{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"import gc\nimport psutil\nimport random\nfrom tqdm.notebook import tqdm\n\nimport numpy as np\nimport pandas as pd\n\nfrom sklearn.metrics import roc_auc_score\n\nimport torch\nimport torch.nn as nn\nfrom torch.utils.data import Dataset, DataLoader","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# dtypes = {\n#     'row_id': 'int64',\n#     'timestamp': 'int64',\n#     'user_id': 'int32',\n#     'content_id': 'int16',\n#     'content_type_id': 'int8',\n#     'task_container_id': 'int16',\n#     'user_answer': 'int8',\n#     'answered_correctly':'int8',\n#     'prior_question_elapsed_time': 'float32',\n#     'prior_question_had_explanation': 'boolean'\n# }","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"MAX_SEQ = 50\nNUM_HEADS = 10\nNUM_EMBED = 128  # length of the embedding layer\nn_skill = 13523  # total number of questions","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"%%time\n\ndtypes = {'timestamp': 'int64', \n          'user_id': 'int32' ,\n          'content_id': 'int16',\n          'content_type_id': 'int8',\n          'answered_correctly':'int8'}\n\ntrain_df = pd.read_csv('/kaggle/input/riiid-test-answer-prediction/train.csv', usecols=[1, 2, 3, 4, 7], dtype=dtypes)\n                   ","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# max_timestamp_u = train[['user_id','timestamp']].groupby(['user_id']).max()\n# max_timestamp_u.head(20)\n# max_timestamp_u.columns = ['max_timestamp']\n# max_timestamp_u['interval'] = max_timestamp_u.max_timestamp.max() - max_timestamp_u.max_timestamp\n# # max_timestamp_u['random'] = np.random.rand(len(max_timestamp_u))\n# max_timestamp_u['random'] = np.random.beta(a, b, len(max_timestamp_u))\n# # max_timestamp_u['random_timestamp'] = max_timestamp_u.interval * max_timestamp_u.random\n# # max_timestamp_u['random_timestamp'] = max_timestamp_u.random_timestamp.astype(int)\n# # max_timestamp_u.drop(['interval', 'random'], axis=1, inplace=True)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_df = train_df[train_df.content_type_id == False]\ntrain_df = train_df.sort_values(['timestamp'], ascending=True).reset_index(drop=True)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_df.info()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_df.head()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"group = train_df[['user_id', 'content_id', 'answered_correctly']].groupby('user_id').apply(lambda x: (\n            x['content_id'].values,\n            x['answered_correctly'].values))\n\ndel train_df\ngc.collect()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"random.seed(1)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class SAKTDataset(Dataset):\n    def __init__(self, group, n_skill, max_seq=MAX_SEQ): #HDKIM 100\n        super(SAKTDataset, self).__init__()\n        self.max_seq = max_seq\n        self.n_skill = n_skill\n        self.samples = group\n        \n#         self.user_ids = [x for x in group.index]\n        self.user_ids = []\n        for user_id in group.index:\n            q, qa = group[user_id]\n            if len(q) < 2:\n                continue\n            self.user_ids.append(user_id)\n            \n            #HDKIM Memory reduction\n            #if len(q)>self.max_seq:\n            #    group[user_id] = (q[-self.max_seq:],qa[-self.max_seq:])\n\n    def __len__(self):\n        return len(self.user_ids)\n\n    def __getitem__(self, index):\n        user_id = self.user_ids[index]\n        q_, qa_ = self.samples[user_id]\n        seq_len = len(q_)\n\n        q = np.zeros(self.max_seq, dtype=int)\n        qa = np.zeros(self.max_seq, dtype=int)\n        \n        if seq_len >= self.max_seq:\n            if random.random()>0.1:\n                # 90% chance picking a random length of samples from the seq>max_seq\n                start = random.randint(0,(seq_len-self.max_seq))\n                end = start + self.max_seq\n                q[:] = q_[start:end]\n                qa[:] = qa_[start:end]\n            else:\n                # 10% change pick the last max_seq length of samples from the seq\n                q[:] = q_[-self.max_seq:]\n                qa[:] = qa_[-self.max_seq:]\n        else:\n            if random.random()>0.1:\n                # 90% chance picking a random length of samples from the seq>max_seq\n                start = 0\n                end = random.randint(2,seq_len)\n                seq_len = end - start\n                q[-seq_len:] = q_[0:seq_len]\n                qa[-seq_len:] = qa_[0:seq_len]\n            else:\n                # 10% change take all the samples of the seq\n                q[-seq_len:] = q_\n                qa[-seq_len:] = qa_\n\n        \n        target_id = q[1:]\n        label = qa[1:]\n\n        x = np.zeros(self.max_seq-1, dtype=int)\n        x = q[:-1].copy()\n        x += (qa[:-1] == 1) * self.n_skill\n\n        return x, target_id, label","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"dataset = SAKTDataset(group, n_skill)\ndataloader = DataLoader(dataset, batch_size=2048, shuffle=True, num_workers=8)\n\nitem = dataset.__getitem__(5)\n# print(item[1])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class FFN(nn.Module):\n    def __init__(self, state_size=200):\n        super(FFN, self).__init__()\n        self.state_size = state_size\n\n        self.lr1 = nn.Linear(state_size, state_size)\n        self.relu = nn.ReLU()\n        self.lr2 = nn.Linear(state_size, state_size)\n        self.dropout = nn.Dropout(0.2)\n    \n    def forward(self, x):\n        x = self.lr1(x)\n        x = self.relu(x)\n        x = self.lr2(x)\n        return self.dropout(x)\n\ndef future_mask(seq_length):\n    future_mask = np.triu(np.ones((seq_length, seq_length)), k=1).astype('bool')\n    return torch.from_numpy(future_mask)\n\n\nclass SAKTModel(nn.Module):\n    def __init__(self, n_skill, max_seq=MAX_SEQ, embed_dim=128): \n        super(SAKTModel, self).__init__()\n        self.n_skill = n_skill\n        self.embed_dim = embed_dim\n\n        self.embedding = nn.Embedding(2*n_skill+1, embed_dim)\n        self.pos_embedding = nn.Embedding(max_seq-1, embed_dim)\n        self.e_embedding = nn.Embedding(n_skill+1, embed_dim)\n\n        self.multi_att = nn.MultiheadAttention(embed_dim=embed_dim, num_heads=8, dropout=0.2)\n\n        self.dropout = nn.Dropout(0.2)\n        self.layer_normal = nn.LayerNorm(embed_dim) \n\n        self.ffn = FFN(embed_dim)\n        self.pred = nn.Linear(embed_dim, 1)\n    \n    def forward(self, x, question_ids):\n        device = x.device        \n        x = self.embedding(x)\n        pos_id = torch.arange(x.size(1)).unsqueeze(0).to(device)\n\n        pos_x = self.pos_embedding(pos_id)\n        x = x + pos_x\n\n        e = self.e_embedding(question_ids)\n\n        x = x.permute(1, 0, 2) # x: [bs, s_len, embed] => [s_len, bs, embed]\n        e = e.permute(1, 0, 2)\n        att_mask = future_mask(x.size(0)).to(device)\n        att_output, att_weight = self.multi_att(e, x, x, attn_mask=att_mask)\n        att_output = self.layer_normal(att_output + e)\n        att_output = att_output.permute(1, 0, 2) # att_output: [s_len, bs, embed] => [bs, s_len, embed]\n\n        x = self.ffn(att_output)\n        x = self.layer_normal(x + att_output)\n        x = self.pred(x)\n\n        return x.squeeze(-1), att_weight","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n\nmodel = SAKTModel(n_skill, embed_dim=128)\n# optimizer = torch.optim.SGD(model.parameters(), lr=1e-3, momentum=0.99, weight_decay=0.005)\noptimizer = torch.optim.Adam(model.parameters(), lr=1e-3)\ncriterion = nn.BCEWithLogitsLoss()\n\nmodel.to(device)\ncriterion.to(device)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def train_epoch(model, train_iterator, optim, criterion, device=\"cpu\"):\n    model.train()\n\n    train_loss = []\n    num_corrects = 0\n    num_total = 0\n    labels = []\n    outs = []\n\n    tbar = tqdm(train_iterator)\n    for item in tbar:\n        x = item[0].to(device).long()\n        target_id = item[1].to(device).long()\n        label = item[2].to(device).float()\n\n        optim.zero_grad()\n        output, atten_weight = model(x, target_id)\n        loss = criterion(output, label)\n        loss.backward()\n        optim.step()\n        train_loss.append(loss.item())\n\n        output = output[:, -1]\n        label = label[:, -1] \n        pred = (torch.sigmoid(output) >= 0.5).long()\n        \n        num_corrects += (pred == label).sum().item()\n        num_total += len(label)\n\n        labels.extend(label.view(-1).data.cpu().numpy())\n        outs.extend(output.view(-1).data.cpu().numpy())\n\n        tbar.set_description('loss - {:.4f}'.format(loss))\n\n    acc = num_corrects / num_total\n    auc = roc_auc_score(labels, outs)\n    loss = np.mean(train_loss)\n\n    return loss, acc, auc","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"epochs = 35\nfor epoch in range(epochs):\n    loss, acc, auc = train_epoch(model, dataloader, optimizer, criterion, device)\n    print(\"epoch - {} train_loss - {:.2f} acc - {:.3f} auc - {:.3f}\".format(epoch, loss, acc, auc))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"torch.save(model.state_dict(), \"SAKT.pt\")\ndel dataset\ngc.collect()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"class TestDataset(Dataset):\n    def __init__(self, samples, test_df, n_skill, max_seq=MAX_SEQ):\n        super(TestDataset, self).__init__()\n        self.samples = samples\n        self.user_ids = [x for x in test_df[\"user_id\"].unique()]\n        self.test_df = test_df\n        self.n_skill = n_skill\n        self.max_seq = max_seq\n\n    def __len__(self):\n        return self.test_df.shape[0]\n\n    def __getitem__(self, index):\n        test_info = self.test_df.iloc[index]\n\n        user_id = test_info[\"user_id\"]\n        target_id = test_info[\"content_id\"]\n\n        q = np.zeros(self.max_seq, dtype=int)\n        qa = np.zeros(self.max_seq, dtype=int)\n\n        if user_id in self.samples.index:\n            q_, qa_ = self.samples[user_id]\n            \n            seq_len = len(q_)\n\n            if seq_len >= self.max_seq:\n                q = q_[-self.max_seq:]\n                qa = qa_[-self.max_seq:]\n            else:\n                q[-seq_len:] = q_\n                qa[-seq_len:] = qa_          \n        \n        x = np.zeros(self.max_seq-1, dtype=int)\n        x = q[1:].copy()\n        x += (qa[1:] == 1) * self.n_skill\n        \n        questions = np.append(q[2:], [target_id])\n        \n        return x, questions","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"import riiideducation\nenv = riiideducation.make_env()\niter_test = env.iter_test()\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"import psutil\nmodel.eval()\n\nprev_test_df = None","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"for (test_df, sample_prediction_df) in tqdm(iter_test):\n    if (prev_test_df is not None) & (psutil.virtual_memory().percent < 90):\n        prev_test_df['answered_correctly'] = eval(test_df['prior_group_answers_correct'].iloc[0])\n        prev_test_df = prev_test_df[prev_test_df.content_type_id == False]\n        \n        prev_group = prev_test_df[['user_id', \n                                   'content_id', \n                                   'answered_correctly']]\\\n            .groupby('user_id').apply(lambda r: (\n                                            r['content_id'].values,\n                                            r['answered_correctly'].values))\n        for prev_user_id in prev_group.index:\n            if prev_user_id in group.index:\n                group[prev_user_id] = (\n                    np.append(group[prev_user_id][0], \n                              prev_group[prev_user_id][0])[-MAX_SEQ:], \n                    np.append(group[prev_user_id][1], \n                              prev_group[prev_user_id][1])[-MAX_SEQ:]\n                )\n \n            else:\n                group[prev_user_id] = (\n                    prev_group[prev_user_id][0], \n                    prev_group[prev_user_id][1]\n                )\n\n    prev_test_df = test_df.copy()\n    \n    test_df = test_df[test_df.content_type_id == False]\n    test_dataset = TestDataset(group, test_df, n_skill)\n    test_dataloader = DataLoader(test_dataset, batch_size=len(test_df), shuffle=False)\n    \n    outs = []\n\n    for item in tqdm(test_dataloader):\n        x = item[0].to(device).long()\n        target_id = item[1].to(device).long()\n\n        with torch.no_grad():\n            output, att_weight = model(x, target_id)\n        outs.extend(torch.sigmoid(output)[:, -1].view(-1).data.cpu().numpy())\n        \n    test_df['answered_correctly'] = outs\n    env.predict(test_df.loc[test_df['content_type_id'] == 0, \n                            ['row_id', 'answered_correctly']])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# sub = pd.read_csv('../working/submission.csv')\n# sub['answered_correctly'].hist(bins=15)","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat":4,"nbformat_minor":4}