def __init__(self, conf, data=None): assert conf.TIME_UNIT_BASE_SIGNALS_NUMBER <= conf.UNIT_INPUT_HEIGHT self.conf = conf if data: self.unit = Unit(conf, data=data['unit']) else: self.unit = Unit(conf) self._buffer = Buffer([], size=self.conf.UNIT_INPUT_HEIGHT)
class TimeUnit: """ This type of unit accepts only one signal per activation, but can predict next signals """ KEY = 'time_unit' def __init__(self, conf, data=None): assert conf.TIME_UNIT_BASE_SIGNALS_NUMBER <= conf.UNIT_INPUT_HEIGHT self.conf = conf if data: self.unit = Unit(conf, data=data['unit']) else: self.unit = Unit(conf) self._buffer = Buffer([], size=self.conf.UNIT_INPUT_HEIGHT) def get_data(self): return { '_key': self.KEY, 'unit': self.unit.get_data(), } def activate(self, signal, learn=True): """ Put signal to the unit :param signal: signal to activate :param learn: is learning enabled :return: output signal of this unit """ self._buffer.push(signal) learn = learn and all(self._buffer) return self.unit.activate(tuple(self._buffer), learn=learn) def reset(self): """ Forget all last signals (but not patterns). While unit has not enough signals, prediction is not be possible """ self._buffer = Buffer([], size=self.conf.UNIT_INPUT_HEIGHT) def get_prediction(self): """ :return: predicted signals """ signals_to_predict = self.conf.UNIT_INPUT_HEIGHT - self.conf.TIME_UNIT_BASE_SIGNALS_NUMBER signals = list(self._buffer)[signals_to_predict:] + [None] * signals_to_predict restored_signals = self.unit.restore(signals) predicted_signals = restored_signals[-signals_to_predict:] return predicted_signals
def reset(self): """ Forget all last signals (but not patterns). While unit has not enough signals, prediction is not be possible """ self._buffer = Buffer([], size=self.conf.UNIT_INPUT_HEIGHT)