The Cosmograph Widget brings the power of Cosmograph's GPU-accelerated force layout graph visualization right into your Jupyter notebooks. Built on top of Anywidget, this widget provides a seamless, interactive graph exploration experience directly within your data science workflow, making it easier than ever to visualize complex data relationships and embeddings.
- Interactive Visualization: Pan, zoom, select, and hover to explore large, complex network graphs right in your notebook.
- GPU-Acceleration: Powered by @cosmograph/cosmos, it delivers smooth interactions and rapid rendering of large-scale graphs.
- Seamless Integration: Embeds naturally into Jupyter notebooks, JupyterLab, and other notebook-based environments.
- Rich Configuration API: Fine-tune graph appearance, behavior, and layout parameters through easy-to-use Python APIs.
Note: You'll need python 3.10 or greater.
To install the Cosmograph widget, simply run:
pip install cosmographOnce installed, you can start using it in your notebooks immediately.
After installation, you can import and use the widget in any Python-based notebook environment:
import pandas as pd
from cosmograph import cosmo
points = pd.DataFrame({
'id': [1, 2, 3, 4, 5],
'label': ['Node A', 'Node B', 'Node C', 'Node D', 'Node E'],
'value': [10, 20, 15, 25, 30],
'category': ['A', 'B', 'A', 'B', 'A']
})
links = pd.DataFrame({
'source': [1, 2, 3, 1, 2],
'target': [2, 3, 4, 5, 4],
'value': [1.0, 2.0, 1.5, 0.5, 1.8]
})
widget = cosmo(
points=points,
links=links,
point_id_by='id',
link_source_by='source',
link_target_by='target',
point_color_by='category',
point_include_columns=['value'],
point_label_by='label',
link_include_columns=['value'],
)
widgetThe widget will render an interactive graph visualization inline, allowing you to explore and manipulate your data directly.
You also use the widget object to interact with the rendered graph.
widget.fit_view() # recenter the view (often useful when you've lost your graph (or within your graph)
widget.selected_point_ids # if you've selected some points and want to get info about the selection...
# etc.Let's download a big dataset of English words, plus some hyponym-hypernym relationships. (A hyponym-hypernym relationship is a “type-of” relationship where a hyponym is a more specific term (e.g., “dog”) and a hypernym is a broader term (e.g., “animal”).)
import pandas as pd
from cosmograph import cosmo
df = pd.read_parquet('https://www.dropbox.com/scl/fi/4mnk1e2wx31j9mdsjzecy/wordnet_feature_meta.parquet?rlkey=ixjiiso80s1uk4yhx1v38ekhm&dl=1')
hyponyms = pd.read_parquet('https://www.dropbox.com/scl/fi/pl72ixv34soo1o8zanfrz/hyponyms.parquet?rlkey=t4d606fmq1uinn29qmli7bx6r&dl=1')Peep at the data:
print(f"{df.shape=}")
df.iloc[0]print(f"{hyponyms.shape=}")
hyponyms.iloc[0]Let's plot the data using the UMAP projection of the (OpenAI) embeddings of the words, coloring by "part-of-speech" and sizing by the usage frequency of the word.
g = cosmo(
df,
point_id_by='lemma',
point_label_by='word',
point_x_by='umap_x',
point_y_by='umap_y',
point_color_by='pos',
point_size_by='frequency',
point_size_scale=0.01, # often have to play with this number to get the size right
)
g
Zooming in a bit:
And now, let's put some hypernym-hyponym links, and let the network converge to a stable layout using a force-directed simulation (try it yourself, the convergence is pretty!)
h = cosmo(
points=df,
links=hyponyms,
link_source_by='source',
link_target_by='target',
point_id_by='lemma',
point_label_by='word',
# point_x_by='umap_x',
# point_y_by='umap_y',
point_color_by='pos',
point_size_by='frequency',
point_size_scale=0.01, # often have to play with this number to get the size right
space_size=8112
)
h
Zooming in a bit:
Try out the Cosmograph widget in Google Colab with these example notebooks:
- Cosmograph Widget (Colab notebook) ✌️
- Mobius in Cosmograph Widget (Colab notebook)🎗️
- Clusters in Cosmograph (Colab notebook) 🫧
- English Words 🔤
Submit issues to https://github.com/cosmograph-org/py_cosmograph/issues.
🌎 Website