DEV Community

Cover image for Google Colab Tips for Power Users
Amit Chaudhary
Amit Chaudhary

Posted on • Edited on • Originally published at amitness.com

Google Colab Tips for Power Users

#machinelearning #jupyter #datascience

Colab is one of the best products to come from Google. It has made GPUs freely accessible to learners and practitioners like me who otherwise wouldn't be able to afford a high-end GPU.

While the interface is very easy to use, there are many lesser-known and undocumented features in colab. In this post, I will share those features that I've discovered from their official talks and basic usage.

1. Scratchpad Notebook

It's a pretty common scenario that we have a bunch of cluttered untitled notebooks created when we try out temporary stuff on colab.

Clutter of Untitled Notebooks in Colab

To solve this, you can bookmark the link given below. It will open a special scratch notebook and any changes you make to that notebook are not saved to your main account.

https://colab.research.google.com/notebooks/empty.ipynb

2. Timing Execution of Cell

It's pretty common that we manually calculate the difference between start and end times of a piece of code to gauge the time taken.

Colab provides an inbuilt feature to do this. After a cell is executed, just hover over the cell run icon and you will get an estimate of the execution time taken.

Execution Time by hovering on run cell

3. Run part of a cell

You can also run only a part of the cell by selecting it and pressing the Runtime > Run Selection button or using the keyboard shortcut Ctrl + Shift + Enter.

Running specific line in colab

4. Jupyter Notebook Keyboard Shortcuts

If you are familiar with keyboard shortcuts from Jupyter Notebook, they don't work directly in Colab. But I found a mental model to map between them.

Just add Ctrl + M before whatever keyboard shortcut you were using in Jupyter. This rule of thumb works for the majority of common use-cases.

Action Jupyter Notebook Google Colab
Add a cell above A Ctrl + M + A
Add a cell below B Ctrl + M + B
See all keyboard shorcuts H Ctrl + M + H
Change cell to code Y Ctrl + M + Y
Change cell to markdown M Ctrl + M + M
Interrupt the kernel II Ctrl + M + I
Delete a cell DD Ctrl + M + D
Checkpoint notebook Ctrl + S Ctrl + M + S

Below are some notable exceptions to this rule for which either the shortcut is changed completely or kept the same.

Action Jupyter Notebook Google Colab
Restart runtime 00 Ctrl + M + .
Run cell Ctrl + Enter Ctrl + Enter
Run cell and add new cell below Alt + Enter Alt + Enter
Run cell and goto the next cell below Shift + Enter Shift + Enter
Comment current line Ctrl + / Ctrl + /

5. Jump to Class Definition

Similar to an IDE, you can go to a class definition by pressing Ctrl and then clicking a class name. For example, here we view the class definition of the Dense layer in Keras by pressing Ctrl and then clicking the Dense class name.

Demo of jumping to class definition

6. Open Notebooks from GitHub

The Google Colab team provides an official chrome extension to open notebooks on GitHub directly on colab. You can install it from here.

After installation, click the colab icon on any GitHub notebook to open it directly.

Extension for opening github notebook in colab

Alternatively, you can also manually open any GitHub notebook by replacing github.com with colab.research.google.com/github.

https://github.com/fastai/course-v3/blob/master/nbs/dl1/00_notebook_tutorial.ipynb

to

https://colab.research.google.com/github/fastai/course-v3/blob/master/nbs/dl1/00_notebook_tutorial.ipynb

7. Run Flask apps from Colab

With a library called flask-ngrok, you can easily expose a Flask web app running on colab to demo prototypes. First, you need to install flask and flask-ngrok.

!pip install flask-ngrok flask==0.12.2
Enter fullscreen mode Exit fullscreen mode

Then, you just need to pass your flask app object to run_with_ngrok function and it will expose a ngrok endpoint when the server is started.

from flask import Flask
from flask_ngrok import run_with_ngrok

app = Flask(__name__)
run_with_ngrok(app)

@app.route('/')
def hello():
    return 'Hello World!'

if __name__ == '__main__':
    app.run()
Enter fullscreen mode Exit fullscreen mode

Example of running flask-ngrok

You can try this out from the package author's official example on Colab.

8. Switch between Tensorflow versions

You can easily switch between Tensorflow 1 and Tensorflow 2 using this magic flag.

To switch to Tensorflow 1.15.2, use this command:

%tensorflow_version 1.x
Enter fullscreen mode Exit fullscreen mode

To switch to Tensorflow 2.2, run this command:

%tensorflow_version 2.x
Enter fullscreen mode Exit fullscreen mode

You will need to restart the runtime for the effect to take place. Colab recommends using the pre-installed Tensorflow version instead of installing it from pip for performance reasons.

9. Tensorboard Integration

Colab also provides a magic command to use Tensorboard directly from the notebook. You just need to set the logs directory location using the --logdir flag. You can learn to use it from the official notebook.

%load_ext tensorboard
%tensorboard --logdir logs
Enter fullscreen mode Exit fullscreen mode

Embedded Tensorboard in Colab

10. Gauge resource limits

Colab provides the following specs for their free and pro versions. Based on your use case, you can switch to the pro version at $10/month if you need a better runtime, GPU, and memory.

Version GPU GPU Ram RAM Storage CPU Cores Idle Timeout Maximum Runtime
Free Tesla K80 11.44GB 13.7GB 37GB 2 90 min 12 hrs
Pro Tesla P100 16GB 27.4GB 37GB 4 90 min 24 hrs

You can view the GPU you have been assigned by running the following command

!nvidia-smi
Enter fullscreen mode Exit fullscreen mode

For information on the CPU, you can run this command

!cat /proc/cpuinfo
Enter fullscreen mode Exit fullscreen mode

Similarly, you can view the RAM capacity by running

import psutil
ram_gb = psutil.virtual_memory().total / 1e9
print(ram_gb)
Enter fullscreen mode Exit fullscreen mode

11. Use interactive shell

There is no built-in interactive terminal in Colab. But you can use the bash command to try out shell commands interactively. Just run this command and you will get an interactive input.

!bash
Enter fullscreen mode Exit fullscreen mode

Now, you can run any shell command in the given input box.

Using interactive shell in colab

To quit from the shell, just type exit in the input box.

Exiting interactive shell in colab

12. Current memory and storage usage

Colab provides an indicator of RAM and disk usage. If you hover over the indicator, you will get a popup with the current usage and the total capacity.

Showing current memory and ram usage in colab

13. "Open in Colab" Badge

You can add a 'Open in Colab' badge to your README.md or jupyter notebooks using the following markdown code.

Open In Colab

In the markdown code, we're loading an SVG image and then linking it to a colab notebook.

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/notebooks/basic_features_overview.ipynb)
Enter fullscreen mode Exit fullscreen mode

14. Interactive Tables for Pandas

Colab provides a notebook extension to add interactive sorting and filtering capabilities to pandas dataframes. To use it, run the following code.

%load_ext google.colab.data_table
Enter fullscreen mode Exit fullscreen mode

You can see the regular pandas dataframe and the interactive dataframe after loading the extension below.

Regular pandas dataframe output

Interactive pandas dataframe output

15. Setup Conda environment

If you use miniconda as your python environment manager, you can setup it on colab by running this command at the top of your notebook.

# Download Miniconda installation script
!wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

# Make it executable
!chmod +x Miniconda3-latest-Linux-x86_64.sh

# Start installation in silent mode
!bash ./Miniconda3-latest-Linux-x86_64.sh -b -f -p /usr/local

# Make conda packages available in current environment
import sys
sys.path.append('/usr/local/lib/python3.7/site-packages/')
Enter fullscreen mode Exit fullscreen mode

After the cell is executed, you can use conda to install packages as usual.

!conda install -y flask
Enter fullscreen mode Exit fullscreen mode

16. Manage Colab Notebooks from Command Line

You can use a library called colab-cli to easily create and sync colab notebooks with your local notebooks.

colab-cli-demo

17. Run background tasks

There are use-cases when we need to start some web server or background tasks before we can execute our regular program.

To run background tasks, use the nohup command followed by your regular shell command and add & to the end to run it in the background. This makes sure that you can run cells afterward in the notebook without your background task blocking it.

!nohup bash ping.sh &
Enter fullscreen mode Exit fullscreen mode

18. Notify on Training Completion

If you're running a long task such as training a model, you can setup Colab to send a desktop notification once it's completed.

To enable that, goto Tools ⮕ Settings ⮕ Site and enable Show desktop notifications checkbox.

You will get a popup to enable browser notification. Just accept it and colab will notify you on task completion even if you are on another tab, window or application.

19. Run javascript code

You can run javascript code by using the %%javascript magic command.

References

Top comments (0)

Read next

shubhsk profile image

AI Engineer's Tool Review: Haystack

shubhanshu -

ppaanngggg profile image

YOLOv11: A New Breakthrough in Document Layout Analysis

ppaanngggg -

shubhsk profile image

AI Engineer's Tool Review: Fireworks AI

shubhanshu -

td_inc profile image

Skills Required To Become A Machine Learning (ML) Engineer

TechDogs -