KornelDylski

# winpty fix
alias python='winpty python.exe'
alias pip='winpty pip'

# custom aliases
alias rt='reset'
alias ll='ls -la'
alias l='ls -1'
alias la='ls -a'

KornelDylski

async function getImages(numbers) {
  // fetch all at once, return fetch promises
  const allResult = await Promise.all(
    numbers.map((num) => fetch("url/to/img" + num))
  );

  // iterate and return array of json promises
  return Promise.all(
    allResult.map((result) => {
      if (result.ok) {

KornelDylski

def accuracy(input, target, thresh=0.4, na_idx=0):
  valm, argm = input.max(-1)

  # results below threshold are considered as "na" category
  argm[valm < thresh] = na_idx

  return (argm==target).float().mean()

KornelDylski

# apply sigmoid and one-hot
input = input.sigmoid()
target = F.one_hot(target, input.shape[1]).float()

# change all target category “na” (placed at 0 index) to zero
target[:, 0] = 0

# finally count bce loss
loss = F.binary_cross_entropy(input, target)

KornelDylski

# borrowed from https://github.com/clcarwin/focal_loss_pytorch/blob/master/focalloss.py
# added "reduction" param for fastai
import torch
import torch.nn as nn
import torch.nn.functional as F


class FocalLoss(nn.Module):
  def __init__(self, gamma=0.0, alpha=None, reduction='mean'):
    super(FocalLoss, self).__init__()

KornelDylski

import numpy as np
import pandas as pd
import cv2
import os
import uuid
import argparse
from pathlib import Path

from google_images_download.google_images_download import googleimagesdownload