BEAT Tutorial
Rough schedule for 3 hour tutorial:
- Introducing BEAT: ~30min
- Ecosystem walkthrough using the Iris dataset: 1hr 15min
- Explaining the experiment and breaking it down conceptually
- Experiment design
- Introduce the Iris dataset
- Introduce the problem (discriminating across the three species using just measurement samples)
- Propose the used method: LDA
- Breaking down this problem/solution into necessary steps
- Getting the data
- Preprocessing it
- Extracting features
- Training the machine
- Scoring
- Displaying results
- Understanding how/why to split into BEAT blocks
- The MapReduce concept -> blocks & connections
- Synchronization & the "every block is a for-loop" thing
- On-paper toolchain (?)
- Experiment design
- Databases
- Quickly introducing Bob
- Creating a database
- Dataformats
- Intro to pulling stuff from platform
- Creating one/some manually
- Algorithms
- What are "algorithms" - analyzer vs normal
- Anatomy of an algorithm file (functions)
- More Bob intro
- Analyzer explanation & create analyzer
- Toolchains
- Using algorithms to create toolchain blocks
- Copying & cool editor stuff
- Discussing toolchain connections & types
- Putting it all together: Experiments
- Explaining how experiments relate to other things
- Creating an experiment and assigning datasets/blocks/parameters
- Plots
- How do plots work? Plotters & plotterparameters
- Creation of plotters
- Creation of plotterparameters
- Generating plots for an experiment
- Pushing back to the platform (?)
- Explaining the experiment and breaking it down conceptually
- More complex & practical example using AT&T eigenfaces: 30min
- Conceptual breakdown: experiment design
- breaking down eigenface into necessary steps
- understanding how/why to split into BEAT blocks & highlighting reusability stuff
- on-paper toolchain
- toolchain
- Conceptual breakdown: experiment design
- Reusing toolchains & other components with DCT-GMM: 15min
- designing using the above stuff
- "wow look reusing stuff"
- Creating a new environment off of an existing one with
beat.env.python
: 30min- Intro to environments & how that all works
- docker executor
- showing where environments are
- compiling/making an environment
- changing the env & making a new one