Workshop outline

[ricardoV94]

Just discussing potential organization for the workshop sessions. We want to keep it brief and sweet.

Perhaps max 2h per session, half for exposition, half for exercises?

First session

• Overview of PyTensor Conceptual intro #12 One minute example #13
• PyTensor from scratch

Exercises:

• Implementing new Ops
• Implementing and working with new types (tuple)
• Intermediate application (PageRank)

Second session

• Graph transformations: Grad and shape Graph transformation: eager and lazy #29
• Graph analysis: Variable dependency or type inference Variable dependency analysis #30

Exercises

• Estimating number of flops in graph Infer number of flops #17
• Determinant of diagonal rewrite Determinant of diagonal rewrite #6
• Root finding with Scipy Use PyTensor for root finding with Scipy #4
• Build sparse jacobian Build sparse jacobian by evaluating system of equations dependency #5

Third session

• VAE example VAE #11
• Partialling / clone replace with PyTensor Partialling / clone replace application #27

Exercises

• Vectorization and Scalarization transformation Vectorization / scalarization exercise #26
• Advanced DL application Advanced DL exercise #25
• Advanced Statistical application Gibbs Sampler #24