Transform word embeddings into an interactive 3D masterpiece! This project takes the pretrained word2vec-google-news-300 model, reduces its 300-dimensional vectors to 3D with PCA, and visualizes them using Plotly. Words like "king," "queen," "doctor," and "cat" dance in space with dynamic colors, direction vectors, and a semantic core—revealing the hidden structure of language in NLP.
- Interactive 3D Plot: Spin, zoom, and explore word vectors in a browser-based visualization.
- Color Coding: Markers shift from red to blue based on cosine similarity to "king."
- Direction Vectors: Dashed arrows trace each word’s path from the origin.
- Semantic Core: A red diamond marks the average vector position.
- Dual Themes: Choose between a dark cosmic look or a light blue modern vibe.
- Python 3.x
- Gensim: For loading Word2Vec.
- Scikit-learn: PCA dimensionality reduction.
- Plotly: Interactive 3D plotting.
- NumPy: Vector and array operations.
- Clone the Repo:
git clone [https://github.com/ChandraPrakash-Bathula/PCA-in-3D.git]
- Install Requirements:
pip install gensim scikit-learn plotly numpy glove model
- Run the Script:
- Multiple themes available
- Run either script to launch the visualization.
- The Word2Vec model downloads on first run (~1.6 GB, so grab a coffee).
- A browser window opens with the 3D plot—click, drag, and scroll to explore.
- Edit the
wordslist in the code to visualize your own terms.
- Dark Cosmic: Black background, white text, starry grid (
rgba(10, 10, 20, 0.9)). - Light Blue: Soft blue background, black text (
rgba(230, 240, 250, 0.9)).
- Swap PCA for t-SNE for funkier clusters.
- Add clustering to group related words.
- Integrate into an NLP dashboard.
Fork it, tweak it, PR it! Ideas welcome—let’s make this even wilder.
MIT License—use it, share it, just give a nod back.
- Built by a student researcher hooked on ML and NLP.
- Thanks to Gensim, Scikit-learn, and Plotly for the open-source juice.