Python

I have been using Python for a very long time.

Whenever you start a new Python project, always start with a Virtual Environment.

I really need to develop the habit of using virtual environments. Here is the command:

pip freeze | xargs pip uninstall -y

Links

• The Hitchhiker’s Guide to Python is a good book to learn best practices.
• https://realpython.com/ is also really good, has some great resources

Python under the hood

• https://medium.com/swlh/python-memory-management-7181d8a6ae5e
• https://medium.com/@naishasaxena2310/how-python-works-behind-the-scenes-57823eb757bc

Code Style Python Script

Why python is so good

• syntax (no need to declare types upfront)
• Automatic memory management,
• libraries (scipy, numpy, pytorch)
• community support is amazing

Structure Your Project

Check out how this folder is structured.

• Module
• Package
• Object-Oriented Programming
• Decorator
• Dynamic Typing
• Mutable and Immutable Type

Testing Your Code

• Test-Driven Development
• Testing

Documentation

• Documentation

Virtual Environments

• MambaForge

Random Tips

Use .extend() instead of .append() if you want to “append” an array to another array.

Python Design

After taking CS247, I have SUCH a stronger mastery of Object-Oriented Programming. Now, I am actually curious about how does OOP actually work in Python? How do types work in Python?

See https://realpython.com/python-type-checking/

Ahh, so python is ALWAYS dynamically typed. https://stackoverflow.com/questions/4714136/how-to-implement-virtual-methods-in-python. And this is why you don’t even need to declare any function as virtual, since it will NEVER use the static type.

Some new things

In Python, a private member can be defined by using a prefix __ (double underscore). So, in the private modifier’s case, we cannot access the attribute, Learned this at Ericsson

phaseOffsetTableObj.__offsetList  # this will return has no attribute

From Andrej Karpathy, so much I am learning from him:

# For classes, you can overwrite + and -, internally in python:
a + b  # a.__add__(b) is called under the hood
 
# Another example
5 + 6 # (5).__add__(6) 
 
# However, this returns an ERROR
int(5).__add__(float(6)) # NotImplemented Error, since int doesn't know how to add onto floats,
float(5).__add__(int(6)) # But this works!!
 
# Also, this is so interesting, I never thought about this
a + 2 # a.__add__(2) WORKS
2 + a # 2.__add__(a) DOESN'T WORK, unless you implement __radd__(self, other)
 
# so below works because #__radd__ is implemented when for the float values
5 + 11.0 

the __call__ method

class Neuron:
	def __call__(self, x):
		...
 
n = Neuron(2)
n(x) # calls __call__

the zip function is actually super useful. Just in lecture 2 I am learning about using this function to print out the bigrams: With the zip, if any of the two inputs is shorter, it just stops, which is why this works.

for w in words:
	for ch1, ch2 in zip(w, w[1:]):
		print(ch1, ch2)

list(string) behaves differently than I thought. Say w = "emma", then list(w) == ['e', 'm', 'm', 'a']

Thoughts

Python is a really good scripting language, but the more I use Python for OOP, the more I dislike it. There is not strict typing, they do introduce the type thingy like this since like Python 3.8 (I believe?), but that doesn’t provide any optimization.

a: int = 5

Python is still fundamentally dynamically typed for everything, which makes it pretty slow and annoying and easy to make mistakes.

With C++, I already know all the types ahead of time, makes it easier to scale. George Hotz talked about this on a Lex Fridman podcast.

Scope Resolution

if __name__ == '__main__':
    x = 1
    
print x

python vs. sudo python

Ran into this super weird issue, where I need sudo permissions, but don’t want packages to install at the sudo level.

Intended target:

/usr/local/bin/python3.8/dist-packages

Target if you use sudo

/root/.local/bin/python3.8/dist-packages

https://superuser.com/questions/600349/why-sudo-python-and-python-in-terminal-start-two-different-versions-python