although I have seen a lot of explanations for decorators, yours has to be one of the easiest to follow
Ow thatās very kind of you, thank you! Although I should say that most of why itās so easy to follow is thanks to talk itās based on, by James Powell.
Diagnosing Neurodegenerative Diseases with Pytorch
Nice! Weāll be covering decorators next week, so I hope youāll share this excellent post in the lesson wiki thenā¦
Iām a remote participant, but will be in San Francisco next week due to attending a conference. Does anyone know if it is possible for normally remote people to attend a live session?
If so, whatās the specific location?
what are you planning to do with them?
Sure thing. Thanks!
The Possibilities are endless, but for right now, Iām going to use them for them to be servers running the model Iām working on as part of Muziguideās web scraping servers, Also, I just might resurrect this circa 2005 robot I built (which has a camera and sonar ranging), and make it do real-time terrain recognition.
Hey @PierreO !
Thanks for the great posts-both this and the papers one(Iām still going through it)
A small suggestion: the last example(decorating add with ntimes) could be replaced with something better, since the output is same as that of one execution.
Even something silly like
@ntimes(3)
def print_blah(x=āblahā):
print(x)
Also, last line of code has a typo: Y instead of y.
Itās not, unfortunately. But please do drop by our study room!
Thanks for the clarification. Iāll see if I can get away during the day and drop by.
Youāre right, Iāll add a print statement
Many thanks for the typo!
thanks , there are some really inspiring stories . ātransferring it over to my domainā advice resonates more with me.
There is cool pycharm plugin that uses AI for predictive autocomplete.
Codota does the same for Java.
I believe there is a huge opportunity in terms of applying artificial intelligence to software engineering. For instance, there Fast.ai at GitHub. Worth watching:
Good stuff. Also, a good practice is to use functools.wraps
to fix the name mangling. Explained here. https://realpython.com/primer-on-python-decorators/#who-are-you-really
Inspired by lesson 2 of part 1 , i have created image dataset downloader.
If you search for keywords first time , it may take some time ( I cache the result , so that next time it will be faster).
I have already cached following keywords ( teddy bear, grizzly bear, black bear, polar bear)
(dog,cat).
Please try it out and let me know.
Dr. Sebastian Ruder has finally uploaded his thesis to his website.
Iāll definitely read this as a homework (Over a few days)
Any luck getting this set up?
L8@1:29:22 ā on initializing layers so that (mean,stdv) ā (0,1) ā huh, interesting. As if normalization layers (regularization too?) were achieving the same end by correcting values after the fact. Sounds like at least basic care for initializing layers should be a standard thing you always do.
Ah. If different layers have (wildly) different (mean, stdev)'s, theyāre all effectively on different playing fields ā good luck getting them to understand one another.
329 pages! Far from the ULMfit 12 pages one Looking forward to reading it.