[WIP] Lesson 3: Overview of building blocks (tensors) of TensorFlow main methods, functions, and datasets

[lillythomas]

Relates to https://github.com/developmentseed/servir-amazonia-2-internal/issues/5

In this lesson, we go through:

• some basics on tensors, how to use them
• tf.function and when to use
• tf.data and how to create datasets from raw data

[gitnotebooks[bot]]

Found 4 changed notebooks. Review the changes at https://gitnotebooks.com/developmentseed/tensorflow-eo-training-2/pull/6

[rbavery]

cna > can

Tip: open the notebook in vscode, it should highlight spelling issues.

Respond and view the context here.

[rbavery]

Nice to have, an answer to the question above.

variable: loss value, changes with each epoch constant: fixed hyper parameters like batch size

Respond and view the context here.

[rbavery]

A "vector" is a "rank-1" tensor and contains something like a list of values > In ML, a "vector" is a "rank-1" tensor and contains a list of values

In mathematics vectors can have more than 1 dimension, but when it comes to ML, vectors typically refer to single dimensional arrays. I think we can omit the last sentence "A vector is characterized by having one axis." since this doesn't hold depending on who you're talking to.

Respond and view the context here.

[lillythomas]

Word. I actually provide an answer a bit later on, discussing how trainable weights should be variables. But I like these suggestions too.

Respond and view the context here.

[rbavery]

overall not a big deal though

Respond and view the context here.

[rbavery]

can delete "and mostly out of scope for now", implied

Respond and view the context here.

[rbavery]

definition of rank happens after the term is introduced, can you shift where it is introduced below this or move these definitions up?

Respond and view the context here.

[rbavery]

I think we can leave out "It's also used in the making of data pipelines."

We don't introduce the concept of functional data pipes in this lesson so this might confuse learners

Respond and view the context here.

[rbavery]

can also remove data pipes here too. unless we introduce TF's concept of it with code.

Respond and view the context here.

[rbavery]

ok here I think is where we map tf.data.dataset to data pipes, should be data pipes for consistency.

can this introduction be moved up? and I think in that case, good to expand and define data pipes and include the references to datapipes.

Respond and view the context here.

[lillythomas]

Reviewed and revised per all of the suggested changes. Merging now.