Comments by "Dr Gamma D" (@DrDeuteron) on "Indently" channel.

  1. Since you didn't ask, here are my tips for functions: docstring for users comment for devs pass arguments as keywords (for devs--from Raymond Hettinger--If you don't know who that is, you're not a pystro) single point return for successful use-cases use guard clause to catch errors early and raise/return early minimize branches (cyclic complexity). only ONE purpose per function: break it down, use private helpers if needed) so devs never scroll to see the whole function. all functions should be PURE: if it depends on state or mutates parameters, make it a class method....and don't mutate someone else's state. only 'type(self) should mutate 'self's attributes.
    39
  2. well now the next video needs to be on custom exceptions.
    37
  3. itertools is written by pystro: Raymond Hettinger, and it is optimized af. Moreover: it reduces your code's cyclomatic complexity. Many for loops, with indented if-blocks disappear. IT doesn't matter that they occur in functions, they are NOT in your code, and your code is less complex by definition of complexity.
    31
  4. list(itertools.compress(people, map((7.).__lt__, map(len, people))) or functools.partial(operator.lt_, 7). which I like because it tells people to stay tf out of your code.
    29
  5. this is awesome, I finally got bumped up to 3.12 (security restrictions)..TODAY, and I am behind on f-strings. I'm concerned they're not pythonic tho, we'll see. I still haven't learned format strings, which I first saw in 1987.
    26
  6. Yes, David Hilbert said if he were awoken in 100 years, his first question would be has the Riemann Hypothesis been proved what version of python is in beta?
    23
  7. Imma tell you the best use for classmethod. One rule they don't tell you is: don't do work in init. It really makes your code more maintainable and extendable. So for instance (ha ha), say you have a class that is the data from a CCSDS packet in a data file. Your procedural instinct is to have dunder init takes the file name and does the reading and unpacking there, and deriving the instance attributes. NO! DO NOT DO IT!!!. Now with dataclasses, it's even easier---you define the CCSDS packet via the data class dunder init default, and the make a class method that is the all important "constructor helper": @classmethod def fromfilename(cls, fname, path=''): with open(os.path.join(path, fname)) as fsrc: return cls.fromefile(fsc) @classmethod def fromfile(cls, fsrc): return cls(*fsrc.readlines())... # or some function thereof. Trust me, it is sooooo much better.
    21
  8. I’m actually dating them rn. Like right. Now.
    20
  9. if I get a script w/o it, I freak out.
    14
  10. just for fun, circa 7:00, you said "sum up until n", which would be interpreted by a native speaker as "sum up to n", your code sums to n-1, which the fix would be to initialize result = n...though that is kinda ugly for those who like ordered math sums, but so is range(n+1) for those who like nice code.
    10
  11. one common pythonic thing is, repr(instance) should return a string so that: >>>instance == eval(repr(instance)) is True, so something _like_: return f"{type(self).___name__}(}" + ", ".join(f'{k}={v}' for k, v in self.___dict___.items()) + ")" Also: note that these dunder methods are strongly typed and MUST return a str. You can also use the "dunder module" static class attribute to assist.
    10
  12. you should never use the class name in an instance method, use "type(self)". If you need to know the exact class then it's time to, ugg, use name mangling. and top speed of "200" is a magic number, it should be a private class attribute: _top_speed = 200 in the class namespace.
    10
  13. that's a different default. except at 6:15 where it is probably preferable.
    9
  14. 5:25 another thing about try/except blocks is minimizing the code in the try block, so the 1st print statement should be before: try: user_float = float(user_input) except ValueError as err: ... except TypeError as err: ... else: print ("etc...") and note, the print statement is removed also, only the cast-to-float lives in try because only it is in question
    9
  15. "integer literals" is their name.
    9
  16. slice is fantastic. I have a complicated space instrument with around 500 channels, and it comes in so fast the electronics can't keep up, so there are 4 separate amplifier chains...which are all identical. but of course they aren't, so each needs to be calibrated separately, The guys were doing it in Matlab with handwritten if chain=='north': idx = [1, 5, ...] else./...repeated everywhere. So I enlightened them (oh, and it had a time stamp in position 0), so I just made a dictionary of slice object vs amplifier chain (named after the cardinal directions): AMP = dict(time=slice(0,1), north=slice(1, None, 4), east=slice(2, None, 4), south=(3, None, 4), west=(4, None, 4)) so a  calibration = {amp: sensor.data[AMP[amp]].mean() for amp in ('north', 'east', 'south', 'west')} gave me a dictionary of time stamped calibration sample averages (when the system was in calibration mode). Then you save that and scale environmental data until the next cal measurment 2 seconds later. `No ifs, no loops. cyclomatic complexity minimized. Nothing is WETT (write everything twice), it's all DRY (don't repeat yourself).
    9
  17. No 15: overloading repr() is a good idea. You can also use "type(self)._name__" or self.__class___.__name_ instead of the explicit "Person" in the f-string.
    9
  18. yield is also an assignment statement: value = yield value coroutines.
    8
  19. | is neither union nor addition, it’s bitwise or.
    8
  20.  @RedShipsofSpainAgain  doesn't the context lib context manager decorator just wrap your function in try/except/finally?
    8
  21. Now this is a separate video on coroutines. yield is not just a 'return' statement. It is also an assignment statement. If in a generator called 'foo', you have: new_value = yield old_value a user, baz, of that generator that has declared: def baz.. bar = foo() can then do: current_result = bar.send(my_value) which gets bars's old_value and stores it locally in baz's current_result while at the same time storing baz's my_value in bar's new_value. bar then waits there until anyone calls next(bar). at which point foo's code restarts execution. confused? good, so am I. But I have used these to write a text "continued line" unwrapper, and a clever spacecraft data packet reader in which the length of the packet is not known until the end of sensor scan--with NO, ZERO, NADA if-then clauses. Cyclic Complexity = 1, but it handles all cases WITHOUT indexing past/present data streams, even though the data reading depends on the values of the data stream at different timestamps (read: indices).
    7
  22. I learned something today.
    6
  23. your matmul should return: type(self)(zip(*[[sum(map(operator.mul, col, row)) for col in self] for row in zip(*other)]))
    6
  24. >>>d1 = dict(a=1, b=2) >>>d2 = dict(c=3, d=4) is much clearer python. It directly types a dictionary with string keys assigned to int literals. Are the type hints just for your IDE? imo human readability comes before your IDE. But why would you type a dict anyway? If you want you special dict[str]=int, just subclass the builtin and assert it.
    6
  25. he mean "Human readability"
    6
  26.  @CripsyFries  or, var_name = list(map(os.remove, glob.glob("*.*"), recursive=True))...or something like that.
    6
  27. 4:45 pro-tip: never ever drop the leading zero on a decimal number that is less than one, and not just in coding. I mean in life.
    6
  28. A comment on the comment section, unrelated to comments: it's pythonic to practice EAFP, while you implement LBYL! What that means is that you would unpack the split string straight into a dict, and don't even have "parts"--which is whole reason you need comment--and that would be in a try/except block for when you didn't look before you leapt.
    6
  29. int enums are good as bit wise flags, but after watching this video, I am going to extend normal integer (Z) operations to raise a not implemented error.
    6
  30. I agree. I remember reading there is no n++ operator b/c Guido never mixes expressions and assignments, but I guess that is passé. Moreover, which order is the dict construction evaluated? I guess left to right, but that is non-pythonic code imho. Also: what if you assign to word more than once in the dict constructor? Sounds buggy af.
    5
  31. I'd go ABC. This seems like it's a way to interact with your IDE.
    5
  32. when I write code that is a tool, let's say something about LEO assets... when python goes wrong, I throw a builtin exception, (and if it;s I/O, I import errors and figure out which I/O thing went wrong), but if it's a domain specific user mistake, I send a custom exception. for example, they set orbit altitude to 2000 km, I'll throw a: >>>class LEOAltitude(ValueError): telling them LEO is 200 - 1600 km. If it's more general, I may go nuts and have an mro that looks like: LEOAltitudeError [or MEO, GEO] AltitudeError. [or Inclination, RightAscension, Eccentricity, MeanAnomaly, ...] OrbitParameterError [or Epoch or Classification... any thing in aTwo Line Element] ValueError. [now it's python's mro from here on out] StandardError Exception BaseException object
    5
  33. forget about speed, just reducing cyclomatic complexity is a win.
    5
  34.  @brunopanizzi  I know, the list comp is lame. I'd rather map to an itertools thing for ZERO loops, but I was in a hurry.
    4
  35. but no. A book is a sequence. iter(book) is the iterator. I mean when you read a page, is it gone forever?
    4
  36. which is why you should never write a function that returns a complicated interface, unless it has a leading underbar in its name, and even then....
    4
  37. View any video by Raymond Hettinger (the creator of itertools), he's a great teacher, esp "super considered super" if you're doing multiple inheritance. Also the somewhat dated "Loop like a native" by Ned Batchelder...if you ever find yourself looping over indices, yeah: that needs a refactor.
    4
  38. yes you need to python to wrap your C/C++/FORTRAN, which is what numpy and scipy do.
    4
  39. one liner for partitioning integers P(n, k): P = lambda n,k=None: ( sum([P(n,j+1) for j in range(n)]) if k is None else 0 if not n or k>n else 1 if k==1 or k==n else P(n-1,k-1)+P(n-k,k) )
    4
  40. I just can’t deal with type annotations on an add function. Is this not DEVOlution ? Next we’ll be calling it “addi”, with “addf”, “adds”, “addl”…..I can keep going…addc,
    4
  41. ["["".join(letter) for letter in word if letter.isalpha()] for word in "why would you say that?".split() ]
    3
  42. afaik it's "new".
    3
  43. the point of a decorator is reuse code...and lambdas don't do that.
    3
  44. So does python care, or is it just for the IDE.?
    3
  45. Python rule: If you think you need meta classes, you don't.
    3
  46. these type hints have got to go. I mean this python ppl. get your duck on.
    3
  47. Wait, what? How long has there been a matmul dunder bound to 🐌?
    3
  48.  @nibblrrr7124  don't forget unicode() from 2x, but those backward quotes `foo` -> repr(foo) were weird. Though emacs knew about them,
    3
  49. for the last one, I think I prefer the code: >>>join_text = str.join it's more pythonic, since functions are 1st class, and: don't reinvent the wheel.
    3
  50.  @TragicGFuel  RTFM
    3