Speed up a slow or laggy Jupyter Notebook

Jupyter Notebook can feel snappy at first and then suddenly become painfully laggy: typing delays, slow cell execution, browser freezes. Let’s look at practical ways to speed things up and keep your notebook usable.

1. Clear heavy outputs and restart

Large DataFrames, long logs, or huge images in many cells can slow down the UI. Try:

• Kernel → Restart & Clear Output.
• Rerun only the cells needed for your current task.
• Avoid printing thousands of lines in loops.

2. Split huge notebooks into smaller ones

A 1,000+ line notebook doing everything from data loading to model training to evaluation will naturally get heavy. Use multiple notebooks:

• 01_eda.ipynb
• 02_training.ipynb
• 03_evaluation.ipynb

3. Offload heavy logic into Python modules

Move complex functions into .py files and import them:

# src/preprocess.py
def clean_data(df):
    # heavy transformations
    return df

# in your notebook
import sys
sys.path.append("src")

from preprocess import clean_data

This keeps the notebook itself lighter and easier to render.

4. Watch memory usage

Too many large variables can cause slowdowns or crashes:

• Delete temporary DataFrames once you’re done: del df_temp.
• Use chunked loading for very large CSVs.
• Consider downcasting numeric dtypes to save RAM.

5. Use a faster environment if needed

If you’re running locally on a low-spec machine, moving the same notebook to Google Colab (with free GPU/TPU for ML workloads) or a beefier server can help.

6. Capture a clean “fast” version with NoteCapsule

Once you’ve cleaned outputs, split notebooks, and optimized memory, take a NoteCapsule snapshot. That way, you have a baseline state that loads quickly and runs reliably.

from notebookcapsule import create_capsule

create_capsule(
    name="optimized-notebook-state",
    notebook_path="notebooks/training.ipynb",
    data_dirs=["./data", "./models"],
    base_dir=".",  # project root
)

This Capsule becomes your “known good” version if later changes slow everything down again.