File size: 10,442 Bytes
60b0ddc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 |
# Custom imports
from batch_sampler import BatchSampler
from image_dataset import ImageDataset
from net import Net, ResNetModel, EfficientNetModel, EfficientNetModel_b7
from train_test import train_model, test_model
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay,classification_report
from visualise_performance_metrics import create_confusion_matrix, ROC_multiclass
from image_dataset_BINARY import ImageDatasetBINARY
from net_BINARY import Net_BINARY
# Torch imports
import torch
import torch.nn as nn
import torch.optim as optim
from torchsummary import summary
# Other imports
import matplotlib.pyplot as plt
from matplotlib.pyplot import figure
import os
import argparse
import plotext
from datetime import datetime
from pathlib import Path
from typing import List
from sklearn.metrics import accuracy_score, precision_score, recall_score, roc_curve
import numpy as np
def main(args: argparse.Namespace, activeloop: bool = True) -> None:
# NOTE: Uncomment the dataset you would like to use. For instance, if you would like to run the Binary Model,
# you would need to comment the ImageDataset Class lines, or viceversa if you are running the other models.
# Load the train and test data set
train_dataset = ImageDataset(Path('dc1/data/X_train.npy'), Path('dc1/data/Y_train.npy'))
test_dataset = ImageDataset(Path('dc1/data/X_test.npy'), Path('dc1/data/Y_test.npy'))
# Load the BINARY train and BINARY test data set
# train_dataset = ImageDatasetBINARY(Path('dc1/data/X_train.npy'), Path('dc1/data/Y_train.npy'))
# test_dataset = ImageDatasetBINARY(Path('dc1/data/X_test.npy'), Path('dc1/data/Y_test.npy'))
# Load the Neural Net.
# NOTE: set number of distinct labels here
# NOTE: uncomment when you need to use one of the models
# Improved Net
# model = Net(n_classes=6)
# ResNet Pre-Trained Model
# model = ResNetModel(n_classes=6)
# EfficientNet Model Pre-Trained == OUR SELECTED MODEL
model = EfficientNetModel(n_classes=6)
# Binary Model
# model = Net_BINARY(n_classes=2)
# Initialize optimizer(s) and loss function(s)
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.1)
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.1)
loss_function = nn.CrossEntropyLoss()
# fetch epoch and batch count from arguments
n_epochs = args.nb_epochs
batch_size = args.batch_size
# IMPORTANT! Set this to True to see actual errors regarding
# the structure of your model (GPU acceleration hides them)!
# Also make sure you set this to False again for actual model training
# as training your model with GPU-acceleration (CUDA/MPS) is much faster.
DEBUG = False
# fpr = {x:[] for x in range(6)}
# tpr = {x:[] for x in range(6)}
# auc = {}
# Moving our model to the right device (CUDA will speed training up significantly!)
if torch.cuda.is_available() and not DEBUG:
print("@@@ CUDA device found, enabling CUDA training...")
device = "cuda"
model.to(device)
# Creating a summary of our model and its layers:
summary(model, (1, 128, 128), device=device)
elif (
torch.backends.mps.is_available() and not DEBUG
): # PyTorch supports Apple Silicon GPU's from version 1.12
print("@@@ Apple silicon device enabled, training with Metal backend...")
device = "mps"
model.to(device)
else:
print("@@@ No GPU boosting device found, training on CPU...")
device = "cpu"
# Creating a summary of our model and its layers:
summary(model, (1, 128, 128), device=device)
# Lets now train and test our model for multiple epochs:
train_sampler = BatchSampler(
batch_size=batch_size, dataset=train_dataset, balanced=args.balanced_batches
)
test_sampler = BatchSampler(
batch_size=100, dataset=test_dataset, balanced=args.balanced_batches
)
mean_losses_train: List[torch.Tensor] = []
mean_losses_test: List[torch.Tensor] = []
for e in range(n_epochs):
if activeloop:
# Training:
losses = train_model(model, train_sampler, optimizer, loss_function, device)
# Calculating and printing statistics:
mean_loss = sum(losses) / len(losses)
mean_losses_train.append(mean_loss)
print(f"\nEpoch {e + 1} training done, loss on train set: {mean_loss}\n")
# Testing:
# losses, y_pred_probs = test_model(model, test_sampler, loss_function, device)
fpr = {x:[] for x in range(6)}
tpr = {x:[] for x in range(6)}
auc = {}
# # Calculating and printing statistics:
losses, y_pred_probs = test_model(model, test_sampler, loss_function, device, fpr, tpr, auc)
# # Calculating and printing statistics:
mean_loss = sum(losses) / len(losses)
mean_losses_test.append(mean_loss)
print(f"\nEpoch {e + 1} testing done, loss on test set: {mean_loss}\n")
print(auc)
### Plotting during training
plotext.clf()
plotext.scatter(mean_losses_train, label="train")
plotext.scatter(mean_losses_test, label="test")
plotext.title("Train and test loss")
plotext.xticks([i for i in range(len(mean_losses_train) + 1)])
plotext.show()
##################################################################################################################
# R O C C U R V E S
# ##################################################################################################################
# NOTE: If you would like to run the ROC function for the Binary dataset, you would need to comment the following code
# and uncomment the # ROC CURVE FOR BINARY. Additionally, in train_test.py file, in order to make the Binary ROC curve
# you would need to comment some lines specified in the file. Please check the file.
# ROC CURVE FOR MULTICLASS
plt.figure(figsize=(8, 6))
colors = plt.cm.get_cmap('viridis', 6).colors
class_names = ['Class 0 (Atelactasis)','Class 1 (Effusion)', 'Class 2 (Infiltration)', 'Class 3 (No Finding)', 'Class 4 (Nodule)', 'Class 5 (Pneumonia)']
for i, color in zip(range(6), colors):
plt.plot(fpr[i], tpr[i], color=color, lw=2, label='{} (AUC = {:.2f})'.format(class_names[i], auc[i]))
plt.plot([0, 1], [0, 1], color='gray', lw=1, linestyle='--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('ROC Curves for 6 Classes')
plt.legend(loc="lower right")
plt.show()
# ROC CURVE FOR BINARY
# for i in range(2):
# plt.plot(fpr[i], tpr[i], label=f'Class {i} (AUC = {auc[i]:.2f})')
# plt.plot([0, 1], [0, 1], 'k--', label='Random chance')
# plt.xlabel('False Positive Rate')
# plt.ylabel('True Positive Rate')
# plt.title('ROC Curves for 2 Classes (1=Sick, 0=Non-sick)')
# plt.legend(loc='lower right')
# plt.show()
# retrieve current time to label artifacts
now = datetime.now()
# check if model_weights/ subdir exists
if not Path("model_weights/").exists():
os.mkdir(Path("model_weights/"))
if not Path("model_weights/").exists():
os.mkdir(Path("model_weights/"))
# Saving the model
torch.save(model.state_dict(), f"model_weights/model_{now.month:02}{now.day:02}{now.hour}_{now.minute:02}.txt")
torch.save(model.state_dict(), f"model_weights/model_{now.month:02}{now.day:02}{now.hour}_{now.minute:02}.txt")
# Create plot of losses
figure(figsize=(9, 10), dpi=80)
fig, (ax1, ax2) = plt.subplots(2, sharex=True)
ax1.plot(range(1, 1 + n_epochs), [x.detach().cpu() for x in mean_losses_train], label="Train", color="blue")
ax2.plot(range(1, 1 + n_epochs), [x.detach().cpu() for x in mean_losses_test], label="Test", color="red")
fig.legend()
# Check if /artifacts/ subdir exists
if not Path("artifacts/").exists():
os.mkdir(Path("artifacts/"))
if not Path("artifacts/").exists():
os.mkdir(Path("artifacts/"))
# save plot of losses
fig.savefig(Path("artifacts") / f"session_{now.month:02}{now.day:02}{now.hour}_{now.minute:02}.png")
##################################################################################################################
# C O N F U S I O N M A T R I X & C L A S S I F I C A T I O N R E P O R T
# ##################################################################################################################
true_labels = test_dataset.get_labels()
# Set the model to evaluation mode
model.eval()
predicted_labels = []
with torch.no_grad():
for inputs, _ in test_dataset:
inputs = inputs.unsqueeze(0).to(device)
outputs = model(inputs)
# Get predicted labels by getting max value (aka most likely)
_, predicted = torch.max(outputs, 1)
predicted_labels.extend(predicted.cpu().numpy())
# Calculate Confusion Matrix
conf_matrix = confusion_matrix(true_labels, predicted_labels)
print("Confusion Matrix:")
print(conf_matrix)
# plot the confusion matrix
# fig, ax = plt.subplots()
# ConfusionMatrixDisplay(confusion_matrix=conf_matrix).plot(ax=ax, cmap="Blues")
# plt.show()
# plt.savefig('confusion_matrix.png')
create_confusion_matrix(true_labels, predicted_labels)
# Classification report (accuracy, precision, f1 etc)
class_report = classification_report(true_labels, predicted_labels)
print("\nClassification Report:")
print(class_report)
fig.savefig(Path("artifacts") / f"session_{now.month:02}{now.day:02}{now.hour}_{now.minute:02}.png")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--nb_epochs", help="number of training iterations", default=1, type=int)
parser.add_argument("--batch_size", help="batch_size", default=25, type=int)
parser.add_argument(
"--balanced_batches",
help="whether to balance batches for class labels",
default=True,
type=bool,
)
args = parser.parse_args()
main(args) |