This repository was archived by the owner on Apr 23, 2025. It is now read-only.
Use #file to get Swift script location.
#20
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Using `CommandLine.arguments[0]` to get the Swift script location is unrobust. `arguments[0]` differs when running `swift` directly vs. invoking `swiftc` then running the compiled binary.
pschuh
pushed a commit
to pschuh/swift-models
that referenced
this pull request
Jul 30, 2019
Many deep learning models are composed of sequential layers stacked one on
top of each other. It can be relatively tedious to write out the explicit
`applied(to:)` function because it's fairly repetitive and the underlying
intent is relatively obscured. (It can be especially bothersome because
it's the 2nd (or 3rd) time you're writing out all the layers. (The first time
is to declare all the instance variables, and the second time (if necessary)
is in the initializer.)
Fortunately, with helper functions, we can make everything both type
safe as well as convenient and easily expressible & readable!
This commit adds a family of `sequenced(in:through:)` functions that take
in a context, an input, and a variable number of layers. It chains through the
output of one layer into the input of the next.
This API approach has a number of advantages:
1. It avoids introducing new symbolic operators, which can be very confusing
to new users.
2. It works with today's AutoDiff implementation. (Yay!)
3. It is very readable and clean.
4. It avoids users "getting stuck". Concretely, if someone implemented a model
using my previously proposed `>>>` operator, if they wanted to add a
residual (or skip) connection, they would have to basically re-write their
whole model using a struct, etc. With this API structure, only "local"
changes are required. (e.g. If only one skip-connection is required, they
can split the sequential chain into two pieces.)
Downsides of this approach:
1. It doesn't DRY-out the types required to define a model. (I have some
thoughts here, but there isn't enough room in this
margin^H^H^H^H^H^Hcommit message.)
2. We should think hard about how things should look when we have loops.
3. We should switch to gyb to generate the code for all the different arities.
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Using
CommandLine.arguments[0]to get the Swift script location isunrobust.
arguments[0]differs when runningswiftdirectly vs.invoking
swiftcthen running the compiled binary.