Lexie88rus

# Extract words from kernel titles
end_of_sentence = '.' # symbol to denote the end of the sentence
def extract_words(title):
    '''
    Function which transforms kernel title into a list of words ending with 'end_of_sentence' word.
    '''
    title = clean_title(title)
    words = title.split(' ')

    return words

Lexie88rus

import re

# Lowercase, remove punctuation and numbers from kernel titles
def clean_title(title):
    '''
    Function to lowercase, remove punctuation and numbers from kernel titles
    '''
    # lowercase
    title = str(title).lower()
    # replace punctuation into spaces

Lexie88rus

# Merge kernels and versions to retreive kernel title and total votes for kernel
kernels_trc = kernels[['CurrentKernelVersionId', 'TotalVotes']].rename(columns={'CurrentKernelVersionId' : 'Id'})
kernel_version_trc = kernel_versions[['Id', 'Title']]
kernels_titles_votes = kernels_trc.merge(kernel_version_trc)

# Sort titles by the number of votes
kernels_titles_votes = kernels_titles_votes.sort_values(by=['TotalVotes'])

# Retreive the list of popular kernel titles (at leat 1 vote)
popular_kernel_titles = kernels_titles_votes[kernels_titles_votes['TotalVotes'] > 0]['Title'].unique().tolist()

Lexie88rus

import pandas as pd

# Load data
# Load kernel versions
kernel_versions = pd.read_csv('../input/KernelVersions.csv')
# Load kernels (to retreive TotalVotes)
kernels = pd.read_csv('../input/Kernels.csv')

Lexie88rus

# Define the demo dataset
class DogDataset3(Dataset):
    '''
    Sample dataset for Augmentor demonstration.
    The dataset will consist of just one sample image.
    '''

    def __init__(self, image):
        self.image = image

Lexie88rus

# Import package
import Augmentor

# Initialize pipeline
p = Augmentor.DataPipeline([[np.array(image), np.array(mask)]])

# Apply augmentations
p.rotate(1, max_left_rotation=3, max_right_rotation=3)
p.shear(1, max_shear_left = 3, max_shear_right = 3)
p.zoom_random(1, percentage_area=0.9)

Lexie88rus

# Import pytorch utilities from albumentations
from albumentations.pytorch import ToTensor

# Define the augmentation pipeline
augmentation_pipeline = A.Compose(
    [
        A.HorizontalFlip(p = 0.5), # apply horizontal flip to 50% of images
        A.OneOf(
            [
                # apply one of transforms to 50% of images

Lexie88rus

# Compose a complex augmentation pipeline
augmentation_pipeline = A.Compose(
    [
        A.HorizontalFlip(p = 0.5), # apply horizontal flip to 50% of images
        A.OneOf(
            [
                # apply one of transforms to 50% of images
                A.RandomContrast(), # apply random contrast
                A.RandomGamma(), # apply random gamma
                A.RandomBrightness(), # apply random brightness

Lexie88rus

# Import patches from matplotlib to draw rectangle for bounding box
import matplotlib.patches as patches

# Visualize augmented image and bbox
fig, ax = plt.subplots(1,2, figsize = (15, 10))

# Plot the original image and bounding box
ax[0].axis('off')
ax[0].imshow(image)
rect = patches.Rectangle((bboxes[0],bboxes[1]), bboxes[2], bboxes[3],linewidth=1,edgecolor='r',facecolor='none')

Lexie88rus

# Create bounding boxes from mask with cv2
import cv2
mask = cv2.cvtColor(mask, cv2.COLOR_BGR2GRAY)
bboxes = cv2.boundingRect(cv2.findNonZero(mask))

# Augment image and bounding box
augmented_boxes = aug_pipeline(image = image, bboxes = [bboxes])

# Access augmented image and bounding box
image_aug = augmented_boxes['image'][0]