ium_464863/IUM_13.ipynb at c8951d8159c0eae3cecd22c21f98502cfa5335dc

Paweł Łączkowski 2ec0020b28 IUM_13 - project for overleaf

2024-06-10 12:20:34 +02:00

470 KiB

Raw Blame History

Experiments - neural networks in breast cancer classification problem

# Data manipulation
import numpy as np
import pandas as pd

# Data visualization
import matplotlib.pyplot as plt
import seaborn as sns
sns.set_style('whitegrid')

# Data preprocessing
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

# Metrics
from sklearn.metrics import confusion_matrix, classification_report
from sklearn.metrics import accuracy_score

# Deep learning
import torch
import torch.nn as nn
import torch.optim as optim

Methods for visualizing confusion matrix and classification report

# Plot confusion matrix
def plot_confusion_matrix(cm, classes, title='Confusion matrix', cmap='Blues', figsize=(10, 6), axis=None):
    """
    Plot the confusion matrix.
    """
    if axis is None:
        fig, ax = plt.subplots(figsize=figsize)
    else:
        ax = axis
        
    sns.heatmap(cm, annot=True, fmt='d', xticklabels=classes, yticklabels=classes, cmap=cmap, ax=ax)
    
    ax.set_title(title)
    ax.set_xlabel('Predicted label')
    ax.set_ylabel('True label')
    
    if axis is None:
        plt.show()  
    
# Plot classification report
def plot_classification_report(report, title='Classification report', axis=None):
    """
    Plot the classification report.
    """
    if axis is None:
        fig, ax = plt.subplots(figsize=(10, 6))
    else:
        ax = axis
        
    sns.heatmap(pd.DataFrame(report).iloc[:-1, :].T, annot=True, cmap='Blues', ax=ax)
    
    ax.set_title('Classification report')
    ax.set_xlabel('Metrics')
    ax.set_ylabel('Classes')
    
    if axis is None:
        plt.show()

Load data

# Load data
data = pd.read_csv('datasets/data.csv')

# Delete unnecessary columns
data.drop(['id'], axis=1, inplace=True)

data

	diagnosis	radius_mean	texture_mean	perimeter_mean	area_mean	smoothness_mean	compactness_mean	concavity_mean	concave points_mean	symmetry_mean	...	radius_worst	texture_worst	perimeter_worst	area_worst	smoothness_worst	compactness_worst	concavity_worst	concave points_worst	symmetry_worst	fractal_dimension_worst
0	1.0	0.521037	0.022658	0.545989	0.363733	0.593753	0.792037	0.703140	0.731113	0.686364	...	0.620776	0.141525	0.668310	0.450698	0.601136	0.619292	0.568610	0.912027	0.598462	0.418864
1	1.0	0.643144	0.272574	0.615783	0.501591	0.289880	0.181768	0.203608	0.348757	0.379798	...	0.606901	0.303571	0.539818	0.435214	0.347553	0.154563	0.192971	0.639175	0.233590	0.222878
2	1.0	0.601496	0.390260	0.595743	0.449417	0.514309	0.431017	0.462512	0.635686	0.509596	...	0.556386	0.360075	0.508442	0.374508	0.483590	0.385375	0.359744	0.835052	0.403706	0.213433
3	1.0	0.210090	0.360839	0.233501	0.102906	0.811321	0.811361	0.565604	0.522863	0.776263	...	0.248310	0.385928	0.241347	0.094008	0.915472	0.814012	0.548642	0.884880	1.000000	0.773711
4	1.0	0.629893	0.156578	0.630986	0.489290	0.430351	0.347893	0.463918	0.518390	0.378283	...	0.519744	0.123934	0.506948	0.341575	0.437364	0.172415	0.319489	0.558419	0.157500	0.142595
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
564	1.0	0.690000	0.428813	0.678668	0.566490	0.526948	0.296055	0.571462	0.690358	0.336364	...	0.623266	0.383262	0.576174	0.452664	0.461137	0.178527	0.328035	0.761512	0.097575	0.105667
565	1.0	0.622320	0.626987	0.604036	0.474019	0.407782	0.257714	0.337395	0.486630	0.349495	...	0.560655	0.699094	0.520892	0.379915	0.300007	0.159997	0.256789	0.559450	0.198502	0.074315
566	1.0	0.455251	0.621238	0.445788	0.303118	0.288165	0.254340	0.216753	0.263519	0.267677	...	0.393099	0.589019	0.379949	0.230731	0.282177	0.273705	0.271805	0.487285	0.128721	0.151909
567	1.0	0.644564	0.663510	0.665538	0.475716	0.588336	0.790197	0.823336	0.755467	0.675253	...	0.633582	0.730277	0.668310	0.402035	0.619626	0.815758	0.749760	0.910653	0.497142	0.452315
568	0.0	0.036869	0.501522	0.028540	0.015907	0.000000	0.074351	0.000000	0.000000	0.266162	...	0.054287	0.489072	0.043578	0.020497	0.124084	0.036043	0.000000	0.000000	0.257441	0.100682

569 rows × 31 columns

Data preprocessing

# Split data into training and testing sets
X = data.iloc[:, 1:]
y = data.iloc[:, 0]

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Standardize the data
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

# Convert data to PyTorch tensors
X_train = torch.FloatTensor(X_train)
X_test = torch.FloatTensor(X_test)
y_train = torch.FloatTensor(y_train.values).view(-1, 1)
y_test = torch.FloatTensor(y_test.values).view(-1, 1)

# Transfer data to GPU if available
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

X_train = X_train.to(device)
X_test = X_test.to(device)
y_train = y_train.to(device)
y_test = y_test.to(device)

Neural network architectures

# V1
# Three fully connected layers with ReLU activation function
# Output layer with Sigmoid activation function
class NeuralNetworkV1(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(NeuralNetworkV1, self).__init__()
        
        self.fc1 = nn.Linear(input_size, hidden_size)
        self.fc2 = nn.Linear(hidden_size, hidden_size // 2)
        self.fc3 = nn.Linear(hidden_size // 2, 1)

        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        out = self.fc1(x)
        out = self.relu(out)
        out = self.fc2(out)
        out = self.relu(out)
        out = self.fc3(out)
        out = self.sigmoid(out)
        return out

# V2
# Four fully connected layers with ReLU activation function and dropout layers
# Output layer with Sigmoid activation function
class NeuralNetworkV2(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(NeuralNetworkV2, self).__init__()
        
        self.fc1 = nn.Linear(input_size, hidden_size)
        self.dropout1 = nn.Dropout(0.5)
        self.fc2 = nn.Linear(hidden_size, hidden_size // 2)
        self.dropout2 = nn.Dropout(0.5)
        self.fc3 = nn.Linear(hidden_size // 2, hidden_size // 4)
        self.fc4 = nn.Linear(hidden_size // 4, 1)

        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        out = self.fc1(x)
        out = self.relu(out)
        out = self.dropout1(out)
        out = self.fc2(out)
        out = self.relu(out)
        out = self.dropout2(out)
        out = self.fc3(out)
        out = self.relu(out)
        out = self.fc4(out)
        out = self.sigmoid(out)
        return out

# V3
# Four fully connected layers with Leaky ReLU activation function
# Output layer with Sigmoid activation function
class NeuralNetworkV3(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(NeuralNetworkV3, self).__init__()
        
        self.fc1 = nn.Linear(input_size, hidden_size)
        self.fc2 = nn.Linear(hidden_size, hidden_size // 2)
        self.fc3 = nn.Linear(hidden_size // 2, hidden_size // 4)
        self.fc4 = nn.Linear(hidden_size // 4, 1)

        self.leaky_relu = nn.LeakyReLU(0.1)
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        out = self.fc1(x)
        out = self.leaky_relu(out)
        out = self.fc2(out)
        out = self.leaky_relu(out)
        out = self.fc3(out)
        out = self.leaky_relu(out)
        out = self.fc4(out)
        out = self.sigmoid(out)
        return out

# V4
# Two convolutional layers with ReLU activation function and max pooling layers
# Two fully connected layers with ReLU activation function
# Output layer with Sigmoid activation function
class NeuralNetworkV4(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(NeuralNetworkV4, self).__init__()
        
        self.conv1 = nn.Conv1d(1, 16, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv1d(16, 32, kernel_size=3, stride=1, padding=1)
        self.pool = nn.MaxPool1d(kernel_size=2, stride=2, padding=0)
        
        # Calculate the output size after the conv and pooling layers
        conv_output_size = input_size // 2  # After two pooling layers with stride 2
        conv_output_size = conv_output_size // 2  # After the second pooling layer
        
        self.fc1 = nn.Linear(32 * conv_output_size, hidden_size)
        self.fc2 = nn.Linear(hidden_size, 1)

        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        x = x.unsqueeze(1)  # Add channel dimension
        out = self.conv1(x)
        out = self.relu(out)
        out = self.pool(out)
        out = self.conv2(out)
        out = self.relu(out)
        out = self.pool(out)
        out = out.view(out.size(0), -1)  # Flatten the tensor
        out = self.fc1(out)
        out = self.relu(out)
        out = self.fc2(out)
        out = self.sigmoid(out)
        return out

# V5
# LSTM layer with ReLU activation function
# Two fully connected layers with ReLU activation function
# Output layer with Sigmoid activation function
class NeuralNetworkV5(nn.Module):
    def __init__(self, input_size, hidden_size):
        super(NeuralNetworkV5, self).__init__()
        
        self.lstm = nn.LSTM(input_size, hidden_size, batch_first=True)
        self.fc1 = nn.Linear(hidden_size, hidden_size // 2)
        self.fc2 = nn.Linear(hidden_size // 2, 1)

        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()
        
    def forward(self, x):
        out, _ = self.lstm(x)
        out = out[:, -1, :]  # Take the last output of the LSTM
        out = self.fc1(out)
        out = self.relu(out)
        out = self.fc2(out)
        out = self.sigmoid(out)
        return out

Training and evaluation

# Training function
def train(model, X_train, y_train, criterion, optimizer, epochs=100):
    """
    Train the neural network.
    """
    for epoch in range(epochs):
        optimizer.zero_grad()
        y_pred = model(X_train)
        loss = criterion(y_pred, y_train)
        loss.backward()
        optimizer.step()
        
        if (epoch + 1) % 10 == 0:
            print(f'Epoch {epoch + 1}/{epochs}, Loss: {loss.item()}')

# Evaluation function
def evaluate(model, X_test, y_test):
    """
    Evaluate the neural network.
    """
    with torch.no_grad():
        y_pred = model(X_test)
        y_pred = (y_pred > 0.5).float()
        cm = confusion_matrix(y_test.cpu(), y_pred.cpu())
        cr = classification_report(y_test.cpu(), y_pred.cpu(), target_names=['Benign', 'Malignant'], output_dict=True)
        acc = accuracy_score(y_test.cpu(), y_pred.cpu())
        
        return cm, cr, acc

# Neural network parameters
input_size = X_train.shape[1]
hidden_size = 128
learning_rate = 0.001
weight_decay = 0.0001
epochs = 1000

Neural network V1

# Neural network V1
model_v1 = NeuralNetworkV1(input_size, hidden_size).to(device)
criterion_v1 = nn.BCELoss()
optimizer_v1 = optim.Adam(model_v1.parameters(), lr=learning_rate, weight_decay=weight_decay)

# Train the model
train(model_v1, X_train, y_train, criterion_v1, optimizer_v1, epochs)

# Evaluate the model
cm_v1, cr_v1, acc_v1 = evaluate(model_v1, X_test, y_test)

Epoch 10/1000, Loss: 0.4779872000217438
Epoch 20/1000, Loss: 0.27251550555229187
Epoch 30/1000, Loss: 0.15164388716220856
Epoch 40/1000, Loss: 0.10145729035139084
Epoch 50/1000, Loss: 0.07789547741413116
Epoch 60/1000, Loss: 0.06484830379486084
Epoch 70/1000, Loss: 0.05551866441965103
Epoch 80/1000, Loss: 0.04840008169412613
Epoch 90/1000, Loss: 0.042578838765621185
Epoch 100/1000, Loss: 0.03765585273504257
Epoch 110/1000, Loss: 0.03336893767118454
Epoch 120/1000, Loss: 0.02957029454410076
Epoch 130/1000, Loss: 0.02613999880850315
Epoch 140/1000, Loss: 0.023030269891023636
Epoch 150/1000, Loss: 0.02021847851574421
Epoch 160/1000, Loss: 0.01771070621907711
Epoch 170/1000, Loss: 0.015495861880481243
Epoch 180/1000, Loss: 0.013572530820965767
Epoch 190/1000, Loss: 0.011898759752511978
Epoch 200/1000, Loss: 0.010441687889397144
Epoch 210/1000, Loss: 0.009163236245512962
Epoch 220/1000, Loss: 0.00809294544160366
Epoch 230/1000, Loss: 0.007174020167440176
Epoch 240/1000, Loss: 0.006399359088391066
Epoch 250/1000, Loss: 0.005741355009377003
Epoch 260/1000, Loss: 0.005172951612621546
Epoch 270/1000, Loss: 0.004688368644565344
Epoch 280/1000, Loss: 0.004283885937184095
Epoch 290/1000, Loss: 0.003929890692234039
Epoch 300/1000, Loss: 0.0036243184003978968
Epoch 310/1000, Loss: 0.0033553754910826683
Epoch 320/1000, Loss: 0.0031130558345466852
Epoch 330/1000, Loss: 0.0028987200930714607
Epoch 340/1000, Loss: 0.0027084406465291977
Epoch 350/1000, Loss: 0.0025372877717018127
Epoch 360/1000, Loss: 0.002381594618782401
Epoch 370/1000, Loss: 0.002238793997094035
Epoch 380/1000, Loss: 0.002110145753249526
Epoch 390/1000, Loss: 0.001990949036553502
Epoch 400/1000, Loss: 0.0018805447034537792
Epoch 410/1000, Loss: 0.0017787017859518528
Epoch 420/1000, Loss: 0.0016847399529069662
Epoch 430/1000, Loss: 0.001597930327989161
Epoch 440/1000, Loss: 0.001518208417110145
Epoch 450/1000, Loss: 0.0014447758439928293
Epoch 460/1000, Loss: 0.0013784606708213687
Epoch 470/1000, Loss: 0.0013172627659514546
Epoch 480/1000, Loss: 0.0012608648976311088
Epoch 490/1000, Loss: 0.001208683243021369
Epoch 500/1000, Loss: 0.0011611600639298558
Epoch 510/1000, Loss: 0.0011176610132679343
Epoch 520/1000, Loss: 0.0010775947012007236
Epoch 530/1000, Loss: 0.0010414356365799904
Epoch 540/1000, Loss: 0.0010077828774228692
Epoch 550/1000, Loss: 0.000977047486230731
Epoch 560/1000, Loss: 0.0009483486064709723
Epoch 570/1000, Loss: 0.0009219619678333402
Epoch 580/1000, Loss: 0.000897533493116498
Epoch 590/1000, Loss: 0.0008748812833800912
Epoch 600/1000, Loss: 0.0008537117973901331
Epoch 610/1000, Loss: 0.0008338657789863646
Epoch 620/1000, Loss: 0.0008152445661835372
Epoch 630/1000, Loss: 0.000797846878413111
Epoch 640/1000, Loss: 0.0007811780087649822
Epoch 650/1000, Loss: 0.0007652725907973945
Epoch 660/1000, Loss: 0.0007502862135879695
Epoch 670/1000, Loss: 0.0007362824399024248
Epoch 680/1000, Loss: 0.0007233091746456921
Epoch 690/1000, Loss: 0.0007109907455742359
Epoch 700/1000, Loss: 0.0006993163260631263
Epoch 710/1000, Loss: 0.0006883330643177032
Epoch 720/1000, Loss: 0.000678009819239378
Epoch 730/1000, Loss: 0.0006681602098979056
Epoch 740/1000, Loss: 0.0006588594405911863
Epoch 750/1000, Loss: 0.0006500912713818252
Epoch 760/1000, Loss: 0.0006416388787329197
Epoch 770/1000, Loss: 0.0006337724043987691
Epoch 780/1000, Loss: 0.0006261061644181609
Epoch 790/1000, Loss: 0.0006186887621879578
Epoch 800/1000, Loss: 0.0006119576282799244
Epoch 810/1000, Loss: 0.0006054439581930637
Epoch 820/1000, Loss: 0.0005992540973238647
Epoch 830/1000, Loss: 0.0005932800122536719
Epoch 840/1000, Loss: 0.000587515824008733
Epoch 850/1000, Loss: 0.0005819853395223618
Epoch 860/1000, Loss: 0.0005764576490037143
Epoch 870/1000, Loss: 0.0005712246056646109
Epoch 880/1000, Loss: 0.0005661725299432874
Epoch 890/1000, Loss: 0.0005614019464701414
Epoch 900/1000, Loss: 0.0005567952175624669
Epoch 910/1000, Loss: 0.0005523671861737967
Epoch 920/1000, Loss: 0.000548191019333899
Epoch 930/1000, Loss: 0.0005440027453005314
Epoch 940/1000, Loss: 0.0005400135414674878
Epoch 950/1000, Loss: 0.0005360668292269111
Epoch 960/1000, Loss: 0.0005323308287188411
Epoch 970/1000, Loss: 0.0005282927886582911
Epoch 980/1000, Loss: 0.0005240424652583897
Epoch 990/1000, Loss: 0.0005197781720198691
Epoch 1000/1000, Loss: 0.0005156174884177744

# Plot confusion matrix
plot_confusion_matrix(cm_v1, ['Benign', 'Malignant'], title='Confusion matrix - Neural Network V1')

# Plot classification report
plot_classification_report(cr_v1)

Neural network V2

# Neural network V2
model_v2 = NeuralNetworkV2(input_size, hidden_size).to(device)
criterion_v2 = nn.BCELoss()
optimizer_v2 = optim.Adam(model_v2.parameters(), lr=learning_rate, weight_decay=weight_decay)

# Train the model
train(model_v2, X_train, y_train, criterion_v2, optimizer_v2, epochs)

# Evaluate the model
cm_v2, cr_v2, acc_v2 = evaluate(model_v2, X_test, y_test)

Epoch 10/1000, Loss: 0.6100791096687317
Epoch 20/1000, Loss: 0.40138334035873413
Epoch 30/1000, Loss: 0.2066049426794052
Epoch 40/1000, Loss: 0.1198694109916687
Epoch 50/1000, Loss: 0.10492949932813644
Epoch 60/1000, Loss: 0.08525355905294418
Epoch 70/1000, Loss: 0.07265784591436386
Epoch 80/1000, Loss: 0.07437089085578918
Epoch 90/1000, Loss: 0.04634793847799301
Epoch 100/1000, Loss: 0.04539191350340843
Epoch 110/1000, Loss: 0.03791217878460884
Epoch 120/1000, Loss: 0.056155040860176086
Epoch 130/1000, Loss: 0.02974613755941391
Epoch 140/1000, Loss: 0.028519876301288605
Epoch 150/1000, Loss: 0.02841273508965969
Epoch 160/1000, Loss: 0.02827402390539646
Epoch 170/1000, Loss: 0.03137960284948349
Epoch 180/1000, Loss: 0.021595297381281853
Epoch 190/1000, Loss: 0.03367958217859268
Epoch 200/1000, Loss: 0.03138892352581024
Epoch 210/1000, Loss: 0.024735331535339355
Epoch 220/1000, Loss: 0.013547890819609165
Epoch 230/1000, Loss: 0.016778510063886642
Epoch 240/1000, Loss: 0.013662113808095455
Epoch 250/1000, Loss: 0.014643474481999874
Epoch 260/1000, Loss: 0.04232195019721985
Epoch 270/1000, Loss: 0.011198709718883038
Epoch 280/1000, Loss: 0.014255641028285027
Epoch 290/1000, Loss: 0.017376599833369255
Epoch 300/1000, Loss: 0.006715432740747929
Epoch 310/1000, Loss: 0.015104355290532112
Epoch 320/1000, Loss: 0.005779958330094814
Epoch 330/1000, Loss: 0.006878014653921127
Epoch 340/1000, Loss: 0.010289205238223076
Epoch 350/1000, Loss: 0.008154270239174366
Epoch 360/1000, Loss: 0.0052977693267166615
Epoch 370/1000, Loss: 0.0059393011033535
Epoch 380/1000, Loss: 0.003750022267922759
Epoch 390/1000, Loss: 0.006243106909096241
Epoch 400/1000, Loss: 0.0048174685798585415
Epoch 410/1000, Loss: 0.008404634892940521
Epoch 420/1000, Loss: 0.005285304039716721
Epoch 430/1000, Loss: 0.003210554365068674
Epoch 440/1000, Loss: 0.0030219131149351597
Epoch 450/1000, Loss: 0.003663143841549754
Epoch 460/1000, Loss: 0.004113232716917992
Epoch 470/1000, Loss: 0.011188282631337643
Epoch 480/1000, Loss: 0.008383953012526035
Epoch 490/1000, Loss: 0.005484223831444979
Epoch 500/1000, Loss: 0.001833457383327186
Epoch 510/1000, Loss: 0.0026361148338764906
Epoch 520/1000, Loss: 0.0018964618211612105
Epoch 530/1000, Loss: 0.005411419551819563
Epoch 540/1000, Loss: 0.005162812303751707
Epoch 550/1000, Loss: 0.004074939526617527
Epoch 560/1000, Loss: 0.001993684796616435
Epoch 570/1000, Loss: 0.002496592467650771
Epoch 580/1000, Loss: 0.012827489525079727
Epoch 590/1000, Loss: 0.0010587115539237857
Epoch 600/1000, Loss: 0.0020602247677743435
Epoch 610/1000, Loss: 0.0010980992810800672
Epoch 620/1000, Loss: 0.0023741163313388824
Epoch 630/1000, Loss: 0.00123070168774575
Epoch 640/1000, Loss: 0.011475415900349617
Epoch 650/1000, Loss: 0.00989847257733345
Epoch 660/1000, Loss: 0.0012280159862712026
Epoch 670/1000, Loss: 0.0017485406715422869
Epoch 680/1000, Loss: 0.0012420162092894316
Epoch 690/1000, Loss: 0.0004315624828450382
Epoch 700/1000, Loss: 0.0007627215818502009
Epoch 710/1000, Loss: 0.00213691801764071
Epoch 720/1000, Loss: 0.0021272513549774885
Epoch 730/1000, Loss: 0.0009174205479212105
Epoch 740/1000, Loss: 0.0015678246272727847
Epoch 750/1000, Loss: 0.0018770662136375904
Epoch 760/1000, Loss: 0.00043499123421497643
Epoch 770/1000, Loss: 0.001615240820683539
Epoch 780/1000, Loss: 0.0023441719822585583
Epoch 790/1000, Loss: 0.0004717250994872302
Epoch 800/1000, Loss: 0.0006681744707748294
Epoch 810/1000, Loss: 0.0018840441480278969
Epoch 820/1000, Loss: 0.0016851990949362516
Epoch 830/1000, Loss: 0.006307181902229786
Epoch 840/1000, Loss: 0.00034771396894939244
Epoch 850/1000, Loss: 0.0006758614326827228
Epoch 860/1000, Loss: 0.0003922640171367675
Epoch 870/1000, Loss: 0.0011983055155724287
Epoch 880/1000, Loss: 0.0005702194175682962
Epoch 890/1000, Loss: 0.0027277739718556404
Epoch 900/1000, Loss: 0.000657720142044127
Epoch 910/1000, Loss: 0.0007201721309684217
Epoch 920/1000, Loss: 0.002012366196140647
Epoch 930/1000, Loss: 0.0005625162739306688
Epoch 940/1000, Loss: 0.0008252564002759755
Epoch 950/1000, Loss: 0.0021678118500858545
Epoch 960/1000, Loss: 0.002342545660212636
Epoch 970/1000, Loss: 0.0015022775623947382
Epoch 980/1000, Loss: 0.0007447443204000592
Epoch 990/1000, Loss: 0.00044577810331247747
Epoch 1000/1000, Loss: 0.0006745096179656684

# Plot confusion matrix
plot_confusion_matrix(cm_v2, ['Benign', 'Malignant'], title='Confusion matrix - Neural Network V2')

# Plot classification report
plot_classification_report(cr_v2)

Neural network V3

# Neural network V3
model_v3 = NeuralNetworkV3(input_size, hidden_size).to(device)
criterion_v3 = nn.BCELoss()
optimizer_v3 = optim.Adam(model_v3.parameters(), lr=learning_rate, weight_decay=weight_decay)

# Train the model
train(model_v3, X_train, y_train, criterion_v3, optimizer_v3, epochs)

# Evaluate the model
cm_v3, cr_v3, acc_v3 = evaluate(model_v3, X_test, y_test)

Epoch 10/1000, Loss: 0.5781828761100769
Epoch 20/1000, Loss: 0.35954442620277405
Epoch 30/1000, Loss: 0.16433778405189514
Epoch 40/1000, Loss: 0.09435044229030609
Epoch 50/1000, Loss: 0.06897494941949844
Epoch 60/1000, Loss: 0.05602168291807175
Epoch 70/1000, Loss: 0.046810463070869446
Epoch 80/1000, Loss: 0.03961234167218208
Epoch 90/1000, Loss: 0.033818356692790985
Epoch 100/1000, Loss: 0.02865622565150261
Epoch 110/1000, Loss: 0.023877572268247604
Epoch 120/1000, Loss: 0.019604215398430824
Epoch 130/1000, Loss: 0.01610736735165119
Epoch 140/1000, Loss: 0.01334200520068407
Epoch 150/1000, Loss: 0.011029877699911594
Epoch 160/1000, Loss: 0.009049472399055958
Epoch 170/1000, Loss: 0.007442420814186335
Epoch 180/1000, Loss: 0.0061035449616611
Epoch 190/1000, Loss: 0.004539611749351025
Epoch 200/1000, Loss: 0.0030651914421468973
Epoch 210/1000, Loss: 0.0022714021615684032
Epoch 220/1000, Loss: 0.001738564227707684
Epoch 230/1000, Loss: 0.0013824186753481627
Epoch 240/1000, Loss: 0.0011372618610039353
Epoch 250/1000, Loss: 0.0009611304849386215
Epoch 260/1000, Loss: 0.0008317003957927227
Epoch 270/1000, Loss: 0.0007334426045417786
Epoch 280/1000, Loss: 0.0006564902723766863
Epoch 290/1000, Loss: 0.0005941776908002794
Epoch 300/1000, Loss: 0.0005428834119811654
Epoch 310/1000, Loss: 0.0005009892047382891
Epoch 320/1000, Loss: 0.00046646010014228523
Epoch 330/1000, Loss: 0.00043750536860898137
Epoch 340/1000, Loss: 0.0004131169698666781
Epoch 350/1000, Loss: 0.0003923749318346381
Epoch 360/1000, Loss: 0.00037453131517395377
Epoch 370/1000, Loss: 0.0003590169653762132
Epoch 380/1000, Loss: 0.0003454721299931407
Epoch 390/1000, Loss: 0.00033352847094647586
Epoch 400/1000, Loss: 0.0003230379370506853
Epoch 410/1000, Loss: 0.00031372407102026045
Epoch 420/1000, Loss: 0.00030547339702025056
Epoch 430/1000, Loss: 0.0002980876306537539
Epoch 440/1000, Loss: 0.0002914170909207314
Epoch 450/1000, Loss: 0.00028551131254062057
Epoch 460/1000, Loss: 0.000280180451227352
Epoch 470/1000, Loss: 0.00027529962244443595
Epoch 480/1000, Loss: 0.00027088262140750885
Epoch 490/1000, Loss: 0.00026690459344536066
Epoch 500/1000, Loss: 0.0002632274990901351
Epoch 510/1000, Loss: 0.00025982430088333786
Epoch 520/1000, Loss: 0.000256663013715297
Epoch 530/1000, Loss: 0.00025376983103342354
Epoch 540/1000, Loss: 0.0002510569174773991
Epoch 550/1000, Loss: 0.0002485134173184633
Epoch 560/1000, Loss: 0.0002461685216985643
Epoch 570/1000, Loss: 0.00024397078959736973
Epoch 580/1000, Loss: 0.00024182727793231606
Epoch 590/1000, Loss: 0.0002398582291789353
Epoch 600/1000, Loss: 0.00023796758614480495
Epoch 610/1000, Loss: 0.00023617268016096205
Epoch 620/1000, Loss: 0.00023440059158019722
Epoch 630/1000, Loss: 0.00023271011014003307
Epoch 640/1000, Loss: 0.00023108486493583769
Epoch 650/1000, Loss: 0.00022952101426199079
Epoch 660/1000, Loss: 0.00022804134641774
Epoch 670/1000, Loss: 0.00022659693786408752
Epoch 680/1000, Loss: 0.00022519213962368667
Epoch 690/1000, Loss: 0.0002238261658931151
Epoch 700/1000, Loss: 0.0002225149655714631
Epoch 710/1000, Loss: 0.000221233261981979
Epoch 720/1000, Loss: 0.00022013885609339923
Epoch 730/1000, Loss: 0.0002189427614212036
Epoch 740/1000, Loss: 0.00021769681188743562
Epoch 750/1000, Loss: 0.00021648130496032536
Epoch 760/1000, Loss: 0.00021531064703594893
Epoch 770/1000, Loss: 0.00021418495452962816
Epoch 780/1000, Loss: 0.0002131147193722427
Epoch 790/1000, Loss: 0.00021202574134804308
Epoch 800/1000, Loss: 0.0002110681962221861
Epoch 810/1000, Loss: 0.00021014301455579698
Epoch 820/1000, Loss: 0.0002092804352287203
Epoch 830/1000, Loss: 0.00020846931147389114
Epoch 840/1000, Loss: 0.00020768567628692836
Epoch 850/1000, Loss: 0.00020691761164925992
Epoch 860/1000, Loss: 0.00020612987282220274
Epoch 870/1000, Loss: 0.00020533577480819076
Epoch 880/1000, Loss: 0.00020457926439121366
Epoch 890/1000, Loss: 0.00020382019283715636
Epoch 900/1000, Loss: 0.00020311155822128057
Epoch 910/1000, Loss: 0.00020234761177562177
Epoch 920/1000, Loss: 0.00020164126181043684
Epoch 930/1000, Loss: 0.00020095478976145387
Epoch 940/1000, Loss: 0.0002003153640544042
Epoch 950/1000, Loss: 0.00019972550217062235
Epoch 960/1000, Loss: 0.00019920482009183615
Epoch 970/1000, Loss: 0.0001986794959520921
Epoch 980/1000, Loss: 0.00019817189604509622
Epoch 990/1000, Loss: 0.00019766329205594957
Epoch 1000/1000, Loss: 0.00019715774396900088

# Plot confusion matrix
plot_confusion_matrix(cm_v3, ['Benign', 'Malignant'], title='Confusion matrix - Neural Network V3')

# Plot classification report
plot_classification_report(cr_v3)

Neural network V4

# Neural network V4
model_v4 = NeuralNetworkV4(input_size, hidden_size).to(device)
criterion_v4 = nn.BCELoss()
optimizer_v4 = optim.Adam(model_v4.parameters(), lr=learning_rate, weight_decay=weight_decay)

# Train the model
train(model_v4, X_train, y_train, criterion_v4, optimizer_v4, epochs)

# Evaluate the model
cm_v4, cr_v4, acc_v4 = evaluate(model_v4, X_test, y_test)

Epoch 10/1000, Loss: 0.4186047613620758
Epoch 20/1000, Loss: 0.20972707867622375
Epoch 30/1000, Loss: 0.14643631875514984
Epoch 40/1000, Loss: 0.1106308326125145
Epoch 50/1000, Loss: 0.09039844572544098
Epoch 60/1000, Loss: 0.07973706722259521
Epoch 70/1000, Loss: 0.07261230796575546
Epoch 80/1000, Loss: 0.06730187684297562
Epoch 90/1000, Loss: 0.06239919736981392
Epoch 100/1000, Loss: 0.05727291852235794
Epoch 110/1000, Loss: 0.05158916860818863
Epoch 120/1000, Loss: 0.04558803513646126
Epoch 130/1000, Loss: 0.03927084803581238
Epoch 140/1000, Loss: 0.03285034000873566
Epoch 150/1000, Loss: 0.026732975617051125
Epoch 160/1000, Loss: 0.021154126152396202
Epoch 170/1000, Loss: 0.016432005912065506
Epoch 180/1000, Loss: 0.012630755081772804
Epoch 190/1000, Loss: 0.009745683521032333
Epoch 200/1000, Loss: 0.007611896842718124
Epoch 210/1000, Loss: 0.0060510095208883286
Epoch 220/1000, Loss: 0.004891632590442896
Epoch 230/1000, Loss: 0.004030665848404169
Epoch 240/1000, Loss: 0.003379524452611804
Epoch 250/1000, Loss: 0.002879904117435217
Epoch 260/1000, Loss: 0.002489611506462097
Epoch 270/1000, Loss: 0.002184198470786214
Epoch 280/1000, Loss: 0.0019437369192019105
Epoch 290/1000, Loss: 0.0017541276756674051
Epoch 300/1000, Loss: 0.0016026693629100919
Epoch 310/1000, Loss: 0.0014796099858358502
Epoch 320/1000, Loss: 0.0013786517083644867
Epoch 330/1000, Loss: 0.0012941397726535797
Epoch 340/1000, Loss: 0.0012196070747449994
Epoch 350/1000, Loss: 0.0011414645705372095
Epoch 360/1000, Loss: 0.001061757793650031
Epoch 370/1000, Loss: 0.0009867347544059157
Epoch 380/1000, Loss: 0.0009273262694478035
Epoch 390/1000, Loss: 0.0008805892430245876
Epoch 400/1000, Loss: 0.0008431114838458598
Epoch 410/1000, Loss: 0.0008109930204227567
Epoch 420/1000, Loss: 0.0007825929205864668
Epoch 430/1000, Loss: 0.0007562871905975044
Epoch 440/1000, Loss: 0.0007321131997741759
Epoch 450/1000, Loss: 0.0007092245505191386
Epoch 460/1000, Loss: 0.0006881365552544594
Epoch 470/1000, Loss: 0.0006688928115181625
Epoch 480/1000, Loss: 0.0006500301533378661
Epoch 490/1000, Loss: 0.0006336761871352792
Epoch 500/1000, Loss: 0.0006172554567456245
Epoch 510/1000, Loss: 0.00060228758957237
Epoch 520/1000, Loss: 0.0005888384766876698
Epoch 530/1000, Loss: 0.0005763943772763014
Epoch 540/1000, Loss: 0.0005643281037919223
Epoch 550/1000, Loss: 0.0005541218561120331
Epoch 560/1000, Loss: 0.0005427465075626969
Epoch 570/1000, Loss: 0.0005325390957295895
Epoch 580/1000, Loss: 0.0005224572960287333
Epoch 590/1000, Loss: 0.0005137791740708053
Epoch 600/1000, Loss: 0.0005050148465670645
Epoch 610/1000, Loss: 0.0004971114685758948
Epoch 620/1000, Loss: 0.0004892157157883048
Epoch 630/1000, Loss: 0.00048245518701151013
Epoch 640/1000, Loss: 0.0004754861583933234
Epoch 650/1000, Loss: 0.0004669078625738621
Epoch 660/1000, Loss: 0.0004601963155437261
Epoch 670/1000, Loss: 0.00045247498201206326
Epoch 680/1000, Loss: 0.00044596134102903306
Epoch 690/1000, Loss: 0.0004407193046063185
Epoch 700/1000, Loss: 0.000434816291090101
Epoch 710/1000, Loss: 0.00042790695442818105
Epoch 720/1000, Loss: 0.0004230188496876508
Epoch 730/1000, Loss: 0.00041854308801703155
Epoch 740/1000, Loss: 0.0004127133288420737
Epoch 750/1000, Loss: 0.00040834766696207225
Epoch 760/1000, Loss: 0.0004034344747196883
Epoch 770/1000, Loss: 0.00039839293458499014
Epoch 780/1000, Loss: 0.0003941936884075403
Epoch 790/1000, Loss: 0.00039072500658221543
Epoch 800/1000, Loss: 0.0003858723503071815
Epoch 810/1000, Loss: 0.0003812974609900266
Epoch 820/1000, Loss: 0.00037719475221820176
Epoch 830/1000, Loss: 0.0003737532824743539
Epoch 840/1000, Loss: 0.00036979030119255185
Epoch 850/1000, Loss: 0.00036644612555392087
Epoch 860/1000, Loss: 0.0003625146928243339
Epoch 870/1000, Loss: 0.000359336263500154
Epoch 880/1000, Loss: 0.00035612270585261285
Epoch 890/1000, Loss: 0.00035236674011684954
Epoch 900/1000, Loss: 0.0003493966069072485
Epoch 910/1000, Loss: 0.0003470622468739748
Epoch 920/1000, Loss: 0.00034316032542847097
Epoch 930/1000, Loss: 0.00034029941889457405
Epoch 940/1000, Loss: 0.00033748531132005155
Epoch 950/1000, Loss: 0.0003346360463183373
Epoch 960/1000, Loss: 0.00033203166094608605
Epoch 970/1000, Loss: 0.0003288303851149976
Epoch 980/1000, Loss: 0.0003261156671214849
Epoch 990/1000, Loss: 0.000323760905303061
Epoch 1000/1000, Loss: 0.0003211983712390065

# Plot confusion matrix
plot_confusion_matrix(cm_v4, ['Benign', 'Malignant'], title='Confusion matrix - Neural Network V4')

# Plot classification report
plot_classification_report(cr_v4)

Neural network V5

# Neural network V5
model_v5 = NeuralNetworkV5(input_size, hidden_size).to(device)
criterion_v5 = nn.BCELoss()
optimizer_v5 = optim.Adam(model_v5.parameters(), lr=learning_rate, weight_decay=weight_decay)

# Train the model
train(model_v5, X_train.unsqueeze(1), y_train, criterion_v5, optimizer_v5, epochs)

# Evaluate the model
cm_v5, cr_v5, acc_v5 = evaluate(model_v5, X_test.unsqueeze(1), y_test)

Epoch 10/1000, Loss: 0.6472257971763611
Epoch 20/1000, Loss: 0.5392731428146362
Epoch 30/1000, Loss: 0.36916524171829224
Epoch 40/1000, Loss: 0.21396635472774506
Epoch 50/1000, Loss: 0.13106301426887512
Epoch 60/1000, Loss: 0.09312008321285248
Epoch 70/1000, Loss: 0.07527194172143936
Epoch 80/1000, Loss: 0.06560871005058289
Epoch 90/1000, Loss: 0.05922595039010048
Epoch 100/1000, Loss: 0.054361775517463684
Epoch 110/1000, Loss: 0.050068166106939316
Epoch 120/1000, Loss: 0.04590252414345741
Epoch 130/1000, Loss: 0.042265865951776505
Epoch 140/1000, Loss: 0.03903531655669212
Epoch 150/1000, Loss: 0.03611545264720917
Epoch 160/1000, Loss: 0.03336109220981598
Epoch 170/1000, Loss: 0.030720891430974007
Epoch 180/1000, Loss: 0.028099115937948227
Epoch 190/1000, Loss: 0.02529078908264637
Epoch 200/1000, Loss: 0.022276364266872406
Epoch 210/1000, Loss: 0.01945570483803749
Epoch 220/1000, Loss: 0.016959620639681816
Epoch 230/1000, Loss: 0.014633278362452984
Epoch 240/1000, Loss: 0.012433870695531368
Epoch 250/1000, Loss: 0.01063645537942648
Epoch 260/1000, Loss: 0.009215538389980793
Epoch 270/1000, Loss: 0.008075335994362831
Epoch 280/1000, Loss: 0.007146317046135664
Epoch 290/1000, Loss: 0.006382191088050604
Epoch 300/1000, Loss: 0.005759004037827253
Epoch 310/1000, Loss: 0.005265630315989256
Epoch 320/1000, Loss: 0.004859541077166796
Epoch 330/1000, Loss: 0.0045250700786709785
Epoch 340/1000, Loss: 0.00425321189686656
Epoch 350/1000, Loss: 0.0040284693241119385
Epoch 360/1000, Loss: 0.0038447484839707613
Epoch 370/1000, Loss: 0.0036921321880072355
Epoch 380/1000, Loss: 0.003561016172170639
Epoch 390/1000, Loss: 0.003450624644756317
Epoch 400/1000, Loss: 0.003364400239661336
Epoch 410/1000, Loss: 0.0032908162102103233
Epoch 420/1000, Loss: 0.003222778672352433
Epoch 430/1000, Loss: 0.003165710251778364
Epoch 440/1000, Loss: 0.0031129783019423485
Epoch 450/1000, Loss: 0.0030633530113846064
Epoch 460/1000, Loss: 0.0029811717104166746
Epoch 470/1000, Loss: 0.0028529497794806957
Epoch 480/1000, Loss: 0.0026854067109525204
Epoch 490/1000, Loss: 0.0025055729784071445
Epoch 500/1000, Loss: 0.002334937220439315
Epoch 510/1000, Loss: 0.002182029653340578
Epoch 520/1000, Loss: 0.0020450232550501823
Epoch 530/1000, Loss: 0.0019291980424895883
Epoch 540/1000, Loss: 0.0018309173174202442
Epoch 550/1000, Loss: 0.0017435038462281227
Epoch 560/1000, Loss: 0.0016699342522770166
Epoch 570/1000, Loss: 0.0016051153652369976
Epoch 580/1000, Loss: 0.0015472021186724305
Epoch 590/1000, Loss: 0.0014925338327884674
Epoch 600/1000, Loss: 0.0014427873538807034
Epoch 610/1000, Loss: 0.0013984435936436057
Epoch 620/1000, Loss: 0.0013581964885815978
Epoch 630/1000, Loss: 0.00132227991707623
Epoch 640/1000, Loss: 0.0012905021430924535
Epoch 650/1000, Loss: 0.0012572426348924637
Epoch 660/1000, Loss: 0.001225507934577763
Epoch 670/1000, Loss: 0.0012008383637294173
Epoch 680/1000, Loss: 0.001180665334686637
Epoch 690/1000, Loss: 0.001158036757260561
Epoch 700/1000, Loss: 0.0011395757319405675
Epoch 710/1000, Loss: 0.0011222346220165491
Epoch 720/1000, Loss: 0.0011062605772167444
Epoch 730/1000, Loss: 0.0010908363619819283
Epoch 740/1000, Loss: 0.0010754970135167241
Epoch 750/1000, Loss: 0.0010610275203362107
Epoch 760/1000, Loss: 0.0010471524437889457
Epoch 770/1000, Loss: 0.0010340394219383597
Epoch 780/1000, Loss: 0.0010215329239144921
Epoch 790/1000, Loss: 0.0010098188649863005
Epoch 800/1000, Loss: 0.0009981722105294466
Epoch 810/1000, Loss: 0.000987424748018384
Epoch 820/1000, Loss: 0.0009770964970812201
Epoch 830/1000, Loss: 0.0009671879815869033
Epoch 840/1000, Loss: 0.000957623531576246
Epoch 850/1000, Loss: 0.00094812415773049
Epoch 860/1000, Loss: 0.0009392902138642967
Epoch 870/1000, Loss: 0.0009308967855758965
Epoch 880/1000, Loss: 0.0009222882217727602
Epoch 890/1000, Loss: 0.0009139023022726178
Epoch 900/1000, Loss: 0.0009058943251147866
Epoch 910/1000, Loss: 0.0008985912427306175
Epoch 920/1000, Loss: 0.0008904503774829209
Epoch 930/1000, Loss: 0.0008828166173771024
Epoch 940/1000, Loss: 0.0008756622555665672
Epoch 950/1000, Loss: 0.000868525356054306
Epoch 960/1000, Loss: 0.0008616363047622144
Epoch 970/1000, Loss: 0.0008551458013243973
Epoch 980/1000, Loss: 0.0008486591395922005
Epoch 990/1000, Loss: 0.0008415335905738175
Epoch 1000/1000, Loss: 0.0008350368589162827

# Plot confusion matrix
plot_confusion_matrix(cm_v5, ['Benign', 'Malignant'], title='Confusion matrix - Neural Network V5')

# Plot classification report
plot_classification_report(cr_v5)

# Accuracy comparison
accuracies = [acc_v1, acc_v2, acc_v3, acc_v4, acc_v5]
models = ['Neural Network V1', 'Neural Network V2', 'Neural Network V3', 'Neural Network V4', 'Neural Network V5']

fig, ax = plt.subplots(figsize=(10, 6))
sns.barplot(x=models, y=accuracies, hue=models, ax=ax)

# Add labels
for i, v in enumerate(accuracies):
    ax.text(i, v + 0.01, str(round(v, 2)), ha='center', va='bottom')

ax.set_title('Accuracy comparison')
ax.set_xlabel('Model')
ax.set_ylabel('Accuracy')

plt.show()

acc_v4

0.9824561403508771

470 KiB Raw Blame History Unescape Escape

Experiments - neural networks in breast cancer classification problem

Methods for visualizing confusion matrix and classification report

Load data

Data preprocessing

Neural network architectures

Training and evaluation

Neural network V1

Neural network V2

Neural network V3

Neural network V4

Neural network V5

470 KiB

Raw Blame History