fixes #100 Transform synchronization

[wyli]

Fixes #100.

Description

• upgrades UniformRandomPatch to use randomizable interface
• adds UniformRandomPatchd dictionary-based interface
• upgrades Compose to use randomizable interface
• fixes spelling randomise -> randomize

transform sync is done by setting random augmentation's internal random states
alternatively it could be done by the dictionary-based transform

Status

Ready

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New tests added to cover the changes
• Docstrings/Documentation updated

[atbenmurray]

During discussions that we had about the use of RandomState, we had a signature more like the following:

class ATransform:
    def __init__(self, arg1, arg2, seed=12345678, rng=None):
        self.rng = RandomState(seed) if rng is None else rng
        ...

This allows a user to set a seed or RandomState object, which might be useful when picking up training from a partially trained, deserialised model

[wyli]

During discussions that we had about the use of RandomState, we had a signature more like the following:

class ATransform:
    def __init__(self, arg1, arg2, seed=12345678, rng=None):
        self.rng = RandomState(seed) if rng is None else rng
        ...

This allows a user to set a seed or RandomState object, which might be useful when picking up training from a partially trained, deserialised model

yes that is done by set_random_state(seed=, rng=) of each randomisable transform and merged in a previous PR.

This PR wants to add composed.set_random_state(seed=, rng=) to address ticket 100, where composed is:

composed = monai.transforms.compose.Compose([
        RandTransform_1(), RandTransform_2(), ...
])

[Nic-Ma]

Looks good to me now.