# ########## SET UP MODEL ########## #
model = FullyConnectedClassifier(**network_args).to(DEVICE)
model.load_state_dict(
torch.load(
os.path.join(MODEL_LOAD_PATH, "fold_" + str(fold) + ".weights")))
optimizer = optim.Adam(model.parameters(),
lr=LEARNING_RATE,
weight_decay=WEIGHT_DECAY)
loss_function = nn.BCELoss()
# ########## EVALUATE ########## #
test_args = {
"model": model,
"device": DEVICE,
"test_loader": test_loader,
"loss_function": loss_function,
"debug_mode": False,
"fold_num": fold,
"predictions_save_path":
CROSS_VALIDATION_WHEN_HAND_CRAFTED_FEATURE_PATH,
"variable_batch_size": True,
}
loss, ap = test_model(**test_args)
print("Test loss: {0:.5f} AP: {1:.4f}".format(loss, ap))
all_aps.append(ap)
print("MEAN AP:", np.round(np.mean(all_aps), 3) * 100)
"loss_function": loss_function,
"debug_mode": DEBUG_MODE,
}
loss, ap = train_model(**trainer_args)
print("Train loss: {0:.5f} AP: {1:.4f}".format(loss, ap))
val_args = {
"model": model,
"device": DEVICE,
"test_loader": val_loader,
"loss_function": loss_function,
"debug_mode": DEBUG_MODE,
}
loss, ap = test_model(**val_args)
print("Validation loss: {0:.5f} AP: {1:.4f}".format(loss, ap))
if ap > best_ap:
best_ap = ap
best_model = copy.deepcopy(model)
model_save_name = "fold_" + str(fold) + ".weights"
torch.save(best_model.state_dict(),
os.path.join(MODEL_SAVE_PATH, model_save_name))
test_args = {
"model": best_model,
"device": DEVICE,
"test_loader": test_loader,
"loss_function": loss_function,
"debug_mode": False,