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

1. Overview of PyTensor Conceptual intro #12 One minute example #13
2. PyTensor from scratch

Exercises:

1. Implementing new Ops
2. Implementing and working with new types (tuple)
3. Intermediate application of PyTensor Intermediate difficulty application of PyTensor #18
4. Reimplementing dprint with rich Reimplement dprint with rich #10

Second session

1. Graph transformations: Grad and shape
2. Graph analysis: Variable dependency and type analysis

Exercises

1. Vectorization and Scalarization transformation
2. Estimating number of flops in graph Infer number of flops #17
3. Root finding with Scipy Use PyTensor for root finding with Scipy #4
4. Type analysis and Determinant of diagonal rewrite
5. Build sparse jacobian

Third session

1. Partialling / clone replace with PyTensor
2. VAE example VAE #11

Exercises

1. New sparse operation and rewrite Implement new Sparse Op #9
2. Expand/factor/collect directed rewrites Expand, factor, collect directed rewrites #15
3. Advanced DL application (something where you disable BatchNorm after learning)?
4. Advanced Statistical application Gibbs Sampler #24