Just wanted to say @Alexey that you're a good interviewer, you asked some great questions! I will check out more of your videos
@KamaAmmo
2 жыл бұрын
Valery Babushkij is a really cool guy!
@zubrz
Жыл бұрын
i had exactly the same experience with ML system design at facebook in London in 2016
@divyamshah9
2 жыл бұрын
Can someone share the link to the 16 point document on ML system design talked about in the video?
@zenogantner
2 жыл бұрын
ML Design interview content starts at 14:42: kzitem.info/news/bejne/kYip04agmqRnaXo
@johnsontian2733
2 жыл бұрын
Just to add that precision is a metric that’s not sensitive to class imbalance.
@ValeriiBabushkin
2 жыл бұрын
The problem with precision is that its calculations take both classes into account, precision = TP/(TP + FP) Imagine we have a model which has a probability of 95% to predict that fraud is a fraud and 5% to predict that non-fraud is fraud. Review two scenarios P = number of positive samples, N = number of negative samples P = 10000, N = 10000, Precision = 0.95*10000/(0.95*10000 + 0.05* 10000) = 0.95 P = 1000, N = 10000, Precision = 0.95*1000/(0.95*1000 + 0.05* 10000) = 0.65
@johnsontian2733
2 жыл бұрын
@@ValeriiBabushkin hey, thanks for the reply! I agree precision is still affected by class imbalance, but for the purpose of model selection, I think it’s effective in separating good models that has high probability of correctly detecting positive class in an imbalanced setting from bad models. E.g., if the model in your 2nd example has 99% prob to predict fraud as fraud, the precision will be 90.8% precision, drastically better than the classifier with 95% prob to predict fraud as fraud.
@hihi-wc7dx
Жыл бұрын
@@johnsontian2733 what was his reply
@momen_ai
2 жыл бұрын
Thank you
@CantPickTheNameIwant
Жыл бұрын
So funny to watch when two guys have a conversation and try to mimic language accents: one american english and second one british, knowing the fact that both are actually russians
@user-uv8ef8sn6k
2 жыл бұрын
Nice British accent, Valeriy! :)
@elmirach4706
8 ай бұрын
I wish you were putting that on board or went through one ML System DEsign case, rather than just talk. The whole title is ML system design, but there is no single case walked through.
@beladianeeral
Жыл бұрын
@16:53 - I think you mean starting with logistic regression?
@Gerald-iz7mv
11 ай бұрын
hi how to prepare for system design platform/infra interview?
@hihi-wc7dx
Жыл бұрын
He meant to say logistic regression didn’t he
@shwetashaw9978
Жыл бұрын
I guess so, as they were talking about a binary classification problem
Пікірлер: 20