I really enjoyed Ben-David (and Shalev-Schwartz)'s book "Understanding machine learning". It's essentially about theory though, not "learn all about machine learning in torchsorflow in 10 days".
I am currently going through Statistical Rethinking [0], by R. McElreath. So far it's been great, all the code example and practice problems help comprehension.
I've also heard good things about Probabilistic Programming and Bayesian methods [1] by Cam Davidson Pilon.
You could also point out that he has abandoned the kurds in Syria (Rojava), where they are losing ground against the turkish army and their jihadist mercenaries. They are about to ethnically cleanse the region, with dramatic consequences for the autonomous zone that has developed a seemingly fair democracy.
The military branch of the PYD party was crucial during the fight against ISIS.
Leaving them to die in Erdogan's hands is tragic.
I would not consider your view of the problem as "leftist". I'm not challenging your self-designation, but the way you present the situation.
In my opinion, there are a few points that are dubious in this analysis:
- A country is not like a person. It doesn't make any sense to compare the situation of a state to that of an individual.
- Different countries are engaged in trade together. If one, say Germany, has commercial surplus, it is because others buy its production. Hence, all countries can't produce like Germany : who is going to buy everything?
- According to mainstream economic doctrine (Keynes), cutting government spendings does not lead to growth especially amidst a crisis, whereas you have an alternative that does create growth and does not condemn people to starvation. This is done instead by increasing the spendings in the right way.
- Various officials have falsified economic documents in order for Greece to enter the Eurozone (and was helped by Goldman Sachs [0], for example). Why should people suffer from acts of this nature ?
In my company we use pyenv with virtualenv and everything works seamlessly. VSCode automatically runs the activate script from pyenv in a new terminal (in my case it is even the fish shell), debugging works fine, and I can go to definitions.
Note that a new extension, that works alongside the original python extension, has been developed by Microsoft: Pylance. It speeds things up based on pyright, a python type-checker.
For refactoring you can use rope [0] as a VSCode extension.
Hey, thanks for your input. I will look into using pyenv so that might solve my problem with "go to definitions".
However, I was also having a lot of trouble with environment variables, especially PYTHONPATH. How do you solve that?
Also rope did NOT work well for me. When I tried to refactor (rename) a function definition, it also changed variables in unrelated scopes that had the same name and were not related to the function at all. So I did not gain anything from that which I couldn't achieve using CTRL+F replace.
I've never had any trouble with PYTHONPATH, I guess this is solved by pyenv with virtualenv. You just need to select the project interpreter the first time you open the project and it should just work.
I haven't used rope extensively, it's too bad that it doesn't work well. It's open source so I guess it needs more contributions.
It makes me think of Voyager [1][2], based on Vega and Vega-lite, that leverage a grammar of graphics. A neat trick you can do in Julia is use it interactively to generate plot specs [3].
Sadly, looking at the github repo it seems to be unmaintened, so I am glad to see Muze in that space.
Indeed. Vega has been one of our inspirations. Though, we took a more business use-case focussed approach for the visualizations, and also more focus on performance.
I'm a big fan of vega/altair but the biggest limitation is that it can't scale to larger datasets. It might be interesting if you made some side by side comparisons of implementing examples from the vega gallery using Muze. An Altair like wrapper for python users would be nice too, of course.
This gradient descent visualizer is made with it for example [1] to interpolate the points when the prediction line moves.
[0] https://svelte.dev/docs/svelte-motion
[1] https://gradfront.pages.dev/