Action Spaces

Contains action space definitions

Classes:

ActionSpace(shape[, buffer_shape])	Abstract ActionSpace class.
ContinuousActionSpace(shape[, low, high, ...])	Action space that takes floating point inputs.
DiscreteActionSpace(shape, low, high[, ...])	Action space that takes integer inputs.

class holoocean.spaces.ActionSpace(shape, buffer_shape=None)

Abstract ActionSpace class.

Parameters:

• shape (list of int) – The shape of data that should be input to step or tick.

• buffer_shape (list of int, optional) –

  The shape of the data that will be written to the shared memory.

  Only use this when it is different from shape.

Methods:

get_high()	The maximum value(s) for the action space.
get_low()	The minimum value(s) for the action space.
sample()	Sample from the action space.

Attributes:

shape

Get the shape of the action space.

get_high()

The maximum value(s) for the action space.

Returns:

the action space’s maximum value(s)

Return type:

(list of float or float)

get_low()

The minimum value(s) for the action space.

Returns:

the action space’s minimum value(s)

Return type:

(list of float or float)

sample()

Sample from the action space.

Returns:

A valid command to be input to step or tick.

Return type:

(np.ndarray)

property shape

Get the shape of the action space.

Returns:

The shape of the action space.

Return type:

(list of int)

class holoocean.spaces.ContinuousActionSpace(shape, low=None, high=None, sample_fn=None, buffer_shape=None)

Action space that takes floating point inputs.

Parameters:

• shape (list of int) – The shape of data that should be input to step or tick.

• sample_fn (function, optional) – A function that takes a shape parameter and outputs a sampled command.

• low (list of float or float) – the low value(s) for the action space. Can be a scalar or an array

• high (list of float or float) –

  the high value(s) for the action space. Cand be a scalar or an array

  If this is not given, it will default to sampling from a unit gaussian.

• buffer_shape (list of int, optional) –

  The shape of the data that will be written to the shared memory.

  Only use this when it is different from shape.

Methods:

get_high()	The maximum value(s) for the action space.
get_low()	The minimum value(s) for the action space.
sample()	Sample from the action space.

get_high()

The maximum value(s) for the action space.

Returns:

the action space’s maximum value(s)

Return type:

(list of float or float)

get_low()

The minimum value(s) for the action space.

Returns:

the action space’s minimum value(s)

Return type:

(list of float or float)

sample()

Sample from the action space.

Returns:

A valid command to be input to step or tick.

Return type:

(np.ndarray)

class holoocean.spaces.DiscreteActionSpace(shape, low, high, buffer_shape=None)

Action space that takes integer inputs.

Parameters:

• shape (list of int) – The shape of data that should be input to step or tick.

• low (int) – The lowest value to sample.

• high (int) – The highest value to sample.

• buffer_shape (list of int, optional) –

  The shape of the data that will be written to the shared memory.

  Only use this when it is different from shape.

Methods:

get_high()	The maximum value(s) for the action space.
get_low()	The minimum value(s) for the action space.
sample()	Sample from the action space.

get_high()

The maximum value(s) for the action space.

Returns:

the action space’s maximum value(s)

Return type:

(list of float or float)

get_low()

The minimum value(s) for the action space.

Returns:

the action space’s minimum value(s)

Return type:

(list of float or float)

sample()

Sample from the action space.

Returns:

A valid command to be input to step or tick.

Return type:

(np.ndarray)