Debugging research code

Fig. 11 Debugging the tool used for debugging. “Debugger” by Randall Munroe, licensed CC BY-NC 2.5.

Debugging is an investigation, not a sign that you are bad at programming.

Read tracebacks from the bottom

Traceback (most recent call last):
  ...
ValueError: operands could not be broadcast together with shapes (32,500) (32,)

Start with:

1. Error type: ValueError

2. Error message: incompatible shapes

3. Your line nearest the bottom

Inspect the smallest useful facts

print(type(value))
print(array.shape)
print(frame.columns.tolist())
print(frame.dtypes)
print(path.resolve())
print(path.exists())

Common error patterns

Error	First question
NameError	Was the variable created in this execution?
KeyError	Which dictionary keys or DataFrame columns exist?
IndexError	What is the size of that axis?
AttributeError	What is the object’s actual type?
ModuleNotFoundError	Which interpreter/environment is running?
broadcasting ValueError	What are both shapes?
file not found	What does the resolved path say?

Reduce the example

Replace a large pipeline with a tiny object that reproduces the error:

small = epochs[:2, :3, :10]
print(small.shape)

Ask a high-quality question

Include:

• what you expected;

• what happened;

• the complete error;

• relevant types and shapes;

• a minimal code example;

• what you already tried.