Visualising How your model thinks using TSNE

Deep Learning
1

I was working on a way to display my college archive photos using Machine learning. So i decided to use Deep learning and Visualise the embeddings using TSNE.

ezgif-5-33f4b49517

I found out that tensor board is quite hard to work with , so I just wrote a simple project to use their standalone embedding projector without having to use tensorflow.

Also, you can share your visualisations without having a sever !!

Project Link is https://github.com/harveyslash/TSNE-Embedding-Visualisation
Visualisation link is https://harveyslash.github.io/TSNE-Embedding-Visualisation/

This is a WIP , your feedback, and issues are appreciated!

14 Likes
2

This is cool. What did you use to make that gif?

3

Thanks!

Just a screen recording tool (Quicktime , since i use mac). Then converted using some online converter

4

This is super cool @harveyslash. Very useful. Thanks for sharing.

5

Is it possible to import from pytorch models?

6

Hi,

Could you show an example of what the preds are? I know it is a numpy array (from Keras Docs) but I am just not sure how it is composed. Iā€™d like to know in order to use the model I trained with fastai.

When we use predict() in fastai, it returns a tensor with the probabilities for all classes, but I am not really sure what is the return in Keras.