = open(os.path.join(eval_dir, 'list_eval_partition.txt'))
file = f.readlines()
eval_dict = {'name': [],
'eval': []}
for i in file:
key, value = i.split()
eval_dict['name'].append(key)
eval_dict['eval'].append(int(value))
eval_dict_csv = pd.DataFrame(eval_dict)f = open(os.path.join(anno_dir, "list_landmarks_align_celeba.txt"))
file = f.readlines()
landmark_align = file[1].split() #classes
landmark_align_cor = []
for fi in file[2:]:
li = list(map(int,fi.split()[1:]))
landmark_align_cor.append(li)df = pd.DataFrame(attr, columns=classes)
df.to_csv('attribute.csv', index=False)
df = pd.DataFrame(landmark_align_cor, columns=landmark_align)
df.to_csv('landmark_align.csv', index=False)
eval_dict_csv.to_csv('eval.csv', index=False)andmark_align.csv
landmark_align = pd.read_csv('landmark_align.csv')
landmark_align.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| lefteye_x | lefteye_y | righteye_x | righteye_y | nose_x | nose_y | leftmouth_x | leftmouth_y | rightmouth_x | rightmouth_y | |
|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 69 | 109 | 106 | 113 | 77 | 142 | 73 | 152 | 108 | 154 |
| 1 | 69 | 110 | 107 | 112 | 81 | 135 | 70 | 151 | 108 | 153 |
| 2 | 76 | 112 | 104 | 106 | 108 | 128 | 74 | 156 | 98 | 158 |
| 3 | 72 | 113 | 108 | 108 | 101 | 138 | 71 | 155 | 101 | 151 |
| 4 | 66 | 114 | 112 | 112 | 86 | 119 | 71 | 147 | 104 | 150 |
This landmark_align.csv file give us the coordinates for the location of some features like eyes, nose and mouth. We also draw this to beter visualize what we have
landmark_id = pd.read_csv('landmark_align.csv')
landmark_id.insert(0, 'ID', range(0, len(landmark_id)))import cv2
import matplotlib.pyplot as plt
def show_images(images):
num_rows = 4
num_cols = 16
plt.figure(figsize=(40, 10))
plt.subplots_adjust(wspace=0, hspace=0)
for i in range(len(images)):
plt.subplot(num_rows, num_cols, i + 1)
if images[i].dtype == 'uint8':
plt.imshow(cv2.cvtColor(images[i], cv2.COLOR_BGR2RGB))
else:
plt.imshow(images[i])
plt.axis('off')
plt.show()
def draw_landmark(df, id, img):
id = int(id)
result = df[df['ID'] == id]
for i in range(1, len(result.columns), 2):
point_x = int(result.iloc[0, i])
point_y = int(result.iloc[0, i + 1])
img = cv2.circle(img, (point_x, point_y), radius=4, color=(51, 255, 51), thickness=-1)
return img
images=[]
for image in ['128338.jpg', '100120.jpg', '000090.jpg', '000002.jpg']:
file = os.path.join(PATH, image)
image_ = cv2.imread(file)
image_ = draw_landmark(landmark_id, int(image.replace('.jpg', '')), image_)
images.append(image_)
show_images(images)
attribute = pd.read_csv('attribute.csv')
attribute.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| 5_o_Clock_Shadow | Arched_Eyebrows | Attractive | Bags_Under_Eyes | Bald | Bangs | Big_Lips | Big_Nose | Black_Hair | Blond_Hair | ... | Sideburns | Smiling | Straight_Hair | Wavy_Hair | Wearing_Earrings | Wearing_Hat | Wearing_Lipstick | Wearing_Necklace | Wearing_Necktie | Young | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 1 | 1 | 0 | 1 | 0 | 1 | 0 | 0 | 1 |
| 1 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 1 | 0 | 0 | ... | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 |
| 2 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 |
| 3 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | ... | 0 | 0 | 1 | 0 | 1 | 0 | 1 | 1 | 0 | 1 |
| 4 | 0 | 1 | 1 | 0 | 0 | 0 | 1 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 | 1 |
5 rows × 40 columns
for col in attribute.columns:
print(col)5_o_Clock_Shadow
Arched_Eyebrows
Attractive
Bags_Under_Eyes
Bald
Bangs
Big_Lips
Big_Nose
Black_Hair
Blond_Hair
Blurry
Brown_Hair
Bushy_Eyebrows
Chubby
Double_Chin
Eyeglasses
Goatee
Gray_Hair
Heavy_Makeup
High_Cheekbones
Male
Mouth_Slightly_Open
Mustache
Narrow_Eyes
No_Beard
Oval_Face
Pale_Skin
Pointy_Nose
Receding_Hairline
Rosy_Cheeks
Sideburns
Smiling
Straight_Hair
Wavy_Hair
Wearing_Earrings
Wearing_Hat
Wearing_Lipstick
Wearing_Necklace
Wearing_Necktie
Young
In this phase of our project, we delve into Exploratory Data Analysis (EDA) to gain insights into the distribution of our dataset. Understanding the distribution is fundamental in shaping our understanding of the data and informing subsequent modeling decisions.
counts = attribute.sum()
percentages = (counts / len(attribute)) * 100
percentages_sorted = percentages.sort_values(ascending=False)
plt.figure(figsize=(12, 6))
percentages_sorted.plot(kind='bar', color='skyblue', edgecolor='black')
plt.title('CelebA Attributes Frequency ')
plt.xlabel('Attributes')
plt.ylabel('Percentages')
plt.xticks(rotation=45, ha='right')
plt.show()
Overall the dataset is imbalanced
Now let check some of the attributes.
First is the gender
gender_attr = attribute.loc[:, 'Male']
gender_attr = gender_attr.to_frame()
gender_attr.columns = ['gender']
gender_attr = gender_attr.replace(1, 'Male')
gender_attr = gender_attr.replace(0, 'Female')
total_count = len(gender_attr)
percentage_male = (gender_attr['gender'] == 'Female').sum() / total_count * 100
percentage_female = (gender_attr['gender'] == 'Male').sum() / total_count * 100
plt.title('Gender')
sns.countplot(x='gender', data=gender_attr, color="c")
plt.text(0, percentage_male, f'{percentage_male:.2f}%', ha='center', va='bottom')
plt.text(1, percentage_female, f'{percentage_female:.2f}%', ha='center', va='bottom')
plt.show()
Although the female take more than half of the dataset. But by the number of images we have, this dataset is still very balance.
Age
age_attr = attribute.loc[:, 'Young']
age_attr = age_attr.to_frame()
age_attr.columns = ['age']
age_attr = age_attr.replace(1, 'Young')
age_attr = age_attr.replace(0, 'Old')
total_count = len(age_attr)
percentage_young = (age_attr['age'] == 'Young').sum() / total_count * 100
percentage_old = (age_attr['age'] == 'Old').sum() / total_count * 100
plt.title('Age')
sns.countplot(x='age', data=age_attr, color="c")
plt.text(0, percentage_young, f'{percentage_young:.2f}%', ha='center', va='bottom')
plt.text(1, percentage_old, f'{percentage_old:.2f}%', ha='center', va='bottom')
plt.show()
Most of the people in the dataset are young, this imbalance could make it harder for the model to know the difference between young and old people
Now let see the correlation matrix for age and gender
def plotCorrelationMatrix(df, graphWidth):
df = df[[col for col in df if df[col].nunique() > 1]] # keep columns where there are more than 1 unique values
if df.shape[1] < 2:
print(f'No correlation plots shown: The number of non-NaN or constant columns ({df.shape[1]}) is less than 2')
return
corr = df.corr()
plt.figure(num=None, figsize=(graphWidth, graphWidth), dpi=80, facecolor='w', edgecolor='k')
corrMat = plt.matshow(corr, fignum = 1)
plt.xticks(range(len(corr.columns)), corr.columns, rotation=90)
plt.yticks(range(len(corr.columns)), corr.columns)
plt.gca().xaxis.tick_bottom()
plt.colorbar(corrMat)
plt.show()age_attr = attribute.loc[:, 'Young']
gender_attr = attribute.loc[:, 'Male']
age_gender = pd.DataFrame([gender_attr, age_attr])
age_gender = age_gender.T
age_gender.head()
<style scoped>
.dataframe tbody tr th:only-of-type {
vertical-align: middle;
}
</style>
.dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| Male | Young | |
|---|---|---|
| 0 | 0 | 1 |
| 1 | 0 | 1 |
| 2 | 1 | 1 |
| 3 | 0 | 1 |
| 4 | 0 | 1 |
plotCorrelationMatrix(age_gender, 10)
There is little to no relationship between these two attributes
But how about the whole dataset.
plotCorrelationMatrix(attribute, 10)
Wỉth this correlation matrix we can point out the some stonger relationship, like Heavy_Makeup and Wearing_Lipstick
makeup_attr = attribute.loc[:, 'Heavy_Makeup']
lipstick_attr = attribute.loc[:, 'Wearing_Lipstick']
makeup_lipstick = pd.DataFrame([makeup_attr, lipstick_attr])
makeup_lipstick = makeup_lipstick.T
plotCorrelationMatrix(makeup_lipstick, 10)
The correlation is about 80%, which is easy to understand as when people wear heavy makeup they most likely have lipstick on as well
Multi Label Classification
Multiple Labels: Each instance can belong to zero or more classes or categories. This means that an instance can be associated with multiple labels simultaneously.
Imbalanced Dataset
import torch
from torch.utils.data import Dataset
from torchvision import transforms
from PIL import Image
from torch.utils.data import DataLoader
import os
import pandas as pd
from torch.utils.data import random_split
import torch.nn.functional as F
from sklearn.utils.class_weight import compute_class_weight
import numpy as np
from sklearn.metrics import precision_recall_fscore_support
from sklearn.metrics import f1_scorebuild class weights for imbalanced classes
from sklearn.utils.class_weight import compute_class_weight
from sklearn.preprocessing import MultiLabelBinarizer
def generate_class_weights(class_series, class_labels):
mlb = None
n_samples = len(class_series)
n_classes = len(class_series[0])
class_count = [0] * n_classes
for classes in class_series:
for index in range(n_classes):
if classes[index] != 0:
class_count[index] += 1
class_weights = [n_samples / (n_classes * freq) if freq > 0 else 1 for freq in class_count]
class_labels = range(len(class_weights)) if mlb is None else mlb.classes_
return dict(zip(class_labels, class_weights))class CelebADataset(Dataset):
def __init__(self, data, labels, transform=None):
self.data = data
self.labels = labels
self.transform = transform
self.classes = list(pd.read_csv(label_path).columns)
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
img = Image.open(self.data[idx]).convert('RGB')
label = torch.Tensor(self.labels[idx])
if self.transform:
img = self.transform(img)
sample = {'image': img, 'label': label}
return sampledef CelebA_DataLoader(batch_size, device):
num_workers = 0 if device.type == 'cuda' else 2
pin_memory = True if device.type == 'cuda' else False
classes = train_dataset.classes
train_loader = DataLoader(train_dataset, batch_size=batch_size, num_workers=num_workers, pin_memory=pin_memory,shuffle=True)
val_loader = DataLoader(val_dataset, num_workers=num_workers, pin_memory=pin_memory, batch_size=batch_size, shuffle=False)
test_loader = DataLoader(test_dataset, num_workers=num_workers, pin_memory=pin_memory, batch_size=batch_size, shuffle=False)
return train_loader, val_loader, test_loaderimage_folder_path = r'Z:\ndm\Img-20240114T105609Z-001\Img\img_align_celeba\img_align_celeba'
label_path = r'attribute.csv'
eval_path = r'eval.csv'eval_list = pd.read_csv(eval_path)['eval'].values
eval_name = pd.read_csv(eval_path)['name'].values
labels = pd.read_csv(label_path).valuesindx, indy, recall = [0]*3, [0]*3, 0
for i in eval_list:
if recall == i - 1:
recall = i
indy[recall] += indy[recall - 1] + 1
indx[recall] = indy[recall]
else:
indy[recall] += 1
print(indx,'\n',indy)[0, 162771, 182638]
[162770, 182637, 202599]
We'll reduce the dataset to 50,000 items to save time, as the current volume is quite large.
train_list = [os.path.join(image_folder_path, name) for name in eval_name[indx[0]:50000]]
train_label_list = labels[indx[0]:50000]
val_list = [os.path.join(image_folder_path, name) for name in eval_name[indx[1]:indx[1] + 5000]]
val_label_list = labels[indx[1]:(indx[1] + 5000)]
test_list = [os.path.join(image_folder_path, name) for name in eval_name[indx[2]:indx[2] + 5000]]
test_label_list = labels[indx[2]:(indx[2] + 5000)]Data Transforms
image_size = (64, 64)
data_transform=transforms.Compose([
transforms.Resize(image_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.5, 0.5, 0.5],
std=[0.5, 0.5, 0.5])
])train_dataset = CelebADataset(train_list, train_label_list, data_transform)
val_dataset = CelebADataset(val_list, val_label_list, data_transform)
test_dataset = CelebADataset(test_list, test_label_list, data_transform)calculating class weight
class_weights = generate_class_weights(train_label_list, train_dataset.classes)
class_weights{0: 0.2255096518130976,
1: 0.09460379928857943,
2: 0.04858142246404975,
3: 0.12315270935960591,
4: 1.1130899376669634,
5: 0.16240093542938808,
6: 0.1044451871657754,
7: 0.10559216083797938,
8: 0.1041493084485919,
9: 0.1696755802904846,
10: 0.4767353165522502,
11: 0.1227656648988411,
12: 0.17470300489168414,
13: 0.4334257975034674,
14: 0.5441880713974749,
15: 0.38461538461538464,
16: 0.3950695322376738,
17: 0.6024096385542169,
18: 0.06502965352200603,
19: 0.05559508984166518,
20: 0.05968011458582,
21: 0.05205080158234437,
22: 0.6151574803149606,
23: 0.2132741852926122,
24: 0.02997961386257345,
25: 0.08831425745372333,
26: 0.5781683626271971,
27: 0.09078364441862154,
28: 0.31985670419651996,
29: 0.39123630672926446,
30: 0.4407616361071932,
31: 0.05243728500713147,
32: 0.11974327042820193,
33: 0.0779982528391364,
34: 0.13427865506499087,
35: 0.5087505087505088,
36: 0.05327309921581998,
37: 0.20685090186993216,
38: 0.3404139433551198,
39: 0.0320159823784033}
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")import torch
import torch.nn as nn
from torchvision.models import resnet50, ResNet50_Weights
import torch.optim as optim
from sklearn.metrics import accuracy_score
from sklearn.metrics import f1_score
from sklearn.metrics import multilabel_confusion_matrix
import matplotlib.pyplot as pltThe Hamming score calculates the ratio of the number of correctly predicted labels to the total number of labels, without considering the differences between the prediction and the ground truth for each data sample
def hamming_score(y_true, y_pred):
num_samples = len(y_true)
total_correct = 0
for true_labels, pred_labels in zip(y_true, y_pred):
correct_labels = (true_labels == pred_labels).sum()
total_correct += correct_labels
hamming_score = total_correct / (num_samples * len(y_true[0]))
return hamming_score
def get_multilabel_evaluation(model, test_loader):
all_predictions = []
all_targets = []
model.eval()
with torch.no_grad():
for dir_ in test_loader:
inputs, targets = dir_.values()
inputs = inputs.to(device)
targets = targets.to(device)
outputs = model(inputs)
predictions = (outputs > 0.01).float()
all_predictions.append(predictions.cpu().numpy())
all_targets.append(targets.cpu().numpy())
return all_predictions, all_targetsPlot Evaluation show Losses, Hamming Score of Training and Validation in processing
import matplotlib.pyplot as plt
def plot_evaluation(train_losses, train_hamming_scores, val_losses, val_hamming_scores, precision, recall, f_score):
fig, axes = plt.subplots(1, 3, figsize=(15, 5))
# Plot Losses
axes[0].plot(train_losses, label='Train Losses', marker='o')
axes[0].plot(val_losses, label='Validation Losses', marker='o')
axes[0].set_xlabel('Epochs')
axes[0].set_ylabel('Loss')
axes[0].set_title('Train and Validation Losses vs. Epochs')
axes[0].legend()
axes[0].grid(True)
# Plot Hamming Scores
axes[1].plot(train_hamming_scores, label='Train Hamming Scores', marker='o')
axes[1].plot(val_hamming_scores, label='Validation Hamming Scores', marker='o')
axes[1].set_xlabel('Epochs')
axes[1].set_ylabel('Hamming Score')
axes[1].set_title('Train and Validation Hamming Scores')
axes[1].legend()
axes[1].grid(True)
# Plot Precision, Recall, F-Score
metrics = ['Precision', 'Recall', 'F-Score']
values = [precision, recall, f_score]
axes[2].bar(metrics, values, color=['blue', 'green', 'red'])
axes[2].set_title('Score')
axes[2].set_ylabel('Score')
axes[2].grid(axis='y')
axes[2].set_ylim(0, 1)
plt.tight_layout()
plt.show()from tqdm import tqdm
def train_model(model, device, train_loader, val_loader, criterion, optimizer, num_epochs=5):
train_losses = []
train_hamming_scores = []
val_losses = []
val_hamming_scores = []
for epoch in range(num_epochs):
model.train()
running_loss = 0.0
total_train_loss = 0
train_bar = tqdm(train_loader, desc=f'Epoch {epoch+1}/{num_epochs}', unit='batch')
for dict_ in train_bar:
inputs, labels = dict_.values()
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
predicted_labels = (outputs > 0.5).float()
hamming_score_value = hamming_score(labels.cpu().numpy(), predicted_labels.cpu().numpy())
train_bar.set_postfix(loss=loss.item(), hamming_score=hamming_score_value)
train_loss = running_loss / len(train_loader)
train_losses.append(train_loss)
train_hamming_scores.append(hamming_score_value)
model.eval()
total_val_loss = 0
with torch.no_grad():
for dict_ in val_loader:
inputs, labels = dict_.values()
inputs, labels = inputs.to(device), labels.to(device)
outputs = model(inputs)
loss = criterion(outputs, labels)
total_val_loss += loss.item() * inputs.size(0)
predicted_labels = (outputs > 0.5).float()
val_hamming_score = hamming_score(labels.cpu().numpy(), predicted_labels.cpu().numpy())
val_loss = total_val_loss / len(val_loader.dataset)
val_losses.append(val_loss)
val_hamming_scores.append(val_hamming_score)
print(f"Epoch [{epoch+1}/{num_epochs}], Train Loss: {train_loss:.4f}, Train Hamming Score: {hamming_score_value:.4f}, Val Loss: {val_loss:.4f}, Val Hamming Score: {val_hamming_score:.4f}")
return [train_losses, train_hamming_scores, val_losses, val_hamming_scores]To train the model, we will utilize the ResNet50 architecture with default weights.
Additionally, we'll apply precomputed class weights to handle any class imbalances. We'll then establish evaluation metrics to assess the training process and determine which parameters perform the best.
This process will span 10 epochs, iterating through both the training loader and validation loader.
def define_model(train_loader, val_loader, learning_rate, num_features, dropout_prob, device):
Model = resnet50(weights=ResNet50_Weights.DEFAULT)
Model.fc = nn.Sequential(
nn.Linear(Model.fc.in_features, num_features),
nn.ReLU(inplace=True),
nn.Dropout(dropout_prob),
nn.Linear(num_features, len(train_dataset.classes))
)
Model = Model.to(device)
weight_tensor = torch.tensor(list(class_weights.values()), dtype=torch.float).to(device)
criterion = nn.BCEWithLogitsLoss(weight=weight_tensor)
optimizer = optim.Adam(Model.parameters(), lr=learning_rate)
Metric = []
Metric.append(train_model(Model, device, train_loader, val_loader, criterion, optimizer, num_epochs=10))
true, pred = get_multilabel_evaluation(Model, test_loader)
predictions_np = np.concatenate(pred)
targets_np = np.concatenate(true)
precision, recall, f_score, _ = precision_recall_fscore_support(targets_np, predictions_np, average='weighted')
Metric.append([precision, recall, f_score])
return Model, Metricimport matplotlib.pyplot as plt
def get_activation_maps(model, input_image):
activation_maps = []
activations = []
def hook(module, input, output):
activations.append(output)
hooks = []
for layer in model.children():
if isinstance(layer, nn.Conv2d):
hooks.append(layer.register_forward_hook(hook))
with torch.no_grad():
_ = model(input_image)
for activation in activations:
activation_maps.append(activation.squeeze().cpu())
for hook in hooks:
hook.remove()
return activation_mapsModel = resnet50(weights=ResNet50_Weights.DEFAULT)
Model.eval()
image = Image.open(image_path).convert("RGB")
image_tensor = data_transform(image)
image_tensor = image_tensor.unsqueeze(0)
activation_maps = get_activation_maps(Model, image_tensor)show_images(activation_maps[0])
By examining the activation maps at different convolutional layers, you can identify specific features that the neural network has found in the images.
We will perform fine-tuning to find the parameters that the model performs best on
parameters = {
'batch_size': [64, 128],
'learning_rate': [0.01, 0.001],
'num_features': [512, 1024],
'drop_out': [0.2, 0.5]
}Metric = {}
for batch_size in parameters['batch_size']:
train_loader, val_loader, test_loader = CelebA_DataLoader(batch_size, device)
for learning_rate in parameters['learning_rate']:
for num_features in parameters['num_features']:
for drop_out in parameters['drop_out']:
name_ = 'CelebA' + '_' + str(batch_size) + '_' + str(learning_rate) + '_' + str(num_features) + '_' + str(drop_out) + '.pth'
print(f'Model: | batch_size = {batch_size} learning_rate = {learning_rate}, num_features = {num_features}, drop_out = {drop_out} |')
Model, Metric[name_] = define_model(train_loader, val_loader, learning_rate, num_features, drop_out, device)
torch.save(Model.state_dict(), name_)Model: | batch_size = 64 learning_rate = 0.01, num_features = 512, drop_out = 0.2 |
Epoch 1/10: 100%|███████████| 157/157 [00:21<00:00, 7.19batch/s, hamming_score=0.805, loss=0.0757]
Epoch [1/10], Train Loss: 0.0731, Train Hamming Score: 0.8047, Val Loss: 0.0699, Val Hamming Score: 0.8141
Epoch 2/10: 100%|████████████| 157/157 [00:21<00:00, 7.20batch/s, hamming_score=0.82, loss=0.0677]
Epoch [2/10], Train Loss: 0.0652, Train Hamming Score: 0.8203, Val Loss: 0.0794, Val Hamming Score: 0.8141
Epoch 3/10: 100%|████████████| 157/157 [00:21<00:00, 7.19batch/s, hamming_score=0.85, loss=0.0506]
Epoch [3/10], Train Loss: 0.0589, Train Hamming Score: 0.8500, Val Loss: 0.0597, Val Hamming Score: 0.8281
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.828, loss=0.0656]
Epoch [4/10], Train Loss: 0.0566, Train Hamming Score: 0.8281, Val Loss: 0.1245, Val Hamming Score: 0.8203
Epoch 5/10: 100%|████████████| 157/157 [00:22<00:00, 7.13batch/s, hamming_score=0.853, loss=0.047]
Epoch [5/10], Train Loss: 0.0544, Train Hamming Score: 0.8531, Val Loss: 0.0565, Val Hamming Score: 0.8438
Epoch 6/10: 100%|███████████| 157/157 [00:21<00:00, 7.15batch/s, hamming_score=0.838, loss=0.0626]
Epoch [6/10], Train Loss: 0.0573, Train Hamming Score: 0.8375, Val Loss: 0.0576, Val Hamming Score: 0.8516
Epoch 7/10: 100%|███████████| 157/157 [00:21<00:00, 7.15batch/s, hamming_score=0.838, loss=0.0453]
Epoch [7/10], Train Loss: 0.0559, Train Hamming Score: 0.8375, Val Loss: 0.0731, Val Hamming Score: 0.8469
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.11batch/s, hamming_score=0.852, loss=0.0439]
Epoch [8/10], Train Loss: 0.0537, Train Hamming Score: 0.8516, Val Loss: 0.0540, Val Hamming Score: 0.8375
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 7.11batch/s, hamming_score=0.842, loss=0.0531]
Epoch [9/10], Train Loss: 0.0513, Train Hamming Score: 0.8422, Val Loss: 0.0532, Val Hamming Score: 0.8484
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.09batch/s, hamming_score=0.831, loss=0.0638]
Epoch [10/10], Train Loss: 0.0507, Train Hamming Score: 0.8313, Val Loss: 0.0527, Val Hamming Score: 0.8469
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 64 learning_rate = 0.01, num_features = 512, drop_out = 0.5 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.07batch/s, hamming_score=0.802, loss=0.0868]
Epoch [1/10], Train Loss: 0.0770, Train Hamming Score: 0.8016, Val Loss: 0.0778, Val Hamming Score: 0.8125
Epoch 2/10: 100%|███████████| 157/157 [00:22<00:00, 7.07batch/s, hamming_score=0.827, loss=0.0543]
Epoch [2/10], Train Loss: 0.0687, Train Hamming Score: 0.8266, Val Loss: 0.0682, Val Hamming Score: 0.8344
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.842, loss=0.0688]
Epoch [3/10], Train Loss: 0.0634, Train Hamming Score: 0.8422, Val Loss: 0.1136, Val Hamming Score: 0.8094
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.847, loss=0.0608]
Epoch [4/10], Train Loss: 0.0608, Train Hamming Score: 0.8469, Val Loss: 0.0645, Val Hamming Score: 0.8422
Epoch 5/10: 100%|███████████| 157/157 [00:22<00:00, 7.06batch/s, hamming_score=0.806, loss=0.0577]
Epoch [5/10], Train Loss: 0.0580, Train Hamming Score: 0.8063, Val Loss: 0.0573, Val Hamming Score: 0.8328
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 6.99batch/s, hamming_score=0.841, loss=0.0663]
Epoch [6/10], Train Loss: 0.0557, Train Hamming Score: 0.8406, Val Loss: 0.0619, Val Hamming Score: 0.8391
Epoch 7/10: 100%|███████████| 157/157 [00:22<00:00, 7.06batch/s, hamming_score=0.828, loss=0.0598]
Epoch [7/10], Train Loss: 0.0572, Train Hamming Score: 0.8281, Val Loss: 0.0581, Val Hamming Score: 0.8375
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.07batch/s, hamming_score=0.797, loss=0.0739]
Epoch [8/10], Train Loss: 0.0569, Train Hamming Score: 0.7969, Val Loss: 0.0547, Val Hamming Score: 0.8438
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.858, loss=0.0482]
Epoch [9/10], Train Loss: 0.0547, Train Hamming Score: 0.8578, Val Loss: 0.0537, Val Hamming Score: 0.8422
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.848, loss=0.0704]
Epoch [10/10], Train Loss: 0.0536, Train Hamming Score: 0.8484, Val Loss: 0.0532, Val Hamming Score: 0.8391
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 64 learning_rate = 0.01, num_features = 1024, drop_out = 0.2 |
Epoch 1/10: 100%|████████████| 157/157 [00:22<00:00, 6.97batch/s, hamming_score=0.82, loss=0.0614]
Epoch [1/10], Train Loss: 0.0670, Train Hamming Score: 0.8203, Val Loss: 0.0573, Val Hamming Score: 0.8375
Epoch 2/10: 100%|██████████████| 157/157 [00:22<00:00, 6.89batch/s, hamming_score=0.83, loss=0.06]
Epoch [2/10], Train Loss: 0.0536, Train Hamming Score: 0.8297, Val Loss: 0.0549, Val Hamming Score: 0.8391
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 6.98batch/s, hamming_score=0.844, loss=0.0597]
Epoch [3/10], Train Loss: 0.0500, Train Hamming Score: 0.8438, Val Loss: 0.0505, Val Hamming Score: 0.8609
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.878, loss=0.0387]
Epoch [4/10], Train Loss: 0.0475, Train Hamming Score: 0.8781, Val Loss: 0.0488, Val Hamming Score: 0.8531
Epoch 5/10: 100%|███████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.881, loss=0.0342]
Epoch [5/10], Train Loss: 0.0457, Train Hamming Score: 0.8812, Val Loss: 0.0476, Val Hamming Score: 0.8609
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 6.99batch/s, hamming_score=0.863, loss=0.0545]
Epoch [6/10], Train Loss: 0.0448, Train Hamming Score: 0.8625, Val Loss: 0.0472, Val Hamming Score: 0.8656
Epoch 7/10: 100%|███████████| 157/157 [00:22<00:00, 6.97batch/s, hamming_score=0.887, loss=0.0402]
Epoch [7/10], Train Loss: 0.0433, Train Hamming Score: 0.8875, Val Loss: 0.0481, Val Hamming Score: 0.8750
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.863, loss=0.0635]
Epoch [8/10], Train Loss: 0.0422, Train Hamming Score: 0.8625, Val Loss: 0.0456, Val Hamming Score: 0.8797
Epoch 9/10: 100%|████████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.875, loss=0.038]
Epoch [9/10], Train Loss: 0.0415, Train Hamming Score: 0.8750, Val Loss: 0.0466, Val Hamming Score: 0.8797
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 6.98batch/s, hamming_score=0.866, loss=0.0384]
Epoch [10/10], Train Loss: 0.0400, Train Hamming Score: 0.8656, Val Loss: 0.0455, Val Hamming Score: 0.8781
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 64 learning_rate = 0.01, num_features = 1024, drop_out = 0.5 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.825, loss=0.0561]
Epoch [1/10], Train Loss: 0.0677, Train Hamming Score: 0.8250, Val Loss: 0.1674, Val Hamming Score: 0.8313
Epoch 2/10: 100%|███████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.848, loss=0.0499]
Epoch [2/10], Train Loss: 0.0590, Train Hamming Score: 0.8484, Val Loss: 0.6921, Val Hamming Score: 0.8109
Epoch 3/10: 100%|████████████| 157/157 [00:22<00:00, 6.91batch/s, hamming_score=0.83, loss=0.0618]
Epoch [3/10], Train Loss: 0.0570, Train Hamming Score: 0.8297, Val Loss: 0.0544, Val Hamming Score: 0.8484
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 6.94batch/s, hamming_score=0.834, loss=0.0493]
Epoch [4/10], Train Loss: 0.0540, Train Hamming Score: 0.8344, Val Loss: 0.0529, Val Hamming Score: 0.8531
Epoch 5/10: 100%|███████████| 157/157 [00:22<00:00, 6.96batch/s, hamming_score=0.836, loss=0.0609]
Epoch [5/10], Train Loss: 0.0536, Train Hamming Score: 0.8359, Val Loss: 0.1815, Val Hamming Score: 0.8016
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.811, loss=0.0567]
Epoch [6/10], Train Loss: 0.0585, Train Hamming Score: 0.8109, Val Loss: 0.0559, Val Hamming Score: 0.8391
Epoch 7/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.836, loss=0.0575]
Epoch [7/10], Train Loss: 0.0547, Train Hamming Score: 0.8359, Val Loss: 0.0530, Val Hamming Score: 0.8391
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 6.97batch/s, hamming_score=0.852, loss=0.0563]
Epoch [8/10], Train Loss: 0.0532, Train Hamming Score: 0.8516, Val Loss: 0.0529, Val Hamming Score: 0.8500
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 6.96batch/s, hamming_score=0.839, loss=0.0517]
Epoch [9/10], Train Loss: 0.0523, Train Hamming Score: 0.8391, Val Loss: 0.0604, Val Hamming Score: 0.8391
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.844, loss=0.0594]
Epoch [10/10], Train Loss: 0.0508, Train Hamming Score: 0.8438, Val Loss: 0.0494, Val Hamming Score: 0.8594
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 64 learning_rate = 0.001, num_features = 512, drop_out = 0.2 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.877, loss=0.0485]
Epoch [1/10], Train Loss: 0.0530, Train Hamming Score: 0.8766, Val Loss: 0.0426, Val Hamming Score: 0.8797
Epoch 2/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.887, loss=0.0437]
Epoch [2/10], Train Loss: 0.0393, Train Hamming Score: 0.8875, Val Loss: 0.0410, Val Hamming Score: 0.8750
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.906, loss=0.0306]
Epoch [3/10], Train Loss: 0.0347, Train Hamming Score: 0.9062, Val Loss: 0.0454, Val Hamming Score: 0.8766
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.897, loss=0.0265]
Epoch [4/10], Train Loss: 0.0310, Train Hamming Score: 0.8969, Val Loss: 0.0419, Val Hamming Score: 0.8828
Epoch 5/10: 100%|█████████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.9, loss=0.0327]
Epoch [5/10], Train Loss: 0.0284, Train Hamming Score: 0.9000, Val Loss: 0.0423, Val Hamming Score: 0.8922
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 7.06batch/s, hamming_score=0.925, loss=0.0245]
Epoch [6/10], Train Loss: 0.0254, Train Hamming Score: 0.9250, Val Loss: 0.0474, Val Hamming Score: 0.8859
Epoch 7/10: 100%|████████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.903, loss=0.023]
Epoch [7/10], Train Loss: 0.0236, Train Hamming Score: 0.9031, Val Loss: 0.0466, Val Hamming Score: 0.8969
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.909, loss=0.0216]
Epoch [8/10], Train Loss: 0.0209, Train Hamming Score: 0.9094, Val Loss: 0.0500, Val Hamming Score: 0.8859
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 7.07batch/s, hamming_score=0.923, loss=0.0227]
Epoch [9/10], Train Loss: 0.0189, Train Hamming Score: 0.9234, Val Loss: 0.0500, Val Hamming Score: 0.8906
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.909, loss=0.0226]
Epoch [10/10], Train Loss: 0.0171, Train Hamming Score: 0.9094, Val Loss: 0.0517, Val Hamming Score: 0.8938
Model: | batch_size = 64 learning_rate = 0.001, num_features = 512, drop_out = 0.5 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.863, loss=0.0479]
Epoch [1/10], Train Loss: 0.0552, Train Hamming Score: 0.8625, Val Loss: 0.0439, Val Hamming Score: 0.8766
Epoch 2/10: 100%|████████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.847, loss=0.053]
Epoch [2/10], Train Loss: 0.0419, Train Hamming Score: 0.8469, Val Loss: 0.0437, Val Hamming Score: 0.8828
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.867, loss=0.0412]
Epoch [3/10], Train Loss: 0.0374, Train Hamming Score: 0.8672, Val Loss: 0.0417, Val Hamming Score: 0.8953
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.06batch/s, hamming_score=0.884, loss=0.0406]
Epoch [4/10], Train Loss: 0.0346, Train Hamming Score: 0.8844, Val Loss: 0.0405, Val Hamming Score: 0.8906
Epoch 5/10: 100%|███████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.887, loss=0.0341]
Epoch [5/10], Train Loss: 0.0321, Train Hamming Score: 0.8875, Val Loss: 0.0453, Val Hamming Score: 0.8797
Epoch 6/10: 100%|████████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.877, loss=0.035]
Epoch [6/10], Train Loss: 0.0301, Train Hamming Score: 0.8766, Val Loss: 0.0423, Val Hamming Score: 0.8828
Epoch 7/10: 100%|███████████| 157/157 [00:22<00:00, 7.04batch/s, hamming_score=0.887, loss=0.0298]
Epoch [7/10], Train Loss: 0.0280, Train Hamming Score: 0.8875, Val Loss: 0.0482, Val Hamming Score: 0.8703
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.898, loss=0.0233]
Epoch [8/10], Train Loss: 0.0263, Train Hamming Score: 0.8984, Val Loss: 0.0443, Val Hamming Score: 0.8969
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.898, loss=0.0355]
Epoch [9/10], Train Loss: 0.0242, Train Hamming Score: 0.8984, Val Loss: 0.0470, Val Hamming Score: 0.8891
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.05batch/s, hamming_score=0.898, loss=0.0306]
Epoch [10/10], Train Loss: 0.0233, Train Hamming Score: 0.8984, Val Loss: 0.0493, Val Hamming Score: 0.8891
Model: | batch_size = 64 learning_rate = 0.001, num_features = 1024, drop_out = 0.2 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.864, loss=0.0409]
Epoch [1/10], Train Loss: 0.0515, Train Hamming Score: 0.8641, Val Loss: 0.0434, Val Hamming Score: 0.9031
Epoch 2/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.897, loss=0.0409]
Epoch [2/10], Train Loss: 0.0387, Train Hamming Score: 0.8969, Val Loss: 0.0410, Val Hamming Score: 0.8688
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.875, loss=0.0394]
Epoch [3/10], Train Loss: 0.0341, Train Hamming Score: 0.8750, Val Loss: 0.0438, Val Hamming Score: 0.8906
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 6.98batch/s, hamming_score=0.916, loss=0.0324]
Epoch [4/10], Train Loss: 0.0307, Train Hamming Score: 0.9156, Val Loss: 0.0415, Val Hamming Score: 0.9016
Epoch 5/10: 100%|█████████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.9, loss=0.0307]
Epoch [5/10], Train Loss: 0.0274, Train Hamming Score: 0.9000, Val Loss: 0.0443, Val Hamming Score: 0.8859
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.903, loss=0.0318]
Epoch [6/10], Train Loss: 0.0255, Train Hamming Score: 0.9031, Val Loss: 0.0608, Val Hamming Score: 0.8875
Epoch 7/10: 100%|████████████| 157/157 [00:22<00:00, 6.99batch/s, hamming_score=0.898, loss=0.028]
Epoch [7/10], Train Loss: 0.0226, Train Hamming Score: 0.8984, Val Loss: 0.0457, Val Hamming Score: 0.8859
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.909, loss=0.0189]
Epoch [8/10], Train Loss: 0.0207, Train Hamming Score: 0.9094, Val Loss: 0.0505, Val Hamming Score: 0.8953
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 6.98batch/s, hamming_score=0.909, loss=0.0181]
Epoch [9/10], Train Loss: 0.0183, Train Hamming Score: 0.9094, Val Loss: 0.0527, Val Hamming Score: 0.8797
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.939, loss=0.0162]
Epoch [10/10], Train Loss: 0.0162, Train Hamming Score: 0.9391, Val Loss: 0.0555, Val Hamming Score: 0.8812
Model: | batch_size = 64 learning_rate = 0.001, num_features = 1024, drop_out = 0.5 |
Epoch 1/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.836, loss=0.0488]
Epoch [1/10], Train Loss: 0.0534, Train Hamming Score: 0.8359, Val Loss: 0.0474, Val Hamming Score: 0.8781
Epoch 2/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.884, loss=0.0376]
Epoch [2/10], Train Loss: 0.0407, Train Hamming Score: 0.8844, Val Loss: 0.0419, Val Hamming Score: 0.8938
Epoch 3/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.898, loss=0.0357]
Epoch [3/10], Train Loss: 0.0362, Train Hamming Score: 0.8984, Val Loss: 0.0406, Val Hamming Score: 0.8984
Epoch 4/10: 100%|███████████| 157/157 [00:22<00:00, 7.01batch/s, hamming_score=0.877, loss=0.0377]
Epoch [4/10], Train Loss: 0.0329, Train Hamming Score: 0.8766, Val Loss: 0.0425, Val Hamming Score: 0.8953
Epoch 5/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.897, loss=0.0339]
Epoch [5/10], Train Loss: 0.0303, Train Hamming Score: 0.8969, Val Loss: 0.0433, Val Hamming Score: 0.8875
Epoch 6/10: 100%|███████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.898, loss=0.0317]
Epoch [6/10], Train Loss: 0.0281, Train Hamming Score: 0.8984, Val Loss: 0.0448, Val Hamming Score: 0.9062
Epoch 7/10: 100%|███████████| 157/157 [00:22<00:00, 7.02batch/s, hamming_score=0.906, loss=0.0255]
Epoch [7/10], Train Loss: 0.0257, Train Hamming Score: 0.9062, Val Loss: 0.0448, Val Hamming Score: 0.8938
Epoch 8/10: 100%|███████████| 157/157 [00:22<00:00, 7.03batch/s, hamming_score=0.914, loss=0.0219]
Epoch [8/10], Train Loss: 0.0238, Train Hamming Score: 0.9141, Val Loss: 0.0457, Val Hamming Score: 0.8875
Epoch 9/10: 100%|███████████| 157/157 [00:22<00:00, 6.98batch/s, hamming_score=0.923, loss=0.0249]
Epoch [9/10], Train Loss: 0.0219, Train Hamming Score: 0.9234, Val Loss: 0.0490, Val Hamming Score: 0.8922
Epoch 10/10: 100%|██████████| 157/157 [00:22<00:00, 7.00batch/s, hamming_score=0.902, loss=0.0244]
Epoch [10/10], Train Loss: 0.0200, Train Hamming Score: 0.9016, Val Loss: 0.0490, Val Hamming Score: 0.8891
Model: | batch_size = 128 learning_rate = 0.01, num_features = 512, drop_out = 0.2 |
Epoch 1/10: 100%|██████████████| 79/79 [00:19<00:00, 4.07batch/s, hamming_score=0.83, loss=0.0507]
Epoch [1/10], Train Loss: 0.0763, Train Hamming Score: 0.8297, Val Loss: 0.0646, Val Hamming Score: 0.8297
Epoch 2/10: 100%|██████████████| 79/79 [00:19<00:00, 4.06batch/s, hamming_score=0.83, loss=0.0715]
Epoch [2/10], Train Loss: 0.0568, Train Hamming Score: 0.8297, Val Loss: 0.0583, Val Hamming Score: 0.8453
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.16batch/s, hamming_score=0.844, loss=0.0511]
Epoch [3/10], Train Loss: 0.0512, Train Hamming Score: 0.8438, Val Loss: 0.0515, Val Hamming Score: 0.8466
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.875, loss=0.0402]
Epoch [4/10], Train Loss: 0.0476, Train Hamming Score: 0.8750, Val Loss: 0.0514, Val Hamming Score: 0.8581
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.881, loss=0.0428]
Epoch [5/10], Train Loss: 0.0456, Train Hamming Score: 0.8812, Val Loss: 0.0487, Val Hamming Score: 0.8653
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.17batch/s, hamming_score=0.864, loss=0.0604]
Epoch [6/10], Train Loss: 0.0436, Train Hamming Score: 0.8641, Val Loss: 0.0464, Val Hamming Score: 0.8703
Epoch 7/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.825, loss=0.0542]
Epoch [7/10], Train Loss: 0.0427, Train Hamming Score: 0.8250, Val Loss: 0.0484, Val Hamming Score: 0.8744
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.878, loss=0.0412]
Epoch [8/10], Train Loss: 0.0413, Train Hamming Score: 0.8781, Val Loss: 0.0444, Val Hamming Score: 0.8797
Epoch 9/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.855, loss=0.0514]
Epoch [9/10], Train Loss: 0.0396, Train Hamming Score: 0.8547, Val Loss: 0.0482, Val Hamming Score: 0.8716
Epoch 10/10: 100%|████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.872, loss=0.0436]
Epoch [10/10], Train Loss: 0.0388, Train Hamming Score: 0.8719, Val Loss: 0.0476, Val Hamming Score: 0.8709
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 128 learning_rate = 0.01, num_features = 512, drop_out = 0.5 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.839, loss=0.0543]
Epoch [1/10], Train Loss: 0.0715, Train Hamming Score: 0.8391, Val Loss: 0.0674, Val Hamming Score: 0.8316
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.836, loss=0.0625]
Epoch [2/10], Train Loss: 0.0578, Train Hamming Score: 0.8359, Val Loss: 0.0557, Val Hamming Score: 0.8434
Epoch 3/10: 100%|██████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.827, loss=0.063]
Epoch [3/10], Train Loss: 0.0538, Train Hamming Score: 0.8266, Val Loss: 0.0539, Val Hamming Score: 0.8522
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.827, loss=0.0571]
Epoch [4/10], Train Loss: 0.0511, Train Hamming Score: 0.8266, Val Loss: 0.0517, Val Hamming Score: 0.8547
Epoch 5/10: 100%|██████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.858, loss=0.047]
Epoch [5/10], Train Loss: 0.0490, Train Hamming Score: 0.8578, Val Loss: 0.0519, Val Hamming Score: 0.8606
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.864, loss=0.0475]
Epoch [6/10], Train Loss: 0.0476, Train Hamming Score: 0.8641, Val Loss: 0.0489, Val Hamming Score: 0.8641
Epoch 7/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.872, loss=0.0576]
Epoch [7/10], Train Loss: 0.0457, Train Hamming Score: 0.8719, Val Loss: 0.0505, Val Hamming Score: 0.8619
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.867, loss=0.0434]
Epoch [8/10], Train Loss: 0.0449, Train Hamming Score: 0.8672, Val Loss: 0.0470, Val Hamming Score: 0.8681
Epoch 9/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.866, loss=0.0372]
Epoch [9/10], Train Loss: 0.0439, Train Hamming Score: 0.8656, Val Loss: 0.0491, Val Hamming Score: 0.8662
Epoch 10/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.88, loss=0.0431]
Epoch [10/10], Train Loss: 0.0422, Train Hamming Score: 0.8797, Val Loss: 0.0449, Val Hamming Score: 0.8769
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 128 learning_rate = 0.01, num_features = 1024, drop_out = 0.2 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.26batch/s, hamming_score=0.805, loss=0.0632]
Epoch [1/10], Train Loss: 0.0755, Train Hamming Score: 0.8047, Val Loss: 0.0634, Val Hamming Score: 0.8375
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.841, loss=0.0584]
Epoch [2/10], Train Loss: 0.0563, Train Hamming Score: 0.8406, Val Loss: 0.0561, Val Hamming Score: 0.8397
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.827, loss=0.0497]
Epoch [3/10], Train Loss: 0.0519, Train Hamming Score: 0.8266, Val Loss: 0.0525, Val Hamming Score: 0.8575
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.852, loss=0.0485]
Epoch [4/10], Train Loss: 0.0486, Train Hamming Score: 0.8516, Val Loss: 0.0494, Val Hamming Score: 0.8503
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.21batch/s, hamming_score=0.864, loss=0.0389]
Epoch [5/10], Train Loss: 0.0465, Train Hamming Score: 0.8641, Val Loss: 0.0519, Val Hamming Score: 0.8622
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.852, loss=0.0517]
Epoch [6/10], Train Loss: 0.0445, Train Hamming Score: 0.8516, Val Loss: 0.0475, Val Hamming Score: 0.8712
Epoch 7/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.883, loss=0.0438]
Epoch [7/10], Train Loss: 0.0429, Train Hamming Score: 0.8828, Val Loss: 0.0464, Val Hamming Score: 0.8678
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.20batch/s, hamming_score=0.884, loss=0.0431]
Epoch [8/10], Train Loss: 0.0416, Train Hamming Score: 0.8844, Val Loss: 0.0500, Val Hamming Score: 0.8694
Epoch 9/10: 100%|█████████████| 79/79 [00:18<00:00, 4.17batch/s, hamming_score=0.889, loss=0.0414]
Epoch [9/10], Train Loss: 0.0403, Train Hamming Score: 0.8891, Val Loss: 0.0476, Val Hamming Score: 0.8647
Epoch 10/10: 100%|████████████| 79/79 [00:18<00:00, 4.21batch/s, hamming_score=0.853, loss=0.0545]
Epoch [10/10], Train Loss: 0.0397, Train Hamming Score: 0.8531, Val Loss: 0.0486, Val Hamming Score: 0.8703
Model: | batch_size = 128 learning_rate = 0.01, num_features = 1024, drop_out = 0.5 |
Epoch 1/10: 100%|██████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.819, loss=0.052]
Epoch [1/10], Train Loss: 0.0761, Train Hamming Score: 0.8187, Val Loss: 0.0632, Val Hamming Score: 0.8297
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.859, loss=0.0468]
Epoch [2/10], Train Loss: 0.0574, Train Hamming Score: 0.8594, Val Loss: 0.0547, Val Hamming Score: 0.8472
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.855, loss=0.0569]
Epoch [3/10], Train Loss: 0.0523, Train Hamming Score: 0.8547, Val Loss: 0.0505, Val Hamming Score: 0.8509
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.863, loss=0.0433]
Epoch [4/10], Train Loss: 0.0494, Train Hamming Score: 0.8625, Val Loss: 0.0509, Val Hamming Score: 0.8525
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.814, loss=0.0539]
Epoch [5/10], Train Loss: 0.0470, Train Hamming Score: 0.8141, Val Loss: 0.0485, Val Hamming Score: 0.8606
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.856, loss=0.0411]
Epoch [6/10], Train Loss: 0.0453, Train Hamming Score: 0.8562, Val Loss: 0.0487, Val Hamming Score: 0.8669
Epoch 7/10: 100%|█████████████| 79/79 [00:19<00:00, 4.05batch/s, hamming_score=0.877, loss=0.0415]
Epoch [7/10], Train Loss: 0.0439, Train Hamming Score: 0.8766, Val Loss: 0.0474, Val Hamming Score: 0.8575
Epoch 8/10: 100%|█████████████| 79/79 [00:19<00:00, 4.06batch/s, hamming_score=0.873, loss=0.0396]
Epoch [8/10], Train Loss: 0.0426, Train Hamming Score: 0.8734, Val Loss: 0.0492, Val Hamming Score: 0.8619
Epoch 9/10: 100%|█████████████| 79/79 [00:19<00:00, 4.02batch/s, hamming_score=0.883, loss=0.0396]
Epoch [9/10], Train Loss: 0.0417, Train Hamming Score: 0.8828, Val Loss: 0.0483, Val Hamming Score: 0.8709
Epoch 10/10: 100%|████████████| 79/79 [00:19<00:00, 4.16batch/s, hamming_score=0.852, loss=0.0475]
Epoch [10/10], Train Loss: 0.0406, Train Hamming Score: 0.8516, Val Loss: 0.0461, Val Hamming Score: 0.8722
C:\Python311\Lib\site-packages\sklearn\metrics\_classification.py:1471: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
_warn_prf(average, modifier, msg_start, len(result))
Model: | batch_size = 128 learning_rate = 0.001, num_features = 512, drop_out = 0.2 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.856, loss=0.0491]
Epoch [1/10], Train Loss: 0.0567, Train Hamming Score: 0.8562, Val Loss: 0.0457, Val Hamming Score: 0.8675
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.863, loss=0.0446]
Epoch [2/10], Train Loss: 0.0397, Train Hamming Score: 0.8625, Val Loss: 0.0395, Val Hamming Score: 0.8881
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.21batch/s, hamming_score=0.906, loss=0.0313]
Epoch [3/10], Train Loss: 0.0334, Train Hamming Score: 0.9062, Val Loss: 0.0434, Val Hamming Score: 0.8900
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.912, loss=0.0247]
Epoch [4/10], Train Loss: 0.0290, Train Hamming Score: 0.9125, Val Loss: 0.0443, Val Hamming Score: 0.8856
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.897, loss=0.0329]
Epoch [5/10], Train Loss: 0.0262, Train Hamming Score: 0.8969, Val Loss: 0.0488, Val Hamming Score: 0.8862
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.925, loss=0.0187]
Epoch [6/10], Train Loss: 0.0230, Train Hamming Score: 0.9250, Val Loss: 0.0489, Val Hamming Score: 0.8953
Epoch 7/10: 100%|██████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.88, loss=0.0333]
Epoch [7/10], Train Loss: 0.0201, Train Hamming Score: 0.8797, Val Loss: 0.0511, Val Hamming Score: 0.8834
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.906, loss=0.0201]
Epoch [8/10], Train Loss: 0.0189, Train Hamming Score: 0.9062, Val Loss: 0.0527, Val Hamming Score: 0.8912
Epoch 9/10: 100%|██████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.909, loss=0.025]
Epoch [9/10], Train Loss: 0.0162, Train Hamming Score: 0.9094, Val Loss: 0.0553, Val Hamming Score: 0.9006
Epoch 10/10: 100%|██████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.9, loss=0.0173]
Epoch [10/10], Train Loss: 0.0146, Train Hamming Score: 0.9000, Val Loss: 0.0553, Val Hamming Score: 0.8959
Model: | batch_size = 128 learning_rate = 0.001, num_features = 512, drop_out = 0.5 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.26batch/s, hamming_score=0.856, loss=0.0499]
Epoch [1/10], Train Loss: 0.0606, Train Hamming Score: 0.8562, Val Loss: 0.0462, Val Hamming Score: 0.8581
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.26batch/s, hamming_score=0.881, loss=0.0376]
Epoch [2/10], Train Loss: 0.0425, Train Hamming Score: 0.8812, Val Loss: 0.0411, Val Hamming Score: 0.8812
Epoch 3/10: 100%|█████████████| 79/79 [00:19<00:00, 4.14batch/s, hamming_score=0.889, loss=0.0373]
Epoch [3/10], Train Loss: 0.0363, Train Hamming Score: 0.8891, Val Loss: 0.0443, Val Hamming Score: 0.8866
Epoch 4/10: 100%|██████████████| 79/79 [00:18<00:00, 4.20batch/s, hamming_score=0.897, loss=0.031]
Epoch [4/10], Train Loss: 0.0322, Train Hamming Score: 0.8969, Val Loss: 0.0426, Val Hamming Score: 0.8866
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.887, loss=0.0325]
Epoch [5/10], Train Loss: 0.0296, Train Hamming Score: 0.8875, Val Loss: 0.0437, Val Hamming Score: 0.8872
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.906, loss=0.0245]
Epoch [6/10], Train Loss: 0.0271, Train Hamming Score: 0.9062, Val Loss: 0.0478, Val Hamming Score: 0.8866
Epoch 7/10: 100%|█████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.909, loss=0.0215]
Epoch [7/10], Train Loss: 0.0250, Train Hamming Score: 0.9094, Val Loss: 0.0445, Val Hamming Score: 0.8878
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.24batch/s, hamming_score=0.933, loss=0.0224]
Epoch [8/10], Train Loss: 0.0224, Train Hamming Score: 0.9328, Val Loss: 0.0476, Val Hamming Score: 0.8931
Epoch 9/10: 100%|██████████████| 79/79 [00:19<00:00, 4.07batch/s, hamming_score=0.92, loss=0.0174]
Epoch [9/10], Train Loss: 0.0214, Train Hamming Score: 0.9203, Val Loss: 0.0500, Val Hamming Score: 0.8891
Epoch 10/10: 100%|████████████| 79/79 [00:19<00:00, 4.00batch/s, hamming_score=0.902, loss=0.0217]
Epoch [10/10], Train Loss: 0.0196, Train Hamming Score: 0.9016, Val Loss: 0.0487, Val Hamming Score: 0.8941
Model: | batch_size = 128 learning_rate = 0.001, num_features = 1024, drop_out = 0.2 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.872, loss=0.0426]
Epoch [1/10], Train Loss: 0.0549, Train Hamming Score: 0.8719, Val Loss: 0.0440, Val Hamming Score: 0.8684
Epoch 2/10: 100%|█████████████| 79/79 [00:18<00:00, 4.18batch/s, hamming_score=0.887, loss=0.0318]
Epoch [2/10], Train Loss: 0.0385, Train Hamming Score: 0.8875, Val Loss: 0.0397, Val Hamming Score: 0.8959
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.884, loss=0.0344]
Epoch [3/10], Train Loss: 0.0326, Train Hamming Score: 0.8844, Val Loss: 0.0424, Val Hamming Score: 0.8934
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.883, loss=0.0375]
Epoch [4/10], Train Loss: 0.0285, Train Hamming Score: 0.8828, Val Loss: 0.0431, Val Hamming Score: 0.8928
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.908, loss=0.0211]
Epoch [5/10], Train Loss: 0.0254, Train Hamming Score: 0.9078, Val Loss: 0.0463, Val Hamming Score: 0.8897
Epoch 6/10: 100%|███████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.9, loss=0.0313]
Epoch [6/10], Train Loss: 0.0219, Train Hamming Score: 0.9000, Val Loss: 0.0470, Val Hamming Score: 0.8925
Epoch 7/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.902, loss=0.0224]
Epoch [7/10], Train Loss: 0.0193, Train Hamming Score: 0.9016, Val Loss: 0.0489, Val Hamming Score: 0.8875
Epoch 8/10: 100%|██████████████| 79/79 [00:18<00:00, 4.21batch/s, hamming_score=0.916, loss=0.022]
Epoch [8/10], Train Loss: 0.0167, Train Hamming Score: 0.9156, Val Loss: 0.0529, Val Hamming Score: 0.8984
Epoch 9/10: 100%|██████████████| 79/79 [00:18<00:00, 4.20batch/s, hamming_score=0.947, loss=0.019]
Epoch [9/10], Train Loss: 0.0147, Train Hamming Score: 0.9469, Val Loss: 0.0582, Val Hamming Score: 0.8900
Epoch 10/10: 100%|████████████| 79/79 [00:18<00:00, 4.17batch/s, hamming_score=0.934, loss=0.0208]
Epoch [10/10], Train Loss: 0.0138, Train Hamming Score: 0.9344, Val Loss: 0.0585, Val Hamming Score: 0.8862
Model: | batch_size = 128 learning_rate = 0.001, num_features = 1024, drop_out = 0.5 |
Epoch 1/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.861, loss=0.0475]
Epoch [1/10], Train Loss: 0.0562, Train Hamming Score: 0.8609, Val Loss: 0.0507, Val Hamming Score: 0.8625
Epoch 2/10: 100%|██████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.898, loss=0.048]
Epoch [2/10], Train Loss: 0.0404, Train Hamming Score: 0.8984, Val Loss: 0.0430, Val Hamming Score: 0.8725
Epoch 3/10: 100%|█████████████| 79/79 [00:18<00:00, 4.20batch/s, hamming_score=0.877, loss=0.0425]
Epoch [3/10], Train Loss: 0.0359, Train Hamming Score: 0.8766, Val Loss: 0.0424, Val Hamming Score: 0.8919
Epoch 4/10: 100%|█████████████| 79/79 [00:18<00:00, 4.23batch/s, hamming_score=0.878, loss=0.0501]
Epoch [4/10], Train Loss: 0.0310, Train Hamming Score: 0.8781, Val Loss: 0.0418, Val Hamming Score: 0.8894
Epoch 5/10: 100%|█████████████| 79/79 [00:18<00:00, 4.22batch/s, hamming_score=0.905, loss=0.0251]
Epoch [5/10], Train Loss: 0.0295, Train Hamming Score: 0.9047, Val Loss: 0.0421, Val Hamming Score: 0.8925
Epoch 6/10: 100%|█████████████| 79/79 [00:18<00:00, 4.21batch/s, hamming_score=0.908, loss=0.0352]
Epoch [6/10], Train Loss: 0.0259, Train Hamming Score: 0.9078, Val Loss: 0.0463, Val Hamming Score: 0.8978
Epoch 7/10: 100%|██████████████| 79/79 [00:18<00:00, 4.25batch/s, hamming_score=0.908, loss=0.036]
Epoch [7/10], Train Loss: 0.0237, Train Hamming Score: 0.9078, Val Loss: 0.0478, Val Hamming Score: 0.8884
Epoch 8/10: 100%|█████████████| 79/79 [00:18<00:00, 4.16batch/s, hamming_score=0.909, loss=0.0252]
Epoch [8/10], Train Loss: 0.0218, Train Hamming Score: 0.9094, Val Loss: 0.0490, Val Hamming Score: 0.8950
Epoch 9/10: 100%|█████████████| 79/79 [00:18<00:00, 4.20batch/s, hamming_score=0.934, loss=0.0262]
Epoch [9/10], Train Loss: 0.0198, Train Hamming Score: 0.9344, Val Loss: 0.0470, Val Hamming Score: 0.8891
Epoch 10/10: 100%|████████████| 79/79 [00:19<00:00, 4.14batch/s, hamming_score=0.909, loss=0.0264]
Epoch [10/10], Train Loss: 0.0182, Train Hamming Score: 0.9094, Val Loss: 0.0545, Val Hamming Score: 0.8978
import json
file_path = "data.json"
with open(file_path, "w") as json_file:
json.dump(Metric, json_file)
print("Data saved to", file_path)Data saved to data.json
import json
with open('data.json', 'r') as file:
data = json.load(file)
name_models = list(data.keys())
for name_ in name_models:
parts = name_.split('_')
batch_size = parts[1]
learning_rate = parts[2]
num_features = parts[3]
dropout = parts[4].replace('.pth','')
print(f'batch Size = {batch_size} | learning Rate = {learning_rate} | num_features = {num_features} | drop out = {dropout}')
train_losses, train_hamming_scores, val_losses, val_hamming_scores = data[name_][0]
precision, recall, f_score = data[name_][1]
plot_evaluation(train_losses, train_hamming_scores, val_losses, val_hamming_scores, precision, recall, f_score)batch Size = 64 | learning Rate = 0.01 | num_features = 512 | drop out = 0.2
batch Size = 64 | learning Rate = 0.01 | num_features = 512 | drop out = 0.5
batch Size = 64 | learning Rate = 0.01 | num_features = 1024 | drop out = 0.2
batch Size = 64 | learning Rate = 0.01 | num_features = 1024 | drop out = 0.5
batch Size = 64 | learning Rate = 0.001 | num_features = 512 | drop out = 0.2
batch Size = 64 | learning Rate = 0.001 | num_features = 512 | drop out = 0.5
batch Size = 64 | learning Rate = 0.001 | num_features = 1024 | drop out = 0.2
batch Size = 64 | learning Rate = 0.001 | num_features = 1024 | drop out = 0.5
batch Size = 128 | learning Rate = 0.01 | num_features = 512 | drop out = 0.2
batch Size = 128 | learning Rate = 0.01 | num_features = 512 | drop out = 0.5
batch Size = 128 | learning Rate = 0.01 | num_features = 1024 | drop out = 0.2
batch Size = 128 | learning Rate = 0.01 | num_features = 1024 | drop out = 0.5
batch Size = 128 | learning Rate = 0.001 | num_features = 512 | drop out = 0.2
batch Size = 128 | learning Rate = 0.001 | num_features = 512 | drop out = 0.5
batch Size = 128 | learning Rate = 0.001 | num_features = 1024 | drop out = 0.2
batch Size = 128 | learning Rate = 0.001 | num_features = 1024 | drop out = 0.5
Through the fine-tuning process, we have carefully evaluated and identified the model that best suits our needs for deployment and testing with external datasets.
Based on the charts, we will use the model with the following parameters: batch size = 128, learning rate = 0.001, num_features = 1024, dropout = 0.2
import torch
def model_load(num_features, dropout_prob, device, name):
Model = resnet50(weights=ResNet50_Weights.DEFAULT)
Model.fc = nn.Sequential(
nn.Linear(Model.fc.in_features, num_features),
nn.ReLU(inplace=True),
nn.Dropout(dropout_prob),
nn.Linear(num_features, 40)
)
Model = Model.to(device)
Model.load_state_dict(torch.load(name))
return Model
Model = model_load(1024, 0.2, 'device', 'CelebA_128_0.001_1024_0.2.pth')from PIL import Image
import torch.nn.functional as F
def predict(image_path):
Model.eval()
image = Image.open(image_path).convert("RGB")
image_tensor = data_transform(image)
image_tensor = image_tensor.unsqueeze(0)
image_tensor = image_tensor.to(device)
with torch.no_grad():
output = Model(image_tensor)
probabilities = F.softmax(output, dim=1)
predicted_classes = torch.zeros_like(probabilities, dtype=torch.int)
threshold = 0.3 # 0.5
predicted_classes[probabilities >= threshold] = 1
predicted = predicted_classes[0].cpu().tolist()
plt.imshow(image)
for ind in range(40):
if predicted[ind] == 1:
print(train_dataset.classes[ind])Let's test on some image of Vietnamese celebrities
predict(r"Z:\ndm\note\test\1.jpg")5_o_Clock_Shadow
Attractive
Big_Lips
Big_Nose
Black_Hair
Bushy_Eyebrows
Male
Mouth_Slightly_Open
Narrow_Eyes
No_Beard
Pointy_Nose
Smiling
Straight_Hair
Young
predict(r"Z:\ndm\note\test\2.jpg")Bags_Under_Eyes
Big_Nose
Goatee
Gray_Hair
Male
Mustache
Sideburns
predict(r"Z:\ndm\note\test\3.png")Big_Nose
Black_Hair
Bushy_Eyebrows
Eyeglasses
High_Cheekbones
Male
No_Beard
Pointy_Nose
Wavy_Hair
Young
predict(r"Z:\ndm\note\test\5.png")Attractive
High_Cheekbones
Narrow_Eyes
No_Beard
Receding_Hairline
Smiling
Wearing_Earrings
Wearing_Lipstick
Wearing_Necklace
Young
- Through the training and experimentation process, it has been demonstrated that this dataset is highly suitable for facial recognition purposes, particularly for security and surveillance applications, or for extracting features for image processing tasks.
- Although the training processes resulted in a reduction of the dataset by 40% compared to the original dataset, experimental results still show that the model performs well in real-world scenarios.
- If there is a desire to further improve the model or embark on larger projects, it is advisable to utilize the entire dataset to optimize performance.