Python 3, an illustrated tour Transcripts
Chapter: Type Annotations
Lecture: mypy (type consistency verification)
0:00 So let's look at an example of using the mypy tool.
0:03 I'm going to add typing to a little project I have, it's a markov chain,
0:08 so you can check out this GitHub repository if you want to look at it,
0:11 but here's how I do it, I'm in a virtual environment
0:14 and I say pip install mypy, that's going to go out and fetch the mypy tool
0:17 and I'm going to clone this GitHub repository that I have,
0:20 and I'll change into that directory,
0:22 and in there, there's a file called markov.py, I'm just going to run mypy,
0:26 which gets installed as a binary when I install the mypy tool
0:30 and I run that on markov.py and it will return no output.
0:34 And again, why this returns no output is because mypy supports gradual typing
0:39 it ignores code that doesn't have annotations and this code didn't have any annotations,
0:44 so it's not going to have any output there.
0:46 If I want to get a little bit more ambitious, I can put -- strict after mypy
0:50 that turns on a bunch of features
0:53 and I'm going to get a bunch of warnings or errors from the results here,
0:57 it's going to say this function is missing a type annotation
0:59 we're calling some other functions in a type context and they're not typed.
1:04 And so these are the sorts of things that mypy can find for us.
1:08 Again, note that it also supports this gradual typing
1:11 and so if we leave off the strict,
1:13 it's just going to ignore anything that we haven't annotated.
1:16 So here are a few hints for adding annotations,
1:18 2 ways that you can do it, you can start from the outside code that gets called
1:22 and calls other code and start calling annotating this outer code,
1:26 alternatively you can start wrapping inside code that gets called and annotating that first.
1:32 Either one of those will work.
1:35 What is important for me is if I've got a public interface,
1:38 I want to make sure that there's typing around it
1:41 and that it's clear what comes in and out.
1:44 So I'm going to start annotating something that I think is important
1:48 and I'm going to run the mypy on some file.
1:51 It might complain because it's going to start type checking where I've annotated
1:55 and then I might need to go in and fix things or add more annotations.
1:59 And if I want to get ambitious again, I can use this -- strict
2:03 and that will turn on a bunch of flags and add a bunch more checks for me.
2:08 But basically, after I've gone through this process on my markov file here,
2:12 I'll have a dif that looks something like this.
2:15 So I'm going to end up importing from the typing module the dict and list types
2:21 and I'm going to make a table result variable here or type
2:26 and it's going to be this structure here.
2:29 It's going to be a dictionary that maps a string to another dictionary
2:33 and inside that dictionary, we map a string to account.
2:36 So this code if you're not familiar with it, it creates a markov chain
2:40 a markov chain takes input and gives you some output based on what your input is,
2:45 and in this case, markov chain is typically used in text prediction
2:49 or if you're typing, predicting what characters to come next
2:53 and so you can feed a paragraph or a bunch of text into this
2:56 and it will be able to tell you if I have a, what comes after a,
3:00 after a comes maybe p because we're spelling apple or something like that.
3:05 That's the tooling that the markov chain allows you to do.
3:09 And so here in my constructor here, I've got data that's coming in
3:12 and I've got size that's an optional value here.
3:15 And when I annotate that, I'm going to say data is going to be a string,
3:20 size is going to be an int and my constructor returns none.
3:24 This is the way that you annotate a constructor.
3:28 Also note that I've got a variable here, an instance variable called self.tables,
3:33 and I am annotating that and that is going to be a list of table results.
3:37 So maybe you can see the reason why I made this table result variable here or type
3:42 is because it makes it a little bit more clear
3:44 I would have this nested list of dictionaries of dictionaries
3:48 and I can just clearly read that this is a list of table results.
3:51 Here's another method that got type annotated.
3:54 So predict takes a string of input.
3:57 So we've annotated that and returns the string that's going to come after that input
4:02 if we feed an a we should get p out, something like that
4:05 and you'll note that I annotated just the method parameters
4:09 and the method what it returns, but there is one more annotation in here.
4:13 I didn't annotate a bunch of the variables inside of here
4:16 because mypy didn't complain about those,
4:18 but it did complain about this guy down here
4:21 and the reason is because I've got a variable called result
4:25 that is looping over this options.items collection,
4:29 and then I'm also reusing that same variable result down later
4:33 to randomly choose out of my possible guys what comes next
4:38 because I'm looping over something that might be empty,
4:42 in this case result could be none and that confuses mypy
4:47 but what's really happening here is this is actually indicated
4:50 that my reuse of this variable, this was a bug on my part,
4:55 I shouldn't have reused this variable name
4:57 and so mypy said, well, you've either got to type it or change the name.
5:01 So in this case, I add the typing and mypy doesn't complain about it anymore.
5:05 But the correct thing to do here would be to actually change that variable name.
5:09 You could call this, this is the input and count rather than the resulting count there.
5:15 Here's another example of the annotation that I added
5:19 this get table function accepts a line, which is a string
5:22 and the number of characters that we're going to process as input.
5:26 So we could process a single character after a comes p,
5:29 but we could also say I want to process a and p
5:31 and after a and p comes another p for apple or whatnot,
5:34 if you add more memory to this markov chain,
5:37 it makes better predictions and can make sentences or paragraphs or that sort of thing.
5:41 And we're going to also say that this get table returns a table result,
5:45 recall that I defined this table result couple slides back, which is a nested dictionary here.
5:50 But again, it's a lot more readable to have this table result defined
5:54 and reuse that table result rather than throwing this nested code around all over the place
6:01 table result is very clear and should make sense.
6:05 So after doing that, I think my code is more clear, it should be more clear
6:09 and people who are coming to it should have a very good understanding
6:13 of what is the input and what is the output.
6:16 I also found a possible bug by reusing the result variable
6:19 so I could annotate that, in retrospect I should have just renamed the variable
6:23 but mypy can help you find these sorts of issues.