Python machine learning

[Jet_ski]

Which python based machine learning platforms do you recommend? I’ve read a little bit about the difference between TensorFlow and PyTorch but I know how to interpret the stability and graph discussions. I’m interested in advice based on your own experience.

 

thanks!

[Nano Adam]

My way to pick something would be to build a few projects in both frameworks then see which one you caught on with. 

[tspdm]

It depends on what you want to do. Is compatability with Mobiles etc important to you ? How deep into ML do you want to dive ? 
From my Experience Tensorflow and espacially the KERAS interface is REALLY easy to get started in.

[steelo]

14 minutes ago, Jet_ski said:

Which python based machine learning platforms do you recommend? I’ve read a little bit about the difference between TensorFlow and PyTorch but I know how to interpret the stability and graph discussions. I’m interested in advice based on your own experience.

 

thanks!

A good 'beginner' project IMO is creating a tracking camera. There are a lot of youtube projects where people made things like facial recognition and tracking turrets. I'd make sure you have all of the basics down with python though.

[Jet_ski]

7 minutes ago, tspdm said:

It depends on what you want to do. Is compatability with Mobiles etc important to you ? How deep into ML do you want to dive ? 
From my Experience Tensorflow and espacially the KERAS interface is REALLY easy to get started in.

Ideally deep enough to put it on my resume and be able to try my own algorithms.

[steelo]

6 minutes ago, Jet_ski said:

Ideally deep enough to put it on my resume and be able to try my own algorithms.

I'm new to python and have similar interests but am very weak with math. I've read that you can get a lot of the 'mathy' stuff online but you still need to understand what the algorithm is doing. That is probably my biggest hurdle.

[igormp]

26 minutes ago, Jet_ski said:

Which python based machine learning platforms do you recommend? I’ve read a little bit about the difference between TensorFlow and PyTorch but I know how to interpret the stability and graph discussions. I’m interested in advice based on your own experience.

 

thanks!

PyTorch is easier and has a nicer API, it's also being more used than tensorflow from what I've seen. Can't go further than that since I only used it superficially.

 

Tf on the other hand is kinda of a mess to look for resources, since the API changes way faster than you'd like to, and some tutorials end up outdated rather quickly. Keras has made tf 2.0 nicer to use, being more like PyTorch, but still not as easy. The great advantage of tensorflow comes when you want to deploy it to embedded devices, FPGAs or even TPUs for faster inference (or even training!). I'm also more used to it already, so there's that.

 

14 minutes ago, steelo said:

A good 'beginner' project IMO is creating a tracking camera. There are a lot of youtube projects where people made things like facial recognition and tracking turrets. I'd make sure you have all of the basics down with python though.

Usually the "hello world" for ML is playing with MNIST or the Titanic dataset.

[Jet_ski]

2 hours ago, steelo said:

I'm new to python and have similar interests but am very weak with math. I've read that you can get a lot of the 'mathy' stuff online but you still need to understand what the algorithm is doing. That is probably my biggest hurdle.

Don’t let math discourage you from programming. Programming algorithms aren’t the same as math algorithms. You don’t need to be able to mathematically solve the computational efficiency of an algorithm to learn how it works and use it effectively.

 

• An example of a math algorithm is factorizing a matrix A into an upper and lower triangular matrix: A=LU
• An example of a programming algorithm is Insertion Sort which takes a list and sorts it’s elements by moving one element at a time.

Try this python book is free and easy to read

https://greenteapress.com/wp/think-python/

[steelo]

20 minutes ago, Jet_ski said:

Don’t let math discourage you from programming. Programming algorithms aren’t the same as math algorithms. You don’t need to be able to mathematically solve the computational efficiency of an algorithm to learn how it works and use it effectively.

 

• An example of a math algorithm is factorizing a matrix A into an upper and lower triangular matrix: A=LU
• An example of a programming algorithm is Insertion Sort which takes a list and sorts it’s elements by moving one element at a time.

Try this python book is free and easy to read

https://greenteapress.com/wp/think-python/

Thank you, I'll check that book out. Right now I'm reading 'Automating the boring stuff with Python'...it's a really good read!

[EffeectFTW]

The answer depends on what you want to make, the one truth about Machine Learning is that you will spend 90% of your time adapting the data input and output rather than working on the model and training.

 

If you want a good starting point to Machine Learning, I would recommend looking at some more high level solutions and then move onto TensorFlow and PyTorch. Two projects I really love is https://runwayml.com/ and https://ml5js.org/ which allow you to test out different models and solutions relatively quickly, while these solutions are limited, I think they will give you a much better understanding of how Machine Learning actually works.