Dev Sprint July18

Visualisations

There are myriad resources available on the web for visualising various machine learning algorithms. We are looking for some library which can be used for our own customised algorithms. Here are a few awesome examples: -

• The best visualisation I came across are Victor Powell's explained visually.
• For neural network Google have a nice implementation on their tensorflow playground.

Visualisation Frameworks -

• Like PCA (principle component analysis) we have a dimensionality reduction technique - t-SNE(t-Distributed Stochastic Neighbor Embedding) for the visualization of high-dimensional datasets.
• For plotting we want to use interactive libraries.
  • Matplotlib - We are using matplotlib for our visualisation already. Though any suggestions are welcomed.
  • Bokeh - This is the library I am currently trying to adapt as it does a pretty good job in making interactive plots.

Cherries

The cake being our repository, cherries are the toppings we need to make user feel that our repository shows genuine results and is trustworthy.

• LIME - Explaining the predictions of any machine learning classifier. We want to know which feature our algorithm preferred the most to come to a certain conclusion. For example - Suppose we want our algorithm to judge whether a wolf is in the input image or not. Now consider that most of our input images have wolf in a snowy environment. If our image classifier is picking the snowy background as a feature every time. No matter how good our accuracy is, our model is not performing well. You can watch their official video on how it works here