def multi_task_loss(y, t): cross_entropy = categorical_crossentropy(y[:, :num_class], t) regress_predictions = discrete_predict(y[:, -1]) mse = squared_loss(regress_predictions, t) log_loss = cross_entropy.mean() reg_loss = mse.mean() return log_loss, reg_loss, log_loss + 3 * reg_loss
def multi_task_loss(y, t): softmax_predictions = categorical_crossentropy(y[:, :num_class], t) regress_predictions = squared_loss(y[:, -1], t) log_loss = softmax_predictions.mean() reg_loss = regress_predictions.mean() return log_loss, reg_loss, log_loss + 0.1 * reg_loss
def multi_task_loss(y, t): softmax_predictions = categorical_crossentropy(y[:, :num_class], t) regress_predictions = squared_loss(y[:, -1], t) log_loss = softmax_predictions.mean() reg_loss = regress_predictions.mean() return log_loss, reg_loss, 0.75*log_loss + 0.25*reg_loss