Method get_action

examples/reinforcement_learning/tutorial_C51.py:214–222

(self, obv)

            self.optimizer = tf.optimizers.Adam(learning_rate=lr)

    def get_action(self, obv):
        eps = epsilon(self.niter)
        if args.train and random.random() < eps:
            return int(random.random() * out_dim)
        else:
            obv = np.expand_dims(obv, 0).astype('float32') * ob_scale
            qdist = np.exp(self._qvalues_func(obv).numpy())
            qvalues = (qdist * vrange).sum(-1)
            return qvalues.argmax(1)[0]

    @tf.function
    def _qvalues_func(self, obv):