[BUG] FedPAC logic error: IndexError: list index out of range

[JY132]

I got the IndexError: list index out of range on FedPAC when training on 30 clients with join rate 0.5 on some dataset with 0.1 Dirichlet parameter.

Regarding these 2 functions in FedPAC server

    def receive_models(self):
        assert (len(self.selected_clients) > 0)

        active_clients = random.sample(
            self.selected_clients, int((1-self.client_drop_rate) * self.current_num_join_clients))

        self.uploaded_ids = []
        self.uploaded_weights = []
        self.uploaded_models = []
        self.uploaded_heads = []
        tot_samples = 0
        for client in active_clients:
            try:
                client_time_cost = client.train_time_cost['total_cost'] / client.train_time_cost['num_rounds'] + \
                        client.send_time_cost['total_cost'] / client.send_time_cost['num_rounds']
            except ZeroDivisionError:
                client_time_cost = 0
            if client_time_cost <= self.time_threthold:
                tot_samples += client.train_samples
                self.uploaded_ids.append(client.id)
                self.uploaded_weights.append(client.train_samples)
                self.uploaded_models.append(client.model.base)
                self.uploaded_heads.append(client.model.head)
        for i, w in enumerate(self.uploaded_weights):
            self.uploaded_weights[i] = w / tot_samples

    def aggregate_and_send_heads(self):
        head_weights = solve_quadratic(len(self.uploaded_ids), self.Vars, self.Hs)

        for idx, cid in enumerate(self.uploaded_ids):
            print('(Client {}) Weights of Classifier Head'.format(cid))
            print(head_weights[idx],'\n')

            if head_weights[idx] is not None:
                new_head = self.add_heads(head_weights[idx])
            else:
                new_head = self.uploaded_heads[cid]

            self.clients[cid].set_head(new_head)

if the head_weights[idx] is None, it is impossible to obtain the new_head using self.uploaded_heads[cid] (In my case, with join rate 0.5)

Here, cid is a client ID, not the position in the list.
That means you're treating uploaded_heads like a dictionary, but it's actually a list?