Sebastian Mantey

[Sebastian Mantey @ManteySebastian]

@itsshehri Currently I am working on something else. So, I haven’t really planned to do this. Sorry!

[Sebastian Mantey @ManteySebastian]

@prajukrishna You can send me a DM on Twitter.

[Sebastian Mantey @ManteySebastian]

@jainpulkit25 I have responded to your comment on the YouTube video: https://t.co/JkU6oBIt7C

[Sebastian Mantey @ManteySebastian]

@babbarsome Thanks! I think you need to include the line “app.config.suppress_callback_exceptions = True”. You can read about it here: https://t.co/XoRDLWhqDV

[Sebastian Mantey @ManteySebastian]

@kierisi I recently listened to the @tferriss podcast with @SafiBahcall. And he has an interesting concept called FBR (at 1:06:42 in the video): https://t.co/zlOfOVToK3

[Ian Goodfellow @goodfellow_ian]

These style-based generator results look great: https://t.co/RL825n0yNP

[Sebastian Mantey @ManteySebastian]

@jeremyphoward Are there any particularly resources you can recommend for learning how to use it properly and take full advantage of it?

[Sebastian Mantey @ManteySebastian]

@laurafbec Sure!

[Sebastian Mantey @ManteySebastian]

@ValechaJagrati Maybe, it’s also a question of greedy algorithm vs. dynamic programming. But I don’t really have any expertise about that.

[Sebastian Mantey @ManteySebastian]

@ValechaJagrati Normally, it shouldn’t improve the accuracy if you consider all the other potential splits. After all, we are choosing THE best split. The sklearn library probably uses many more potential parameters than just the max depth or min samples. So, you might look into them.

[Sebastian Mantey @ManteySebastian]

@jakevdp I just started to play around with plotly and dash to create my own personal dashboard. And there I also have to tinker a little bit with CSS. So, it is somewhat relieving to hear that even non-beginners/experts don’t think it is easy or straight-forward.

[Sebastian Mantey @ManteySebastian]

@amir_ali_cheema @sirajraval Just in case you haven't seen it, but @iamtrask tweeted many good ideas for potential #100DaysOfMLCode projects.

[Sebastian Mantey @ManteySebastian]

@ValechaJagrati I had a look at it now. And I’m not really sure why you would want to implement all the combinations and not just the best one.

[Sebastian Mantey @ManteySebastian]

@ValechaJagrati I will have a look at it on the weekend.

[Sebastian Mantey @ManteySebastian]

@hugobowne I really like your podcast. It’s so interesting to learn about the experiences of people working in data science. It’s also really great that your interview partners come from so many different industries and backgrounds.

Can I ask you how you record the interviews?

[Sebastian Mantey @ManteySebastian]

@SaketPallav @sirajraval He touched on that in this video: https://t.co/YTrRLUaJhw

[Sebastian Mantey @ManteySebastian]

@iamtrask … But, I think, that is not really the problem (think ReLUs). The bigger problem is that the derivative of the step function is zero everywhere else. So, because of the chain rule in backprop, the weight updates would always be zero.

[Sebastian Mantey @ManteySebastian]

@iamtrask … in introductions to deep learning, the step function is often used first to explain the purpose of activation functions. Then, it is replaced with the sigmoid function since the step function is not differentiable at the step…