PERF001-heavy-import

Dimension: Performance Severity: Medium Kind: FileLevel (Python only)

Description

Detects top-level import or from ... import statements for heavyweight packages (numpy, pandas, tensorflow, torch, scipy, matplotlib, cv2, sklearn) in Python library files.

Importing a Python module executes all top-level statements in that module at import time. When a library imports a heavyweight package at the top level, every consumer of that library must wait for the heavyweight package to initialize — even if the feature that requires it is never called. For packages like tensorflow or torch, this load time can be hundreds of milliseconds to several seconds.

Examples

Flagged

import pandas           # PERF001: imposes ~100ms+ load on every import of this lib
import torch            # PERF001: imposes ~1s+ load on every import of this lib

def compute(df):
    return df.sum()

Not flagged

def compute(data):
    import pandas as pd  # lazy import — cost paid only when compute() is called
    df = pd.DataFrame(data)
    return df.sum()

Affected packages

Package	Typical load time
tensorflow	1–5 s
torch	0.5–2 s
numpy	100–300 ms
pandas	200–500 ms
scipy	200–400 ms
matplotlib	200–400 ms
cv2 (OpenCV)	100–300 ms
sklearn (scikit-learn)	200–500 ms

Fix

Move the import inside the function or method that uses it (lazy-import pattern):

def train_model(data):
    import torch          # imported only when train_model() is called
    ...

Or use TYPE_CHECKING for type-annotation-only imports:

from __future__ import annotations
from typing import TYPE_CHECKING

if TYPE_CHECKING:
    import pandas as pd   # not executed at runtime

Suppression

# zuit: ignore PERF001-heavy-import
import numpy  # justified: this module is always used with numpy