type check in unit test

[Trinkle23897]

Fix #195: Add mypy test in .github/workflows/docs_and_lint.yml.

Also remove the out-of-the-date api

[Trinkle23897]

UPDATE: resolved

There are many errors like

tianshou/policy/imitation/base.py:45: error: Return type "Dict[str, float]" of "learn" incompatible with return type "Dict[str, Union[float, List[float]]]" in supertype "BasePolicy"  [override]

Possibly a bug in mypy:

from typing import Union

class A:
    def test(self) -> Union[float, int]:
        pass

class B(A):
    def test(self) -> int:
        pass

This works in mypy check. However,

from typing import Dict, Union

class A:
    def test(self) -> Dict[str, Union[float, int]]:
        pass

class B(A):
    def test(self) -> Dict[str, int]:
        pass

This cannot pass the check:

$ mypy try.py
try.py: note: In member "test" of class "B":
try.py:8: error: Return type "Dict[str, int]" of "test" incompatible with return type "Dict[str, Union[float, int]]" in supertype "A"  [override]
Found 1 error in 1 file (checked 1 source file)

However, here says narrow type in subclass is acceptable.

python/mypy#9418

[codecov-commenter]

Codecov Report

Merging #200 into master will increase coverage by 0.32%.
The diff coverage is 95.45%.

@@            Coverage Diff             @@
##           master     #200      +/-   ##
==========================================
+ Coverage   94.09%   94.42%   +0.32%     
==========================================
  Files          40       39       -1     
  Lines        2457     2438      -19     
==========================================
- Hits         2312     2302      -10     
+ Misses        145      136       -9     

Flag	Coverage Δ
#unittests	94.42% <95.45%> (+0.32%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
tianshou/exploration/__init__.py	100.00% <ø> (ø)
tianshou/env/worker/dummy.py	91.30% <83.33%> (ø)
tianshou/policy/modelfree/sac.py	87.50% <84.21%> (+1.04%)	⬆️
tianshou/env/worker/ray.py	85.29% <87.50%> (+2.43%)	⬆️
tianshou/env/worker/subproc.py	91.21% <88.46%> (+0.05%)	⬆️
tianshou/policy/modelfree/ppo.py	96.47% <92.85%> (-0.05%)	⬇️
tianshou/data/utils/segtree.py	70.27% <93.33%> (+4.64%)	⬆️
tianshou/exploration/random.py	97.29% <93.33%> (ø)
tianshou/policy/multiagent/mapolicy.py	93.75% <93.33%> (ø)
tianshou/data/collector.py	94.44% <95.45%> (+2.95%)	⬆️
... and 21 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 380e9e9...62d875a. Read the comment docs.

[duburcqa]

My overall comment is that typing checking should make things more robust and clearer. It is clearly not the case at the current point, so I do not recommand satisfying all its recommendations at this point. Maybe it will get better in the future but it does not look viable for now. Just my opinion though.

[Trinkle23897]

My overall comment is that typing checking should make things more robust and clearer. It is clearly not the case at the current point, so I do not recommand satisfying all its recommendations at this point. Maybe it will get better in the future but it does not look viable for now. Just my opinion though.

Agree most part. But I think it’s a tradeoff, especially in Python such a dynamic programming language. If we use other type checker, possibly the same things will happen again. Also, mypy is continuing improving, so I don’t pin a specific version and once the advanced features are added into mypy (like variable type re-definition), we can modify our code with minimal changes.
I think the most important part of type checker is to find the potential bugs, thus we need it. But we cannot notice the potential bugs under many type errors.