LS
It's ovefits but on test data it's bad again
Size: a a a
2019 July 22
2019 July 23
1u
А с чего это перешли на английский? Не умеете общаться на русском - ищите своих Саксонов
DP
А с чего это перешли на английский? Не умеете общаться на русском - ищите своих Саксонов
английский знает около 2 млрд людей
русский знает около 0.4 млрд людей
ну и все кто хочет изучать ИИ скорее всего должен изучать английский тоже, на англ все работы по ИИ
русский знает около 0.4 млрд людей
ну и все кто хочет изучать ИИ скорее всего должен изучать английский тоже, на англ все работы по ИИ
SA
How to Develop an Information Maximizing GAN (InfoGAN) in Keras
https://machinelearningmastery.com/how-to-develop-an-information-maximizing-generative-adversarial-network-infogan-in-keras/
https://machinelearningmastery.com/how-to-develop-an-information-maximizing-generative-adversarial-network-infogan-in-keras/
BK
I have very small dataset of 1800 rows.
Enough for a small network with proper regularisation
BK
Also try UMAP for dimensionality reduction
LS
Umap works for mixed data?
LS
I tried famd already. Was not suitable
BK
Umap works for mixed data?
Onehot first
LS
Okay will try umap
LS
Thank you
DP
Enough for a small network with proper regularisation
he wants something explainable. neural network's structure cant be explained in a presentation
BK
he wants something explainable. neural network's structure cant be explained in a presentation
Yes, they can be
LS
What nn would you suggest for a 1000 rows?
DP
Yes, they can be
i mean all of the connectome, not just layers
BK
i mean all of the connectome, not just layers
NN is not a black box when you try hard enough
DP
NN is not a black box when you try hard enough
it is if you just have a month
LS
I have only 12 days lol
BK
What nn would you suggest for a 1000 rows?
Try two or three layered fully connected relu activated with sigmoid output
DP
What nn would you suggest for a 1000 rows?
i would suggest fully connected layers, very few layers (2-3), removal of weak weights (0.01=0)