rust: add data_compat module
#4340
Conversation
Summary: This module performs conversions from legacy on-disk data formats. It will correspond to both the `data_compat` and `dataclass_compat` modules from Python TensorBoard. For now, we have just one function, to determine the initial summary metadata of a time series. And, for now, we only implement this for scalars, to keep things simple. Test Plan: Unit tests included. wchargin-branch: rust-data-compat wchargin-source: ef510b422b00a66a23efc1406947844fa8d1a798
wchargin-branch: rust-data-compat wchargin-source: 467974fa4e0e55dea0872bb1394a768a38811ac5
| match (md, value) { | ||
| // Any summary metadata that sets its own data class is expected to already be in the right | ||
| // form. | ||
| (Some(md), _) if md.data_class != i32::from(pb::DataClass::Unknown) => Box::new(md), |
There was a problem hiding this comment.
Interesting, I did not know that the if <condition> could be part of the arm's matching pattern (on the left side of the =>)! The examples in the Rust reference / book are less advanced.
There was a problem hiding this comment.
Yep! This is called a match guard. It is occasionally quite useful
when you want to continue to match multiple kinds of other patterns when
the guard fails, as in this example.
| } | ||
| } | ||
|
|
||
| mod unknown { |
There was a problem hiding this comment.
I see, so these mods are used like describe(... blocks in Jasmine in JS.
There was a problem hiding this comment.
Yeah, I’m using them as such, since they show up in the output like this:
test data_compat::tests::scalars::test_tf1x_simple_value ... ok
test data_compat::tests::scalars::test_tf2x_scalar_tensor_without_dataclass ... ok
test data_compat::tests::unknown::test_custom_plugin_with_dataclass ... ok
test data_compat::tests::unknown::test_empty ... ok
test data_compat::tests::unknown::test_unknown_plugin_no_dataclass ... ok
I haven’t done this before and don’t have strong opinions about it, but
it seemed pretty reasonable, so I figured that I’d try it.
| match (md, value) { | ||
| // Any summary metadata that sets its own data class is expected to already be in the right | ||
| // form. | ||
| (Some(md), _) if md.data_class != i32::from(pb::DataClass::Unknown) => Box::new(md), |
There was a problem hiding this comment.
Yep! This is called a match guard. It is occasionally quite useful
when you want to continue to match multiple kinds of other patterns when
the guard fails, as in this example.
| } | ||
| } | ||
|
|
||
| mod unknown { |
There was a problem hiding this comment.
Yeah, I’m using them as such, since they show up in the output like this:
test data_compat::tests::scalars::test_tf1x_simple_value ... ok
test data_compat::tests::scalars::test_tf2x_scalar_tensor_without_dataclass ... ok
test data_compat::tests::unknown::test_custom_plugin_with_dataclass ... ok
test data_compat::tests::unknown::test_empty ... ok
test data_compat::tests::unknown::test_unknown_plugin_no_dataclass ... ok
I haven’t done this before and don’t have strong opinions about it, but
it seemed pretty reasonable, so I figured that I’d try it.
Summary:
This module performs conversions from legacy on-disk data formats. It
will correspond to both the
data_compatanddataclass_compatmodulesfrom Python TensorBoard. For now, we have just one function, to
determine the initial summary metadata of a time series. And, for now,
we only implement this for scalars, to keep things simple.
Test Plan:
Unit tests included.
wchargin-branch: rust-data-compat