sizhky

__all__ = ['B','BB','C','choose','crop_from_bb','cv2', 'dumpdill','df2bbs', 'FName','glob','Glob',
 'line','loaddill','logger','extn', 'np', 'now','os','pd','parent','Path','pdb',
 'plt','puttext','randint', 'rand', 'read','rect','see','show','stem','tqdm','Tqdm']

import cv2, glob, numpy as np, pandas as pd, tqdm, os
import matplotlib#; matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.patheffects as path_effects
import pdb, datetime, dill
from pathlib import Path

sizhky

{
    "snippets" : [
        {
            "name" : "example",
            "code" : [
                "# This is an example snippet!",
                "# To create your own, add a new snippet block to the",
                "# snippets.json file in your jupyter nbextensions directory:",
                "# /nbextensions/snippets/snippets.json",
                "import this"

sizhky

git clone https://github.com/sizhky/classical-python-problems
git status
git add folder1 folder2 folder3 file1 file2 file*
git status
git commit -m "1.1.1"
git push origin master

sizhky

# !jbck train.ipynb -> folder.of.train.ipynb/18:[email protected]
ip="train.ipynb"
!echo {ip}
!mkdir -p folder.of.{ip}
!jupyter nbconvert {ip}
!mv {stem(ip)}.html folder.of.{ip}/$(date +%H:%M@%d%b%y)

sizhky

>>>inspect(listofcollections)
'''
0
list `[1,2,3,...]` of len `1345`

1
list `[np.ndarray([[1,2],[3,2],...]) of size (1345x2)
       np.ndarray([list `[2,3,4,...]` of len `7`, list `[2,2,1,...]` of len `15`])
       torch.tensor([[[0.0000,...]]] of size (1345x1x100x100))

sizhky

i = 0
while True:
    try:
        i += 1
        if i == 10: i/0
    except ZeroDivisionError:
        print('Go Free!!')
        break

sizhky

import time
info = lambda report: '\t'.join([f'{k}: {v:.3f}' for k,v in report.items()])

def report_metrics(pos, **report):
    elapsed = time.time()-start
    end = report.pop('end','\n')
    elapsed = '\t({:.2f}s - {:.2f}s remaining)'.format(time.time()-start, ((n_epochs-pos)/pos)*elapsed)
    current_iteration = f'EPOCH: {pos:.3f}\t'
    print(current_iteration + info(report) + elapsed, end=end)
    

sizhky

Inception Networks. Role of Auxiliary loss in Training and Inference

• Deep networks suffer the problem of vanishing gradient, i.e., gradients near to the loss layer are much larger and as they get distributed to the previous layers backward, the magnitude of gradients diminish.

Training

• Inception tackles this problem by using the concept of auxiliary losses where (see fig) it branches out convolution outputs from 4a and 4d into their respective mini classification heads that employ standard softmax - predicting on the same classes as the main task

• Hence during training, there will be a strong gradient flow from loss:0 to adjust layers 4a (and preceding layers) more aggressively. Similar is the case with layers "4d,4c and 4b" where the gradients are high from "loss:1" and for layers 4e onwards which are closer to "loss:2".

sizhky

8x+4y+z = 96
3x+2y+4z = 53
9x+2y+z = 119

sizhky