Getting Started

• Access: https://colab.research.google.com
• Sign in: Google account required.
• Create new: File → New Notebook.
• Open from GitHub: File → Open → GitHub (paste repo URL).
• Upload: File → Upload notebook (from local or Drive).

---

Keyboard Shortcuts

Most Jupyter shortcuts work, plus a few Colab-specific ones.

Shortcut	Action
Ctrl+Enter	Run cell
Shift+Enter	Run cell and select below
Ctrl+M B	Insert cell below
Ctrl+M A	Insert cell above
Ctrl+M D	Delete selected cell
Ctrl+M Z	Undo cell deletion
Ctrl+M Y	Change to Code cell
Ctrl+M M	Change to Markdown cell
Ctrl+M R	Change to Raw cell
Ctrl+M H	Show shortcuts
Ctrl+S	Save (automatic, but manual available)
Ctrl+Shift+P	Command palette
Ctrl+Shift+C	Copy cell

---

Hardware Acceleration

Enable GPU or TPU for faster training.

# Menu: Runtime → Change runtime type
Hardware accelerator: None / GPU / TPU

# Check if GPU is available
import tensorflow as tf
tf.config.list_physical_devices('GPU')

# Check TPU
import os
if 'COLAB_TPU_ADDR' in os.environ:
    print('TPU available')

# CPU info
!cat /proc/cpuinfo | grep "model name" | head -1

# GPU info (NVIDIA)
!nvidia-smi

# RAM
!free -h

---

File System and Data Access

Colab provides a temporary VM with a Linux filesystem.

Mount Google Drive

# Mount Drive for persistent storage
from google.colab import drive
drive.mount('/content/drive')

# Now you can read/write to /content/drive/MyDrive/
!ls /content/drive/MyDrive/

# Example: Load CSV from Drive
import pandas as pd
df = pd.read_csv('/content/drive/MyDrive/data.csv')

Upload/Download Files

# Upload via UI: Files → Upload

# Or programmatically
from google.colab import files
uploaded = files.upload()   # opens file chooser

# Download files
files.download('output.csv')

Access External Data

# From URL (wget/curl)
!wget https://example.com/data.zip
!unzip data.zip

# From GitHub raw
!wget https://raw.githubusercontent.com/user/repo/main/data.csv

---

Magic Commands

Same as IPython/Jupyter with some Colab-specific additions.

System Commands (!)

!ls -la
!pip install numpy
!git clone https://github.com/user/repo
!python script.py

# Capture output
output = !ls -la
print(output)

Line Magics (%)

%time
%timeit
%run script.py
%matplotlib inline
%cd /content/drive/MyDrive/

Cell Magics (%%)

%%bash
echo "Hello from bash"
ls

%%html
HTML output

%%writefile myfile.txt
content to write

---

Environment Variables

# Set environment variables
import os
os.environ['MY_VAR'] = 'value'

# Or use %env
%env MY_VAR=value

# Access
print(os.environ['MY_VAR'])

# Preserve between sessions? Use Drive mount or secrets

---

Secret Management

Store secrets (API keys, passwords) in Colab Secrets.

# Access via userdata
from google.colab import userdata
api_key = userdata.get('API_KEY')

# Set in UI: Keys (left sidebar) → Add key
# Not visible in notebook, secure.

---

Collaboration

• Share: Click "Share" button (top right) → set permissions.
• View-only / Edit / Comment modes.
• Live collaboration – multiple people can edit simultaneously (like Google Docs).
• Comments – add comments on specific cells.
• Version history – File → Revision history.

---

Running Cells with Different Runtimes

• Restart runtime – Runtime → Restart runtime (clears all variables).
• Restart and run all – Runtime → Run all (restarts first).
• Interrupt – Runtime → Interrupt execution (or click stop button).
• Factory reset – Runtime → Factory reset runtime (for clean state).

---

Installing Packages

# Standard pip
!pip install pandas numpy matplotlib

# Install specific version
!pip install tensorflow==2.15.0

# From GitHub
!pip install git+https://github.com/user/repo.git

# Conda (if needed)
!conda install -c conda-forge opencv -y

# Apt-get (for system libraries)
!apt-get update && apt-get install -y ffmpeg

---

Visualisation

# Matplotlib (inline)
%matplotlib inline
import matplotlib.pyplot as plt
plt.plot([1, 2, 3, 4])
plt.show()

# Plotly (interactive)
import plotly.express as px
fig = px.scatter(...)
fig.show()  # renders in Colab

# Seaborn
import seaborn as sns
sns.heatmap(df.corr())

---

Handling Large Files and Data

Colab has limited disk (≈100GB) and RAM (≈12-25GB depending on runtime).

• Use Drive – mount and read/write directly; avoid copying large datasets into VM.
• Use cloud storage – load from S3, GCS, or BigQuery.
• Streaming – use pandas.read_csv() with chunksize.
• Delete unused files – !rm -rf large_file.
• Check disk usage – !df -h /content.
• Garbage collect – import gc; gc.collect().

---

Colab Pro / Pro+ Features

• Better GPUs – A100, V100, T4 (depending on plan).
• More RAM – higher memory instances.
• Longer runtimes – up to 24 hours (vs 12 for free).
• Background execution – notebook keeps running even if browser closed (Pro+).

---

Export / Download

• Save to Drive – File → Save a copy in Drive.
• Download – File → Download .ipynb, .py, .html, etc.
• GitHub – File → Save a copy to GitHub (requires auth).

---

Common Code Snippets

Mount Drive and change working directory

from google.colab import drive
drive.mount('/content/drive')
import os
os.chdir('/content/drive/MyDrive/Colab Notebooks')
!pwd

Clone a GitHub repo

!git clone https://github.com/user/repo.git
%cd repo

Unzip a file

!unzip -q archive.zip -d extracted/

List files in Drive

!ls -l /content/drive/MyDrive/

Read CSV from Drive

import pandas as pd
df = pd.read_csv('/content/drive/MyDrive/data.csv')

Save DataFrame to CSV in Drive

df.to_csv('/content/drive/MyDrive/output.csv', index=False)

---

Best Practices

• Mount Drive at start – for persistent data access.
• Use environment variables – for configurable paths.
• Install dependencies upfront – in the first cell.
• Use %%capture to suppress excessive output.
• Clear output – to reduce notebook size (Edit → Clear all outputs).
• Save to GitHub – version control your notebooks.
• Use userdata – for secrets instead of hard-coding.
• Monitor runtime – use !nvidia-smi for GPU usage.
• Set %matplotlib inline – for plots.
• Restart runtime – before sharing to ensure reproducibility.
• Use !pip install with --quiet to reduce log noise.

---

Troubleshooting

• GPU not detected? – ensure runtime type is GPU; check !nvidia-smi.
• Out of memory? – reduce batch size, use garbage collection, mount Drive.
• Drive not mounted? – re‑run drive.mount() and allow access.
• Package version conflicts? – use !pip install package==version.
• Timeout? – keep notebook active; reconnect if idle.
• Not saving? – Colab auto‑saves, but also use File → Save.