iTranslated by AI
Revisiting Google Colab in 2025: Discovering Overlooked Features
Introduction
This document is a memo of my "re-investigation" into the features of Google Colab as of 2025/09/25, which I used to pay for in the past. For those using Google Colab for the first time, please refer to other introductory documents. Also, since I am currently not subscribed, paid features have not been verified.
Screen Effects
There are settings to apply screen effects. They provide a glimmer of hope when you're fed up with everything.
In the menu at the top of the screen, select Tools → Settings to display the settings dialog.
Select "Miscellaneous"
Power Level
You can make sparks fly while coding.
Corgi Mode/Cat Mode/Crab Mode
Corgis, cats, and crabs will move across the top of the screen.
Changing the Runtime Type
Runtime Versions
It is now possible to select past runtimes.
Previously, there were cases where code stopped working due to version updates, but with the ability to select past runtimes, reproducing past code has become easier.
Past runtime versions are available for one year after their release.
Frequently Asked Questions - Past Runtime Versions
Runtime Types
In addition to Python 3, R and Julia are now available.
Example of Python
Code
import numpy as np
# Sample matrix/vector (Same as Python)
A = np.array([[3, 1, 2],
[0, 1, 4],
[2, 3, 1]], dtype=float)
B = np.array([[1, 2, 3],
[4, 5, 6],
[7, 8, 9]], dtype=float)
b = np.array([1, 2, 3], dtype=float)
# 1) Matrix multiplication
C = A @ B
# 2) Transpose
AT = A.T
# 3) Determinant
detA = np.linalg.det(A)
# 4) Inverse matrix
invA = np.linalg.inv(A)
# 5) Solution to the system of linear equations A x = b
x = np.linalg.solve(A, b)
# 6) Eigenvalues (if you want eigenvectors, it's the 2nd return value of eig)
w, v = np.linalg.eig(A) # w: eigenvalues, v: column vectors are eigenvectors
# 7) Element-wise product (Hadamard product)
hadamard = A * B
print("C=\n", C)
print("A^T=\n", AT)
print("det(A)=", detA)
print("inv(A)=\n", invA)
print("x=\n", x)
print("eigvals(A)=\n", w)
print("A .* B=\n", hadamard)
Output
C=
[[21. 27. 33.]
[32. 37. 42.]
[21. 27. 33.]]
A^T=
[[3. 0. 2.]
[1. 1. 3.]
[2. 4. 1.]]
det(A)= -29.000000000000018
inv(A)=
[[ 0.37931034 -0.17241379 -0.06896552]
[-0.27586207 0.03448276 0.4137931 ]
[ 0.06896552 0.24137931 -0.10344828]]
x=
[-0.17241379 1.03448276 0.24137931]
eigvals(A)=
[-2.62721301 1.94136684 5.68584617]
A .* B=
[[ 3. 2. 6.]
[ 0. 5. 24.]
[14. 24. 9.]]
Example of R
Code
# Sample matrix/vector (Same as Python)
A <- matrix(c(3,1,2,
0,1,4,
2,3,1), nrow=3, byrow=TRUE)
B <- matrix(c(1,2,3,
4,5,6,
7,8,9), nrow=3, byrow=TRUE)
b <- c(1,2,3)
# 1) Matrix multiplication
C <- A %*% B
# 2) Transpose
AT <- t(A)
# 3) Determinant
detA <- det(A)
# 4) Inverse matrix
invA <- solve(A)
# 5) Solution to the system of linear equations A x = b
x <- solve(A, b)
# 6) Eigenvalues (Eigenvectors are in $vectors)
eig <- eigen(A) # eig$values, eig$vectors
# 7) Element-wise product (Hadamard product)
hadamard <- A * B
print(C)
print(AT)
print(detA)
print(invA)
print(x)
print(eig$values)
print(hadamard)
Output
The order of eigenvalues may differ from Python.
[,1] [,2] [,3]
[1,] 21 27 33
[2,] 32 37 42
[3,] 21 27 33
[,1] [,2] [,3]
[1,] 3 0 2
[2,] 1 1 3
[3,] 2 4 1
[1] -29
[,1] [,2] [,3]
[1,] 0.37931034 -0.17241379 -0.06896552
[2,] -0.27586207 0.03448276 0.41379310
[3,] 0.06896552 0.24137931 -0.10344828
[1] -0.1724138 1.0344828 0.2413793
[1] 5.685846 -2.627213 1.941367
[,1] [,2] [,3]
[1,] 3 2 6
[2,] 0 5 24
[3,] 14 24 9
Example of Julia
Example Code
using LinearAlgebra # det, inv, eigen, etc.
# Sample matrix/vector (Same as Python)
A = [3.0 1.0 2.0;
0.0 1.0 4.0;
2.0 3.0 1.0]
B = [1.0 2.0 3.0;
4.0 5.0 6.0;
7.0 8.0 9.0]
b = [1.0, 2.0, 3.0]
# 1) Matrix multiplication
C = A * B
# 2) Transpose (non-conjugate transpose)
AT = transpose(A)
# 3) Determinant
detA = det(A)
# 4) Inverse matrix
invA = inv(A)
# 5) Solution to the system of linear equations A x = b (\ is recommended in Julia)
x = A \ b
# 6) Eigenvalues (Eigenvectors are in .vectors)
eig = eigen(A) # eig.values, eig.vectors
# 7) Element-wise product (Hadamard product)
hadamard = A .* B
println("C=\n", C)
println("A^T=\n", AT)
println("det(A)=", detA)
println("inv(A)=\n", invA)
println("x=\n", x)
println("eigvals(A)=\n", eig.values)
println("A .* B=\n", hadamard)
Output
C=
[21.0 27.0 33.0; 32.0 37.0 42.0; 21.0 27.0 33.0]
A^T=
[3.0 0.0 2.0; 1.0 1.0 3.0; 2.0 4.0 1.0]
det(A)=-29.000000000000004
inv(A)=
[0.3793103448275862 -0.17241379310344823 -0.06896551724137932; -0.27586206896551724 0.03448275862068964 0.41379310344827586; 0.0689655172413793 0.24137931034482757 -0.10344827586206895]
x=
[-0.1724137931034482, 1.0344827586206895, 0.24137931034482757]
eigvals(A)=
[-2.6272130052966602, 1.9413668397423212, 5.685846165554339]
A .* B=
[3.0 2.0 6.0; 0.0 5.0 24.0; 14.0 24.0 9.0]
Type Checking
There is an option to perform type checking.
By default, type checking is not performed, as shown below.
Open the settings dialog from the menu: Tools → Settings.
In the settings dialog, go to Editor → Code diagnostics and select "Syntax and type checking".
This enables type checking.
Secret Keys
You can set secret keys by selecting the icon in the left navigation menu.
Example of retrieving a secret key in Python
from google.colab import userdata
userdata.get('test_key')
It is likely unavailable when using R or Julia runtimes.
Methods I tried that didn't work
While you can run Python from R and other languages, the Python instance running is not the Colab Python, so it seems impossible to retrieve secrets.
library(reticulate)
out <- py_capture_output({
py_run_string("
from google.colab import userdata
try:
print(userdata.get('test_key'))
except Exception as e:
print(e)
")
})
cat(out)
'NoneType' object has no attribute 'kernel'
Also, no library corresponding to google.colab was found.
GitHub Integration
Integration with GitHub, including private repositories, is possible through the following steps:
- Select Tools → Settings to open the settings dialog.
- Select the GitHub tab in the settings dialog.
The steps to open a notebook are as follows:
File menu → Open notebook
Select the GitHub tab and choose the repository and branch.
Saving the file allows you to commit to GitHub.
Slideshow
You can run a slideshow for presentations.
View → Start slideshow
Opening in Playground Mode
The notebook opens in read-only mode, and edits are made to a copy. No changes are made to the original repository or Drive.
Notebook Diff
In the menu at the top of the screen: Tools → Notebook diff
Summary
By the way, most of the features I found this time have existed for more than a year.
- Screen effects existed in 2019
- Selecting past runtimes September 2025
- R runtime existed around 2018
- Julia runtime around March 2025
- Type checking existed in 2024
- Secret keys November 2023
- GitHub integration existed in 2021
- Slideshow around July 2025
- Playground mode existed in 2018
- Notebook diff existed in 2019
It's surprising how much you don't notice when you're just using it out of habit.
Discussion