Skip to content

Instantly share code, notes, and snippets.

View ivanpanshin's full-sized avatar

Ivan Panshin ivanpanshin

54 followers · 9 following
View GitHub Profile
@ivanpanshin
ivanpanshin / transforms.py
Created April 12, 2020 12:43
train and valid transforms for cancer detection
train_transforms = transforms.Compose([
transforms.Resize((196, 196)),
transforms.RandomChoice([
transforms.ColorJitter(brightness=0.5),
transforms.ColorJitter(contrast=0.5),
transforms.ColorJitter(saturation=0.5),
transforms.ColorJitter(hue=0.5),
transforms.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),
transforms.ColorJitter(brightness=0.3, contrast=0.3, saturation=0.3, hue=0.3),
transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5),
@ivanpanshin
ivanpanshin / train_transform.py
Created April 12, 2020 13:30
train and valid transforms for cancer classification
train_transforms = transforms.Compose([
transforms.Resize((196, 196)),
transforms.RandomChoice([
transforms.ColorJitter(brightness=0.5),
transforms.ColorJitter(contrast=0.5),
transforms.ColorJitter(saturation=0.5),
transforms.ColorJitter(hue=0.5),
transforms.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),
transforms.ColorJitter(brightness=0.3, contrast=0.3, saturation=0.3, hue=0.3),
transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5),
@ivanpanshin
ivanpanshin / validation.py
Last active April 12, 2020 13:32
group k-fold validation for cancer classification
train_labels = pd.read_csv(os.path.join(data_dir, 'train_labels.csv'))
wsi = pd.read_csv(os.path.join(data_dir, 'patch_id_wsi_full.csv'))
wsi_ids = wsi.wsi.values
group_kfold = GroupKFold(n_splits=n_splits)
for fold_index, (train_index, valid_index) in enumerate(group_kfold.split(train_labels.id, train_labels.label, wsi_ids)):
X_train, X_valid = train_labels.id[train_index], train_labels.id[valid_index]
y_train, y_valid = train_labels.label[train_index], train_labels.label[valid_index]
@ivanpanshin
ivanpanshin / eda.py
Last active April 12, 2020 17:54
eda for cancer detection
import os
from PIL import Image
import matplotlib.pyplot as plt
# Basic stats like amount of images, classes balance, image size
n_train_images = len(os.listdir('data/train'))
n_test_images = len(os.listdir('data/test'))
train_imgs = os.listdir("data/train")
print('Train/Test images: {}/{}'.format(n_train_images, n_test_images))
@ivanpanshin
ivanpanshin / models.csv
Last active April 12, 2020 19:01
models comparison
Name of the model CV (Fold) LB (Private Kaggle)
SE_ResNet-50 center crop (32 px) 0.91 0.89
SE_ResNet-50 original size (96 px) 0.96 0.95
SE_ResNet-50 resize (196 px) 0.980 0.966
SE_ResNet-50 resize + TTA (196 px + 4TTA) 0.985 0.969
SE_ResNet-50 resize + TTA + folds (196 px + 4TTA + 5 folds) — 0.977
previous model + EfficientNet-B3 resize + TTA + folds (224 px + 4TTA + 5 folds) — 0.981
@ivanpanshin
ivanpanshin / app.py
Created April 28, 2020 08:31
from flask import Flask
server = Flask(__name__)
@server.route('/')
def hello_world():
return 'hello world!'
if __name__ == '__main__':
server.run(host='0.0.0.0', port=8000)
@ivanpanshin
ivanpanshin / app.py
Last active April 29, 2020 20:20
Basic Flask App
from flask import Flask
server = Flask(__name__)
@server.route('/')
def hello_world():
return 'hello world!'
@ivanpanshin
ivanpanshin / wsgi.py
Last active April 29, 2020 20:22
WSGI entry point
from app import server
if __name__ == "__main__":
server.run(host='0.0.0.0', port=8000)
@ivanpanshin
ivanpanshin / Dockerfile
Created April 28, 2020 08:49
Dockerfile for Flask and Gunicorn
FROM python:3.6.7
WORKDIR usr/src/flask_app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
@ivanpanshin
ivanpanshin / Dockerfile
Created April 28, 2020 08:50
Dockerfile for Flask and Gunicorn
FROM python:3.6.7
WORKDIR usr/src/flask_app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .