{"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_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"from Bio import SeqIO\nimport pandas as pd","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-07-06T11:23:25.934027Z","iopub.execute_input":"2023-07-06T11:23:25.934794Z","iopub.status.idle":"2023-07-06T11:23:26.104506Z","shell.execute_reply.started":"2023-07-06T11:23:25.934752Z","shell.execute_reply":"2023-07-06T11:23:26.10328Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"MAIN_DIR = \"/kaggle/input/cafa-5-protein-function-prediction\"\n\n# UTILITARIES\nimport numpy as np\nfrom tqdm import tqdm\nimport time\nimport matplotlib.pyplot as plt\nplt.style.use('ggplot')\n\n# TORCH MODULES FOR METRICS COMPUTATION :\nimport torch\nfrom torch.utils.data import Dataset\nfrom torch import nn\nfrom torch.utils.data import random_split\nfrom torch.optim.lr_scheduler import ReduceLROnPlateau\nfrom torchmetrics.classification import MultilabelF1Score\nfrom torchmetrics.classification import MultilabelAccuracy\n\nimport pytorch_lightning as pl\nfrom pytorch_lightning import Trainer\nfrom pytorch_lightning.loggers import WandbLogger\n\n# WANDB FOR LIGHTNING :\nimport wandb\n\n# FILES VISUALIZATION\nimport os\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        print(os.path.join(dirname, filename))","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:23:30.906492Z","iopub.execute_input":"2023-07-06T11:23:30.906961Z","iopub.status.idle":"2023-07-06T11:23:47.420583Z","shell.execute_reply.started":"2023-07-06T11:23:30.906921Z","shell.execute_reply":"2023-07-06T11:23:47.419588Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class config:\n    train_sequences_path = MAIN_DIR  + \"/Train/train_sequences.fasta\"\n    train_labels_path = MAIN_DIR + \"/Train/train_terms.tsv\"\n    test_sequences_path = MAIN_DIR + \"/Test (Targets)/testsuperset.fasta\"\n    \n    num_labels = 500\n    n_epochs = 20\n    batch_size = 128\n    lr = 0.008\n    \n    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:23:47.422584Z","iopub.execute_input":"2023-07-06T11:23:47.42294Z","iopub.status.idle":"2023-07-06T11:23:47.442798Z","shell.execute_reply.started":"2023-07-06T11:23:47.4229Z","shell.execute_reply":"2023-07-06T11:23:47.441951Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Load ProtBERT Model...\")\n# PROT BERT LOADING :\nfrom transformers import BertModel, BertTokenizer\ntokenizer = BertTokenizer.from_pretrained(\"Rostlab/prot_bert\", do_lower_case=False )\nmodel = BertModel.from_pretrained(\"Rostlab/prot_bert\").to(config.device)","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:23:47.444147Z","iopub.execute_input":"2023-07-06T11:23:47.444651Z","iopub.status.idle":"2023-07-06T11:24:10.004926Z","shell.execute_reply.started":"2023-07-06T11:23:47.444619Z","shell.execute_reply":"2023-07-06T11:24:10.003917Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"GENERATE TARGETS FOR ENTRY IDS (\"+str(config.num_labels)+\" MOST COMMON GO TERMS)\")\nids = np.load(\"/kaggle/input/protbert-embeddings-for-cafa5/train_ids.npy\")\nlabels = pd.read_csv(config.train_labels_path, sep = \"\\t\")\n\ntop_terms = labels.groupby(\"term\")[\"EntryID\"].count().sort_values(ascending=False)\nlabels_names = top_terms[:config.num_labels].index.values\ntrain_labels_sub = labels[(labels.term.isin(labels_names)) & (labels.EntryID.isin(ids))]\nid_labels = train_labels_sub.groupby('EntryID')['term'].apply(list).to_dict()\n\ngo_terms_map = {label: i for i, label in enumerate(labels_names)}\nlabels_matrix = np.empty((len(ids), len(labels_names)))\n\nfor index, id in tqdm(enumerate(ids)):\n    id_gos_list = id_labels[id]\n    temp = [go_terms_map[go] for go in labels_names if go in id_gos_list]\n    labels_matrix[index, temp] = 1\n\nnp.save(\"/kaggle/working/train_targets_top\"+str(config.num_labels)+\".npy\", np.array(labels_matrix))\nprint(\"GENERATION FINISHED!\")","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:24:10.007791Z","iopub.execute_input":"2023-07-06T11:24:10.008175Z","iopub.status.idle":"2023-07-06T11:25:25.460333Z","shell.execute_reply.started":"2023-07-06T11:24:10.008142Z","shell.execute_reply":"2023-07-06T11:25:25.459284Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class ProteinSequenceDataset(Dataset):\n    \n    def __init__(self, datatype, embeddings_source):\n        super(ProteinSequenceDataset).__init__()\n        self.datatype = datatype\n        \n        if embeddings_source in [\"ProtBERT\", \"EMS2\"]:\n            embeds = np.load(\"/kaggle/input/\"+embeds_map[embeddings_source]+\"/\"+datatype+\"_embeddings.npy\")\n            ids = np.load(\"/kaggle/input/\"+embeds_map[embeddings_source]+\"/\"+datatype+\"_ids.npy\")\n        \n        if embeddings_source == \"T5\":\n            embeds = np.load(\"/kaggle/input/\"+embeds_map[embeddings_source]+\"/\"+datatype+\"_embeds.npy\")\n            ids = np.load(\"/kaggle/input/\"+embeds_map[embeddings_source]+\"/\"+datatype+\"_ids.npy\")\n            \n        embeds_list = []\n        for l in range(embeds.shape[0]):\n            embeds_list.append(embeds[l,:])\n        self.df = pd.DataFrame(data={\"EntryID\": ids, \"embed\" : embeds_list})\n        \n        if datatype==\"train\":\n            df_labels = np.load(\"/kaggle/working/train_targets_top\"+str(config.num_labels)+\".npy\")\n            self.df[\"labels_vect\"] = df_labels.tolist()\n            #self.df = self.df.merge(df_labels, on=\"EntryID\")\n        \n    def __len__(self):\n        return len(self.df)\n    \n    def __getitem__(self, index):\n        embed = torch.tensor(self.df.iloc[index][\"embed\"] , dtype = torch.float32)\n        if self.datatype==\"train\":\n            targets = torch.tensor(self.df.iloc[index][\"labels_vect\"], dtype = torch.float32)\n            return embed, targets\n        if self.datatype==\"test\":\n            id = self.df.iloc[index][\"EntryID\"]\n            return embed, id\n        ","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:25:25.461791Z","iopub.execute_input":"2023-07-06T11:25:25.462426Z","iopub.status.idle":"2023-07-06T11:25:25.476476Z","shell.execute_reply.started":"2023-07-06T11:25:25.462392Z","shell.execute_reply":"2023-07-06T11:25:25.475471Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class MultiLayerPerceptron(torch.nn.Module):\n\n    def __init__(self, input_dim, num_classes):\n        super(MultiLayerPerceptron, self).__init__()\n\n        self.linear1 = torch.nn.Linear(input_dim, 1012)\n        self.activation1 = torch.nn.ReLU()\n        self.linear2 = torch.nn.Linear(1012, 712)\n        self.activation2 = torch.nn.ReLU()\n        self.linear3 = torch.nn.Linear(712, 356)\n        self.activation3 = torch.nn.ReLU\n        self.linear4 = torch.nn.Linear(356,num_classes)\n\n    def forward(self, x):\n        x = self.linear1(x)\n        x = self.activation1(x)\n        x = self.linear2(x)\n        x = self.activation2(x)\n        x = self.linear3(x)\n        x = self.activation3(x)\n        x = self.linear4(x)\n        return x","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:25:25.477956Z","iopub.execute_input":"2023-07-06T11:25:25.479042Z","iopub.status.idle":"2023-07-06T11:25:25.49518Z","shell.execute_reply.started":"2023-07-06T11:25:25.479004Z","shell.execute_reply":"2023-07-06T11:25:25.494222Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CNN1D(nn.Module):\n    def __init__(self, input_dim, num_classes):\n        super(CNN1D, self).__init__()\n        # (batch_size, channels, embed_size)\n        self.conv1 = nn.Conv1d(in_channels=1, out_channels=3, kernel_size=3, dilation=1, padding=1, stride=1)\n        # (batch_size, 3, embed_size)\n        self.pool1 = nn.MaxPool1d(kernel_size=2, stride=2)\n        # (batch_size, 3, embed_size/2 = 512)\n        self.conv2 = nn.Conv1d(in_channels=3, out_channels=8, kernel_size=3, dilation=1, padding=1, stride=1)\n        # (batch_size, 8, embed_size/2 = 512)\n        self.pool2 = nn.MaxPool1d(kernel_size=2, stride=2)\n        # (batch_size, 8, embed_size/4 = 256)\n        self.fc1 = nn.Linear(in_features=int(8 * input_dim/4), out_features=128)\n        self.fc2 = nn.Linear(in_features=128, out_features=num_classes)\n\n    def forward(self, x):\n        x = x.reshape(x.shape[0], 1, x.shape[1])\n        x = self.pool1(nn.functional.relu(self.conv1(x)))\n        x = self.pool2(nn.functional.relu(self.conv2(x)))\n        x = torch.flatten(x, 1)\n        x = nn.functional.relu(self.fc1(x))\n        x = self.fc2(x)\n        return x","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:25:25.496815Z","iopub.execute_input":"2023-07-06T11:25:25.497191Z","iopub.status.idle":"2023-07-06T11:25:25.511285Z","shell.execute_reply.started":"2023-07-06T11:25:25.497127Z","shell.execute_reply":"2023-07-06T11:25:25.5104Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class MixedModel(nn.Module):\n    def __init__(self, input_dim, num_classes):\n        super(MixedModel, self).__init__()\n\n        self.mlp = MultiLayerPerceptron(input_dim, num_classes)\n        self.cnn = CNN1D(input_dim, num_classes)\n\n    def forward(self, x):\n        mlp_output = self.mlp(x)\n        cnn_output = self.cnn(x)\n\n        # Perform some operation to combine the outputs\n        combined_output = mlp_output + cnn_output\n\n        return combined_output","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:52:41.116752Z","iopub.execute_input":"2023-07-06T11:52:41.117783Z","iopub.status.idle":"2023-07-06T11:52:41.124902Z","shell.execute_reply.started":"2023-07-06T11:52:41.117722Z","shell.execute_reply":"2023-07-06T11:52:41.123911Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# def train_model(embeddings_source, model_type=\"linear\", train_size=0.9):\n    \n#     train_dataset = ProteinSequenceDataset(datatype=\"train\", embeddings_source = embeddings_source)\n    \n#     train_set, val_set = random_split(train_dataset, lengths = [int(len(train_dataset)*train_size), len(train_dataset)-int(len(train_dataset)*train_size)])\n#     train_dataloader = torch.utils.data.DataLoader(train_set, batch_size=config.batch_size, shuffle=True)\n#     val_dataloader = torch.utils.data.DataLoader(val_set, batch_size=config.batch_size, shuffle=True)\n\n#     if model_type == \"linear\":\n#         model = MultiLayerPerceptron(input_dim=embeds_dim[embeddings_source], num_classes=config.num_labels).to(config.device)\n#     if model_type == \"convolutional\":\n#         model = CNN1D(input_dim=embeds_dim[embeddings_source], num_classes=config.num_labels).to(config.device)\n\n#     optimizer = torch.optim.Adam(model.parameters(), lr = config.lr)\n#     scheduler = ReduceLROnPlateau(optimizer, factor=0.1, patience=1)\n#     CrossEntropy = torch.nn.CrossEntropyLoss()\n#     f1_score = MultilabelF1Score(num_labels=config.num_labels).to(config.device)\n#     n_epochs = config.n_epochs\n\n#     print(\"BEGIN TRAINING...\")\n#     train_loss_history=[]\n#     val_loss_history=[]\n    \n#     train_f1score_history=[]\n#     val_f1score_history=[]\n#     for epoch in range(n_epochs):\n#         print(\"EPOCH \", epoch+1)\n#         ## TRAIN PHASE :\n#         losses = []\n#         scores = []\n#         for embed, targets in tqdm(train_dataloader):\n#             embed, targets = embed.to(config.device), targets.to(config.device)\n#             optimizer.zero_grad()\n#             preds = model(embed)\n#             loss= CrossEntropy(preds, targets)\n#             score=f1_score(preds, targets)\n#             losses.append(loss.item()) \n#             scores.append(score.item())\n#             loss.backward()\n#             optimizer.step()\n#         avg_loss = np.mean(losses)\n#         avg_score = np.mean(scores)\n#         print(\"Running Average TRAIN Loss : \", avg_loss)\n#         print(\"Running Average TRAIN F1-Score : \", avg_score)\n#         train_loss_history.append(avg_loss)\n#         train_f1score_history.append(avg_score)\n        \n#         ## VALIDATION PHASE : \n#         losses = []\n#         scores = []\n#         for embed, targets in val_dataloader:\n#             embed, targets = embed.to(config.device), targets.to(config.device)\n#             preds = model(embed)\n#             loss= CrossEntropy(preds, targets)\n#             score=f1_score(preds, targets)\n#             losses.append(loss.item())\n#             scores.append(score.item())\n#         avg_loss = np.mean(losses)\n#         avg_score = np.mean(scores)\n#         print(\"Running Average VAL Loss : \", avg_loss)\n#         print(\"Running Average VAL F1-Score : \", avg_score)\n#         val_loss_history.append(avg_loss)\n#         val_f1score_history.append(avg_score)\n        \n#         scheduler.step(avg_loss)\n#         print(\"\\n\")\n        \n#     print(\"TRAINING FINISHED\")\n#     print(\"FINAL TRAINING SCORE : \", train_f1score_history[-1])\n#     print(\"FINAL VALIDATION SCORE : \", val_f1score_history[-1])\n    \n#     losses_history = {\"train\" : train_loss_history, \"val\" : val_loss_history}\n#     scores_history = {\"train\" : train_f1score_history, \"val\" : val_f1score_history}\n    \n#     return model, losses_history, scores_history","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:52:41.723841Z","iopub.execute_input":"2023-07-06T11:52:41.724187Z","iopub.status.idle":"2023-07-06T11:52:41.73297Z","shell.execute_reply.started":"2023-07-06T11:52:41.724158Z","shell.execute_reply":"2023-07-06T11:52:41.730126Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def train_model(embeddings_source, model_type=\"mixed\", train_size=0.9):\n    train_dataset = ProteinSequenceDataset(datatype=\"train\", embeddings_source=embeddings_source)\n\n    train_set, val_set = random_split(train_dataset, lengths=[int(len(train_dataset)*train_size), len(train_dataset)-int(len(train_dataset)*train_size)])\n    train_dataloader = torch.utils.data.DataLoader(train_set, batch_size=config.batch_size, shuffle=True)\n    val_dataloader = torch.utils.data.DataLoader(val_set, batch_size=config.batch_size, shuffle=True)\n\n    if model_type == \"linear\":\n        model = MultiLayerPerceptron(input_dim=embeds_dim[embeddings_source], num_classes=config.num_labels).to(config.device)\n    elif model_type == \"convolutional\":\n        model = CNN1D(input_dim=embeds_dim[embeddings_source], num_classes=config.num_labels).to(config.device)\n    elif model_type == \"mixed\":\n        model = MixedModel(input_dim=embeds_dim[embeddings_source], num_classes=config.num_labels).to(config.device)\n    else:\n        raise ValueError(\"Invalid model_type\")\n\n    optimizer = torch.optim.Adam(model.parameters(), lr=config.lr)\n    scheduler = ReduceLROnPlateau(optimizer, factor=0.1, patience=1)\n    CrossEntropy = torch.nn.CrossEntropyLoss()\n    f1_score = MultilabelF1Score(num_labels=config.num_labels).to(config.device)\n    n_epochs = config.n_epochs\n\n    print(\"BEGIN TRAINING...\")\n    train_loss_history = []\n    val_loss_history = []\n    train_f1score_history = []\n    val_f1score_history = []\n\n    for epoch in range(n_epochs):\n        print(\"EPOCH \", epoch+1)\n\n        # TRAIN PHASE :\n        losses = []\n        scores = []\n\n        for embed, targets in tqdm(train_dataloader):\n            embed, targets = embed.to(config.device), targets.to(config.device)\n            optimizer.zero_grad()\n            preds = model(embed)\n            loss = CrossEntropy(preds, targets)\n            score = f1_score(preds, targets)\n            losses.append(loss.item()) \n            scores.append(score.item())\n            loss.backward()\n            optimizer.step()\n\n        avg_loss = np.mean(losses)\n        avg_score = np.mean(scores)\n        print(\"Running Average TRAIN Loss : \", avg_loss)\n        print(\"Running Average TRAIN F1-Score : \", avg_score)\n        train_loss_history.append(avg_loss)\n        train_f1score_history.append(avg_score)\n\n        # VALIDATION PHASE : \n        losses = []\n        scores = []\n\n        for embed, targets in val_dataloader:\n            embed, targets = embed.to(config.device), targets.to(config.device)\n            preds = model(embed)\n            loss = CrossEntropy(preds, targets)\n            score = f1_score(preds, targets)\n            losses.append(loss.item())\n            scores.append(score.item())\n\n        avg_loss = np.mean(losses)\n        avg_score = np.mean(scores)\n        print(\"Running Average VAL Loss : \", avg_loss)\n        print(\"Running Average VAL F1-Score : \", avg_score)\n        val_loss_history.append(avg_loss)\n        val_f1score_history.append(avg_score)\n\n        scheduler.step(avg_loss)\n        print(\"\\n\")\n\n    print(\"TRAINING FINISHED\")\n    print(\"FINAL TRAINING SCORE : \", train_f1score_history[-1])\n    print(\"FINAL VALIDATION SCORE : \", val_f1score_history[-1])\n\n    losses_history = {\"train\": train_loss_history, \"val\": val_loss_history}\n    scores_history = {\"train\": train_f1score_history, \"val\": val_f1score_history}\n\n    return model, losses_history, scores_history\n","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:52:42.036125Z","iopub.execute_input":"2023-07-06T11:52:42.036466Z","iopub.status.idle":"2023-07-06T11:52:42.055247Z","shell.execute_reply.started":"2023-07-06T11:52:42.03643Z","shell.execute_reply":"2023-07-06T11:52:42.05431Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Directories for the different embedding vectors : \nembeds_map = {\n    \"T5\" : \"t5embeds\",\n    \"ProtBERT\" : \"protbert-embeddings-for-cafa5\",\n    \"EMS2\" : \"cafa-5-ems-2-embeddings-numpy\"\n}\n\n# Length of the different embedding vectors :\nembeds_dim = {\n    \"T5\" : 1024,\n    \"ProtBERT\" : 1024,\n    \"EMS2\" : 1280\n}","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:25:25.532474Z","iopub.execute_input":"2023-07-06T11:25:25.532856Z","iopub.status.idle":"2023-07-06T11:25:25.545331Z","shell.execute_reply.started":"2023-07-06T11:25:25.532826Z","shell.execute_reply":"2023-07-06T11:25:25.544358Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"protbert_model, protbert_losses, protbert_scores = train_model(embeddings_source=\"ProtBERT\",model_type=\"mixed\")","metadata":{"execution":{"iopub.status.busy":"2023-07-06T11:52:50.038167Z","iopub.execute_input":"2023-07-06T11:52:50.038553Z","iopub.status.idle":"2023-07-06T12:04:28.570179Z","shell.execute_reply.started":"2023-07-06T11:52:50.038521Z","shell.execute_reply":"2023-07-06T12:04:28.569166Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ems2_model, ems2_losses, ems2_scores = train_model(embeddings_source=\"EMS2\",model_type=\"linear\")","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:04:32.837555Z","iopub.execute_input":"2023-07-06T12:04:32.837923Z","iopub.status.idle":"2023-07-06T12:15:34.459813Z","shell.execute_reply.started":"2023-07-06T12:04:32.837893Z","shell.execute_reply":"2023-07-06T12:15:34.458776Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.figure(figsize = (10, 4))\nplt.plot(ems2_losses[\"val\"], label = \"EMS2\")\nplt.plot(protbert_losses[\"val\"], label = \"ProtBERT\")\nplt.title(\"Validation Losses for # Vector Embeddings\")\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"Average Loss\")\nplt.legend()\nplt.show()\n\nplt.figure(figsize = (10, 4))\nplt.plot(ems2_scores[\"val\"], label = \"EMS2\")\nplt.plot(protbert_scores[\"val\"], label = \"ProtBERT\")\nplt.title(\"Validation F1-Scores for # Vector Embeddings\")\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"Average F1-Score\")\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:18:34.756692Z","iopub.execute_input":"2023-07-06T12:18:34.757656Z","iopub.status.idle":"2023-07-06T12:18:35.424511Z","shell.execute_reply.started":"2023-07-06T12:18:34.757611Z","shell.execute_reply":"2023-07-06T12:18:35.423645Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def predict(embeddings_source):\n    \n    test_dataset = ProteinSequenceDataset(datatype=\"test\", embeddings_source = embeddings_source)\n    test_dataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False)\n    \n    if embeddings_source == \"T5\":\n        model = t5_model\n    if embeddings_source == \"ProtBERT\":\n        model = protbert_model\n    if embeddings_source == \"EMS2\":\n        model = ems2_model\n        \n    model.eval()\n    \n    labels = pd.read_csv(config.train_labels_path, sep = \"\\t\")\n    top_terms = labels.groupby(\"term\")[\"EntryID\"].count().sort_values(ascending=False)\n    labels_names = top_terms[:config.num_labels].index.values\n    print(\"GENERATE PREDICTION FOR TEST SET...\")\n\n    ids_ = np.empty(shape=(len(test_dataloader)*config.num_labels,), dtype=object)\n    go_terms_ = np.empty(shape=(len(test_dataloader)*config.num_labels,), dtype=object)\n    confs_ = np.empty(shape=(len(test_dataloader)*config.num_labels,), dtype=np.float32)\n\n    for i, (embed, id) in tqdm(enumerate(test_dataloader)):\n        embed = embed.to(config.device)\n        confs_[i*config.num_labels:(i+1)*config.num_labels] = torch.nn.functional.sigmoid(model(embed)).squeeze().detach().cpu().numpy()\n        ids_[i*config.num_labels:(i+1)*config.num_labels] = id[0]\n        go_terms_[i*config.num_labels:(i+1)*config.num_labels] = labels_names\n\n    submission_df = pd.DataFrame(data={\"Id\" : ids_, \"GO term\" : go_terms_, \"Confidence\" : confs_})\n    print(\"PREDICTIONS DONE\")\n    return submission_df","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:18:41.312339Z","iopub.execute_input":"2023-07-06T12:18:41.312695Z","iopub.status.idle":"2023-07-06T12:18:41.327919Z","shell.execute_reply.started":"2023-07-06T12:18:41.312666Z","shell.execute_reply":"2023-07-06T12:18:41.326796Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission_df = predict(\"ProtBERT\")","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:18:47.253354Z","iopub.execute_input":"2023-07-06T12:18:47.253707Z","iopub.status.idle":"2023-07-06T12:21:44.494344Z","shell.execute_reply.started":"2023-07-06T12:18:47.253678Z","shell.execute_reply":"2023-07-06T12:21:44.49332Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#np.save(\"pred.npy\", submission_df)","metadata":{"execution":{"iopub.status.busy":"2023-07-04T17:02:43.517044Z","iopub.execute_input":"2023-07-04T17:02:43.517444Z","iopub.status.idle":"2023-07-04T17:03:22.062442Z","shell.execute_reply.started":"2023-07-04T17:02:43.517413Z","shell.execute_reply":"2023-07-04T17:03:22.061294Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission_df","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:22:01.984398Z","iopub.execute_input":"2023-07-06T12:22:01.984777Z","iopub.status.idle":"2023-07-06T12:22:02.003986Z","shell.execute_reply.started":"2023-07-06T12:22:01.984722Z","shell.execute_reply":"2023-07-06T12:22:02.003102Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission_df.to_csv('submission.tsv', sep='\\t', index=False)","metadata":{"execution":{"iopub.status.busy":"2023-07-06T12:22:05.704835Z","iopub.execute_input":"2023-07-06T12:22:05.70521Z","iopub.status.idle":"2023-07-06T12:26:34.340333Z","shell.execute_reply.started":"2023-07-06T12:22:05.705181Z","shell.execute_reply":"2023-07-06T12:26:34.339266Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}