Schedules

Schedules are generators that provide different rates, schedules, decays or series. They’re typically used for batch sizes or learning rates. You can easily implement your own schedules as well: just write your own generator function, that produces whatever series of values you need. A common use case for schedules is within Optimizer objects, which accept iterators for most of their parameters. See the training guide for details.

constant function

Yield a constant rate.

from thinc.api import constant

batch_sizes = constant(0.001)
batch_size = next(batch_sizes)

config[batch_size]
@schedules = "constant.v1"
rate = 0.001

Argument	Type
`rate`	`float`
YIELDS	`float`

constant_then function

Yield a constant rate for N steps, before starting a schedule.

from thinc.api import constant_then, decaying

learn_rates = constant_then(
    0.005,
    1000,
    decaying(0.005, 1e-4)
)
learn_rate = next(learn_rates)

config[learn_rates]
@schedules = "constant_then.v1"
rate = 0.005
steps = 1000

[learn_rates.schedule]
@schedules = "decaying"
base_rate = 0.005
decay = 1e-4

Argument	Type
`rate`	`float`
`steps`	`int`
`schedule`	`Iterable[float]`
YIELDS	`float`

decaying function

Yield an infinite series of linearly decaying values, following the schedule base_rate * 1 / (1 + decay * t).

from thinc.api import decaying

learn_rates = decaying(0.005, 1e-4)
learn_rate = next(learn_rates)  # 0.001
learn_rate = next(learn_rates)  # 0.00999

config[learn_rate]
@schedules = "decaying.v1"
base_rate = 0.005
decay = 1e-4
t = 0

Argument	Type
`base_rate`	`float`
`decay`	`float`
keyword-only
`t`	`int`
YIELDS	`float`

compounding function

Yield an infinite series of compounding values. Each time the generator is called, a value is produced by multiplying the previous value by the compound rate.

from thinc.api import compounding

batch_sizes = compounding(1.0, 32.0, 1.001)
batch_size = next(batch_sizes)  # 1.0
batch_size = next(batch_sizes)  # 1.0 * 1.001

config[batch_size]
@schedules = "compounding.v1"
start = 1.0
stop = 32.0
compound = 1.001
t = 0

Argument	Type
`start`	`float`
`stop`	`float`
`compound`	`float`
keyword-only
`t`	`int`
YIELDS	`float`

warmup_linear function

Generate a series, starting from an initial rate, and then with a warmup period, and then a linear decline. Used for learning rates.

from thinc.api import warmup_linear

learn_rates = warmup_linear(0.01, 3000, 6000)
learn_rate = next(learn_rates)

config[learn_rate]
@schedules = "warmup_linear.v1"
initial_rate = 0.01
warmup_steps = 3000
total_steps = 6000

Argument	Type
`initial_rate`	`float`
`warmup_steps`	`int`
`total_steps`	`int`
YIELDS	`float`

slanted_triangular function

Yield an infinite series of values according to Howard and Ruder’s (2018) “slanted triangular learning rate” schedule.

from thinc.api import slanted_triangular

learn_rates = slanted_triangular(0.1, 5000)
learn_rate = next(learn_rates)

config[learn_rate]
@schedules = "slanted_triangular.v1"
max_rate = 0.1
num_steps = 5000
cut_frac = 0.1
ratio = 32
decay = 1.0
t = 0.1

Argument	Type
`max_rate`	`float`
`num_steps`	`int`
keyword-only
`cut_frac`	`float`
`ratio`	`int`
`decay`	`float`
`t`	`float`
YIELDS	`float`

cyclic_triangular function

Linearly increasing then linearly decreasing the rate at each cycle.

from thinc.api import cyclic_triangular

learn_rates = cyclic_triangular(0.005, 0.001, 1000)
learn_rate = next(learn_rates)

config[learn_rate]
@schedules = "cyclic_triangular.v1"
min_lr = 0.005
max_lr = 0.001
period = 1000

Argument	Type
`min_lr`	`float`
`max_lr`	`float`
`period`	`int`
YIELDS	`float`