janwithb/latent-space-plots

A simple Jupyter notebook to visualize data in latent space using dimensionality reduction techniques.

Latent Space Plots

A simple Jupyter notebook to visualize data in latent space using dimensionality reduction techniques. The notebook focuses on RL with image-based observations.

Data basis

Latent vectors and rewards: (n, k + 1)

Index	Feature_0	Feature_1	Feature_2	Feature_k	Reward
0	...	...	...	...	...
1	...	...	...	...	...
2	...	...	...	...	...
n	...	...	...	...	...

Corresponding images: (n, 3, 64, 64)

Dimensionality reduction options

• Principal Component Analysis (PCA)
• t-Distributed Stochastic Neighbor Embedding (TSNE)

Visualizations

2D plot

3D plot

3D animation

2D plot with images

3D plot with images

3D animation with images