Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add cheatsheet for other frameworks #22

Merged
merged 2 commits into from
Jan 28, 2022
Merged

Add cheatsheet for other frameworks #22

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d96c73d
Add initial cheatsheet
darsnack Jan 15, 2022
aa7506f
Update for cosine annealing
darsnack Jan 28, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 46 additions & 0 deletions docs/cheatsheet.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
# Schedule cheatsheet for other frameworks

If you are coming from PyTorch or Tensorflow, the following table should help you find the corresponding schedule policy in ParameterSchedulers.jl.

!!! note
PyTorch typically wraps an optimizer as the first argument, but we ignore that functionality in the table. To wrap a Flux.jl optimizer with a schedule from the rightmost column, use [`ParameterSchedules.Scheduler`](#).
The variable `lr` in the middle/rightmost column refers to the initial learning rate of the optimizer.

| PyTorch | Tensorflow | ParameterSchedulers.jl |
|:-------------------------------------------------------------------------------|:------------------------------------------------------------------|:------------------------------------------------------|
| `LambdaLR(_, lr_lambda)` | N/A | `lr_lambda` |
| `MultiplicativeLR(_, lr_lambda)` | N/A | N/A |
| `StepLR(_, step_size, gamma)` | `ExponentialDecay(lr, step_size, gamma, True)` | `Step(lr, gamma, step_size)` |
| `MultiStepLR(_, milestones, gamma)` | N/A | `Step(lr, gamma, milestones)` |
| `ConstantLR(_, factor, total_iters)` | N/A | `Sequence(lr * factor => total_iters, lr => nepochs)` |
| `LinearLR(_, start_factor, end_factor, total_iters)` | N/A | `Sequence(Triangle(lr * start_factor, lr * end_factor, 2 * total_iters) => total_iters, lr => nepochs)` |
| `ExponentialLR(_, gamma)` | `ExponentialDecay(lr, 1, gamma, False)` | `Exp(lr, gamma)` |
| N/A | `ExponentialDecay(lr, steps, gamma, False)` | `Interpolator(Exp(lr, gamma), steps)` |
| `CosineAnnealingLR(_, T_max, eta_min)` | `CosineDecay(lr, T_max, eta_min)` | `CosAnneal(lr, eta_min, T_0, false)` |
| `CosineAnnealingRestarts(_, T_0, 1, eta_min)` | `CosineDecayRestarts(lr, first_decay_steps, t_mul, 1, eta_min)` | `CosAnneal(lr, eta_min, T_0)` |
| N/A | `CosineDecayRestarts(lr, first_decay_steps, t_mul, m_mul, alpha)` | N/A |
| `SequentialLR(_, schedulers, milestones)` | N/A | `Sequence(schedulers, milestones)` |
| `ReduceLROnPlateau(_, mode, factor, patience, threshold, 'abs', 0)` | N/A | See [below](# "`ReduceLROnPlateau` style schedules") |
| `CyclicLR(_, base_lr, max_lr, step_size, step_size, 'triangular', _, None)` | N/A | `Triangule(base_lr, max_lr, step_size)` |
| `CyclicLR(_, base_lr, max_lr, step_size, step_size, 'triangular2', _, None)` | N/A | `TrianguleDecay2(base_lr, max_lr, step_size)` |
| `CyclicLR(_, base_lr, max_lr, step_size, step_size, 'exp_range', gamma, None)` | N/A | `TrianguleExp(base_lr, max_lr, step_size, gamma)` |
| `CyclicLR(_, base_lr, max_lr, step_size, step_size, _, _, scale_fn)` | N/A | See [Arbitrary looping schedules](#) |
| N/A | `InverseTimeDecay(lr, 1, decay_rate, False)` | `Inv(lr, decay_rate, 1)` |
| N/A | `InverseTimeDecay(lr, decay_step, decay_rate, False)` | `Interpolator(Inv(lr, decay_rate, 1), decay_step)` |
| N/A | `PolynomialDecay(lr, decay_steps, 0, power, False)` | `Poly(lr, power, decay_steps)` |

## `ReduceLROnPlateau` style schedules

Unlike PyTorch, ParameterSchedulers.jl doesn't create a monolithic schedule to control dynamic schedules. Instead, [`ParameterSchedulers.Stateful`](#) has an `advance` keyword argument that can allow for arbitrary advancement of schedules based on a predicate function. When combined with `Flux.plateau` as the predicate, we get `ReduceLROnPlateau`.
```julia
# the code below is written to match
# ReduceLROnPlateau(_, 'max', factor, patience, threshold, 'abs', 0)
# we also assume accuracy_func() is an accuracy metric that's already given for our model

# this is done to match ReduceLROnPlateau
# but it could be any schedule
s = Exp(lr, factor)
predicate = Flux.plateau(accuracy_func, patience; min_dist = threshold)
ParameterSchedulers.Stateful(s; advance = predicate)
```
Using this approach, we can be more flexible than PyTorch. You can use any schedule (not just exponential decay) and arbitrary predicates. Make sure to check out the [Flux docmentation on "patience helpers"](https://fluxml.ai/Flux.jl/stable/utilities/#Patience-Helpers) for more ways to customize the predicate (e.g. the `'min'` mode for `ReduceLROnPlateau`).
2 changes: 2 additions & 0 deletions toc.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
[Introduction](README.md)

[Schedule cheatsheet](docs/cheatsheet.md)

# Tutorials

* [Getting started](docs/tutorials/getting-started.md)
Expand Down