def predict(self, guest_data):
LOGGER.info("@ start guest predict")
features, labels, instances_indexes = convert_instance_table_to_array(
guest_data)
guest_x = np.squeeze(features)
guest_y = np.expand_dims(labels, axis=1)
LOGGER.debug("guest_x, guest_y: " + str(guest_x.shape) + ", " +
str(guest_y.shape))
host_prob = self._do_get(
name=self.transfer_variable.host_prob.name,
tag=self.transfer_variable.generate_transferid(
self.transfer_variable.host_prob),
idx=-1)[0]
self.guest_model.set_batch(guest_x, guest_y)
pred_prob = self.guest_model.predict(host_prob)
LOGGER.debug("pred_prob: " + str(pred_prob.shape))
self._do_remote(pred_prob,
name=self.transfer_variable.pred_prob.name,
tag=self.transfer_variable.generate_transferid(
self.transfer_variable.pred_prob),
role=consts.HOST,
idx=-1)
return None
def predict(self, host_data, predict_param):
LOGGER.info("@ start host predict")
features, labels, instances_indexes = convert_instance_table_to_array(
host_data)
host_x = np.squeeze(features)
LOGGER.debug("host_x： " + str(host_x.shape))
host_prob = self.host_model.predict(host_x)
self._do_remote(host_prob,
name=self.transfer_variable.host_prob.name,
tag=self.transfer_variable.generate_transferid(
self.transfer_variable.host_prob),
role=consts.GUEST,
idx=-1)
pred_prob = self._do_get(
name=self.transfer_variable.pred_prob.name,
tag=self.transfer_variable.generate_transferid(
self.transfer_variable.pred_prob),
idx=-1)[0]
pred_prob = np.squeeze(pred_prob)
LOGGER.debug("pred_prob: " + str(pred_prob.shape))
pred_prob_table = create_table(pred_prob, instances_indexes)
actual_label_table = create_table(labels, instances_indexes)
pred_label_table = self.classified(pred_prob_table,
predict_param.threshold)
if predict_param.with_proba:
predict_result = actual_label_table.join(
pred_prob_table, lambda label, prob:
(label if label > 0 else 0, prob))
predict_result = predict_result.join(pred_label_table, lambda x, y:
(x[0], x[1], y))
else:
predict_result = actual_label_table.join(
pred_label_table, lambda a_label, p_label:
(a_label, None, p_label))
return predict_result
