How to Recover a Deleted Jupyter Notebook

Unsaved or Deleted Jupyter Notebooks

Your IDE might crash, or you might forget to save your changes, assuming they're already saved. You or someone else might accidentally delete your Jupyter Notebook—one you've spent hours working on. For instance:

you've lost your Jupyter Notebook and don't have a backup

you can't find your deleted Jupyter Notebook in your recycle bin

you don't have .ipynb_checkpoints—which are saved snapshots of your Jupyter Notebook

This guide will help you recover your lost Jupyter Notebook, regardless of the reason for its disappearance. Whether you want to avoid rewriting it or need to back it up quickly, you're in the right place.

Recovering unsaved or deleted Jupyter Notebook using IPython kernel history

Every executed code cell is stored in an SQLite database, which you can use to recover your Jupyter Notebook code cells! Unfortunately, only Python/R/Julia code cells are saved, not markdown cells.

First, navigate to your home directory:

On Linux/WSL2/Mac: ~/.ipython/profile_default

On Windows: C:\Users\<YourUsername>\.ipython\profile_default

Make sure to show hidden files:

On Mac: Press Cmd + Shift + . (dot) to reveal hidden files and directories.

On Windows: In File Explorer, go to the "View" tab and check the "Hidden items" box.

Once you're in the .ipython/profile_default directory, you should see the history.sqlite file. This is where all your executed Jupyter Notebook code cells are stored. You can explore this database using any database explorer or terminal commands to inspect its tables and content. However, the following intuitive Jupyter Notebook code will also help you do this:


import sqlite3 as sql  # necessary to load/read your database
import pandas as pd    # necessary to view/manipulate tables
import nbformat as nbf # necessary to recover/nicely format back your recovered code cells

Once you've imported the required packages (if they're not installed, use pip install nbformat or so), all you need to do is provide the correct path to your history.sqlite file. Knowing when you last executed the lost Jupyter Notebook can help you more easily find the right code.


path_to_sqlite_database = '~/.ipython/profile_default/history.sqlite'
start_date = '2024-09-24 09:00:00'
end_date = '2024-09-24 10:00:00'

conn = sql.connect(path_to_sqlite_database )

df = pd.read_sql_query(f"""
	select 
		s.session, s.start, s.end, h.source
	from sessions as s
	join history  as h using (session)
	where s.start between {start_date} and {end_date}""", conn)
df.head()

Now, explore the dataframe and locate your code snippets in the "source" column. Once you find a snippet, note the corresponding session number from the "session" column—this reflects each time you start or restart the Jupyter Kernel. Remember this session number for the next step.


my_sessions = 6000
df_session_content = pd.read_sql_query(f"""
	select *    
	from history 
	where session = {my_session}""", conn)
df_session_content.head()

After creating the above dataframe, filter it to include only distinct rows based on the "source" column. This ensures that running the code cell multiple times within the same session doesn't result in duplicate rows. Finally, use the nbformat package to write the required rows from the "source" column into a new Jupyter Notebook, as shown below:


df_session_content= df_session_content.drop_duplicates(subset='source')
# Create a new Jupyter Notebook
nb = nbf.v4.new_notebook()

# Add each row in df['source'] as a new cell
cells = []
for row in df_session_content['source']:
    cell = nbf.v4.new_code_cell(row)
    cells.append(cell)

nb['cells'] = cells
# Save the notebook to a file
with open('new_notebook.ipynb', 'w') as f:
    nbf.write(nb, f)
print("Jupyter Notebook created successfully.")

Voilà! You have recovered the code cells of your Jupyter Notebook. However, you'll need to manually input all markdown cells if you had them before.

Troubles are like the night... Rest, don’t despair; morning will surely come.