kengz/SLM-Lab

View on GitHub
slm_lab/agent/algorithm/ppo.py

Summary

Maintainability
D
1 day
Test Coverage
B
89%

Function train has a Cognitive Complexity of 37 (exceeds 10 allowed). Consider refactoring.
Open

    def train(self):
        if util.in_eval_lab_modes():
            return np.nan
        clock = self.body.env.clock
        if self.to_train == 1:
Severity: Minor
Found in slm_lab/agent/algorithm/ppo.py - About 4 hrs to fix

Cognitive Complexity

Cognitive Complexity is a measure of how difficult a unit of code is to intuitively understand. Unlike Cyclomatic Complexity, which determines how difficult your code will be to test, Cognitive Complexity tells you how difficult your code will be to read and comprehend.

A method's cognitive complexity is based on a few simple rules:

  • Code is not considered more complex when it uses shorthand that the language provides for collapsing multiple statements into one
  • Code is considered more complex for each "break in the linear flow of the code"
  • Code is considered more complex when "flow breaking structures are nested"

Further reading

Cyclomatic complexity is too high in method train. (12)
Open

    def train(self):
        if util.in_eval_lab_modes():
            return np.nan
        clock = self.body.env.clock
        if self.to_train == 1:
Severity: Minor
Found in slm_lab/agent/algorithm/ppo.py by radon

Cyclomatic Complexity

Cyclomatic Complexity corresponds to the number of decisions a block of code contains plus 1. This number (also called McCabe number) is equal to the number of linearly independent paths through the code. This number can be used as a guide when testing conditional logic in blocks.

Radon analyzes the AST tree of a Python program to compute Cyclomatic Complexity. Statements have the following effects on Cyclomatic Complexity:

Construct Effect on CC Reasoning
if +1 An if statement is a single decision.
elif +1 The elif statement adds another decision.
else +0 The else statement does not cause a new decision. The decision is at the if.
for +1 There is a decision at the start of the loop.
while +1 There is a decision at the while statement.
except +1 Each except branch adds a new conditional path of execution.
finally +0 The finally block is unconditionally executed.
with +1 The with statement roughly corresponds to a try/except block (see PEP 343 for details).
assert +1 The assert statement internally roughly equals a conditional statement.
Comprehension +1 A list/set/dict comprehension of generator expression is equivalent to a for loop.
Boolean Operator +1 Every boolean operator (and, or) adds a decision point.

Source: http://radon.readthedocs.org/en/latest/intro.html

Avoid deeply nested control flow statements.
Open

                        for k, v in minibatch.items():
                            if k not in ('advs', 'v_targets'):
                                minibatch[k] = math_util.venv_pack(v, self.body.env.num_envs)
                    advs, v_targets = minibatch['advs'], minibatch['v_targets']
Severity: Major
Found in slm_lab/agent/algorithm/ppo.py - About 45 mins to fix

    Identical blocks of code found in 2 locations. Consider refactoring.
    Open

                        if self.shared:  # shared network
                            loss = policy_loss + val_loss
                            self.net.train_step(loss, self.optim, self.lr_scheduler, clock=clock, global_net=self.global_net)
                        else:
                            self.net.train_step(policy_loss, self.optim, self.lr_scheduler, clock=clock, global_net=self.global_net)
    Severity: Major
    Found in slm_lab/agent/algorithm/ppo.py and 1 other location - About 6 hrs to fix
    slm_lab/agent/algorithm/actor_critic.py on lines 291..297

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 107.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 7 locations. Consider refactoring.
    Open

                logger.debug(f'Trained {self.name} at epi: {clock.epi}, frame: {clock.frame}, t: {clock.t}, total_reward so far: {self.body.env.total_reward}, loss: {loss:g}')
    Severity: Major
    Found in slm_lab/agent/algorithm/ppo.py and 6 other locations - About 1 hr to fix
    slm_lab/agent/algorithm/actor_critic.py on lines 300..300
    slm_lab/agent/algorithm/dqn.py on lines 148..148
    slm_lab/agent/algorithm/reinforce.py on lines 160..160
    slm_lab/agent/algorithm/sac.py on lines 225..225
    slm_lab/agent/algorithm/sarsa.py on lines 147..147
    slm_lab/agent/algorithm/sil.py on lines 148..148

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 52.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    Identical blocks of code found in 3 locations. Consider refactoring.
    Open

            if self.entropy_coef_spec is not None:
                self.entropy_coef_scheduler = policy_util.VarScheduler(self.entropy_coef_spec)
                self.body.entropy_coef = self.entropy_coef_scheduler.start_val
    Severity: Minor
    Found in slm_lab/agent/algorithm/ppo.py and 2 other locations - About 30 mins to fix
    slm_lab/agent/algorithm/actor_critic.py on lines 102..104
    slm_lab/agent/algorithm/reinforce.py on lines 69..71

    Duplicated Code

    Duplicated code can lead to software that is hard to understand and difficult to change. The Don't Repeat Yourself (DRY) principle states:

    Every piece of knowledge must have a single, unambiguous, authoritative representation within a system.

    When you violate DRY, bugs and maintenance problems are sure to follow. Duplicated code has a tendency to both continue to replicate and also to diverge (leaving bugs as two similar implementations differ in subtle ways).

    Tuning

    This issue has a mass of 40.

    We set useful threshold defaults for the languages we support but you may want to adjust these settings based on your project guidelines.

    The threshold configuration represents the minimum mass a code block must have to be analyzed for duplication. The lower the threshold, the more fine-grained the comparison.

    If the engine is too easily reporting duplication, try raising the threshold. If you suspect that the engine isn't catching enough duplication, try lowering the threshold. The best setting tends to differ from language to language.

    See codeclimate-duplication's documentation for more information about tuning the mass threshold in your .codeclimate.yml.

    Refactorings

    Further Reading

    There are no issues that match your filters.

    Category
    Status