def evaluate(segmentation_model, images_dir, masks_dir, metrics, size=None):
segmentation_dl = torch.utils.data.DataLoader(
SegmentationDataset(images_dir, masks_dir, size=size, crop=False), 1, shuffle=False)
eval_out = model_metrics(segmentation_model, segmentation_dl, stats=metrics)
print('Segmenation model', eval_out)
return eval_out
def evaluate_gan_mask_generator(model, G, bg_direction, params,
mask_postprocessing, zs, z_noise, num_steps):
mask_generator = MaskGenerator(
G, bg_direction, params, [], mask_postprocessing,
zs=zs, z_noise=z_noise).cuda().eval()
def it():
while True:
sample = mask_generator()
for img, mask in zip(sample[0], sample[1]):
yield img.unsqueeze(0), mask
score = {
DEFAULT_EVAL_KEY:
model_metrics(SegmentationInference(model), it(), num_steps, (F_max,)),
THR_EVAL_KEY:
model_metrics(Threshold(model), it(), num_steps, (IoU, accuracy)),
}
return score
def predict_cat(X_train, X_test, y_train, y_test=None):
# X_train: description data for training
# y_train: corresponding categories for training
# X_test and y_test: description and category for testing
# Vectorizing the train and test data using TfIDf vectorization
# TfIdf - Text Frequency Inverse Document Freqeuncy : vectorizes based on frequency across the current text document but less frequency across multiple documents
train_tfIdf, test_tfIdf = vectorize(0, X_train, X_test)
train_SVC(train_tfIdf, y_train)
classifier = pickle.load(open("linearkernelSVC.sav", "rb"))
predictions = classifier.predict(
test_tfIdf) #predictions made on the unseen data
train_score = classifier.score(train_tfIdf, y_train)
print("\n\nTrain Accuracy:", train_score * 100, "%\n\n")
score = classifier.score(test_tfIdf, y_test)
model_metrics(classifier, y_test, predictions, score)
