diff --git a/src/cs598_dlh_final_project.ipynb b/src/cs598_dlh_final_project.ipynb
index 1fd42899e3d4388ec276c846b0222b943732f323..a6dbd06a4f67e938fc102d339773ff801faccbd6 100644
--- a/src/cs598_dlh_final_project.ipynb
+++ b/src/cs598_dlh_final_project.ipynb
@@ -2193,7 +2193,63 @@
" # print(masks[i_patient, :, ])\n",
" # print(rev_masks[i_patient, :, ])\n",
" \n",
- " return x, masks, rev_x, rev_masks, y\n"
+ " return x, masks, rev_x, rev_masks, y\n",
+ "\n",
+ "def code_emb_per_patient_collate_function(data):\n",
+ " \"\"\"\n",
+ " TODO: Collate the the list of samples into batches. For each patient, you need to pad the diagnosis\n",
+ " sequences to the sample shape (max # visits, max # diagnosis codes). The padding infomation\n",
+ " is stored in `mask`.\n",
+ " \n",
+ " \n",
+ " Arguments:\n",
+ " data: a list of samples fetched from `CustomDataset`\n",
+ " \n",
+ " Outputs:\n",
+ " x: a tensor of shape (# patiens, max # visits, max # diagnosis codes) of type torch.long\n",
+ " masks: a tensor of shape (# patiens, max # visits, max # diagnosis codes) of type torch.bool\n",
+ " rev_x: same as x but in reversed time. This will be used in our RNN model for masking\n",
+ " rev_masks: same as mask but in reversed time. This will be used in our RNN model for masking\n",
+ " y: a tensor of shape (# patiens) of type torch.float\n",
+ " \n",
+ " Note that you can obtains the list of diagnosis codes and the list of hf labels\\n\",\n",
+ " using: `sequences, labels = zip(*data)`\\n\",\n",
+ " \"\"\"\n",
+ " \n",
+ " # sequences, labels = zip(*data)\n",
+ " samples = data\n",
+ " \n",
+ " y = torch.tensor(labels, dtype=torch.float)\n",
+ " \n",
+ " num_patients = len(sequences)\n",
+ " num_visits = [len(patient) for patient in sequences]\n",
+ " num_codes = [len(visit) for patient in sequences for visit in patient]\n",
+ "\n",
+ " max_num_visits = max(num_visits)\n",
+ " max_num_codes = max(num_codes)\n",
+ " \n",
+ " x = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.long)\n",
+ " rev_x = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.long)\n",
+ " masks = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.bool)\n",
+ " rev_masks = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.bool)\n",
+ " for i_patient, patient in enumerate(sequences):\n",
+ " kMaxVisits = len(patient)\n",
+ " for j_visit, visit in enumerate(patient):\n",
+ " \"\"\"\n",
+ " TODO: update `x`, `rev_x`, `masks`, and `rev_masks`\n",
+ " \"\"\"\n",
+ " l = len(visit)\n",
+ " x[i_patient, j_visit, 0:l] = torch.LongTensor(visit)\n",
+ " rev_x[i_patient, kMaxVisits-1-j_visit, 0:l] = torch.LongTensor(visit)\n",
+ " masks[i_patient, j_visit, 0:l] = 1\n",
+ " rev_masks[i_patient, kMaxVisits-1-j_visit, 0:l] = 1\n",
+ " # print(f\"------ v: {kMaxVisits} ------\")\n",
+ " # print(x[i_patient, :, ])\n",
+ " # print(rev_x[i_patient, :, ])\n",
+ " # print(masks[i_patient, :, ])\n",
+ " # print(rev_masks[i_patient, :, ])\n",
+ " \n",
+ " return x, masks, rev_x, rev_masks, y"
]
},
{
@@ -2207,7 +2263,7 @@
},
{
"cell_type": "code",
- "execution_count": 110,
+ "execution_count": 141,
"id": "f6afd1f5-0f3b-4f4b-9c85-f74bb367ef88",
"metadata": {},
"outputs": [],
@@ -2370,7 +2426,7 @@
},
{
"cell_type": "code",
- "execution_count": 138,
+ "execution_count": 140,
"id": "1056b60a-5861-4e47-b694-891053cc8470",
"metadata": {},
"outputs": [
@@ -2381,7 +2437,7 @@
"Some weights of the model checkpoint at bert-base-uncased were not used when initializing BertModel: ['cls.predictions.transform.LayerNorm.bias', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.dense.bias', 'cls.predictions.transform.dense.weight', 'cls.predictions.decoder.weight', 'cls.predictions.bias']\n",
"- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
"- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
- "Generating samples for mortality_pred_task: 100%|█████████████████████████████████████████████████████████| 100/100 [00:00<00:00, 14196.80it/s]\n"
+ "Generating samples for mortality_pred_task: 100%|█████████████████████████████████████████████████████████| 100/100 [00:00<00:00, 12923.05it/s]\n"
]
}
],
@@ -2528,6 +2584,73 @@
"#### CodeEmb Loader"
]
},
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "id": "992467f7-d834-4d4a-872a-714389fa203c",
+ "metadata": {
+ "tags": []
+ },
+ "outputs": [],
+ "source": [
+ "\n",
+ "def code_emb_per_patient_collate_function(data):\n",
+ " \"\"\"\n",
+ " TODO: Collate the the list of samples into batches. For each patient, you need to pad the diagnosis\n",
+ " sequences to the sample shape (max # visits, max # diagnosis codes). The padding infomation\n",
+ " is stored in `mask`.\n",
+ " \n",
+ " \n",
+ " Arguments:\n",
+ " data: a list of samples fetched from `CustomDataset`\n",
+ " \n",
+ " Outputs:\n",
+ " x: a tensor of shape (# patiens, max # visits, max # diagnosis codes) of type torch.long\n",
+ " masks: a tensor of shape (# patiens, max # visits, max # diagnosis codes) of type torch.bool\n",
+ " rev_x: same as x but in reversed time. This will be used in our RNN model for masking\n",
+ " rev_masks: same as mask but in reversed time. This will be used in our RNN model for masking\n",
+ " y: a tensor of shape (# patiens) of type torch.float\n",
+ " \n",
+ " Note that you can obtains the list of diagnosis codes and the list of hf labels\\n\",\n",
+ " using: `sequences, labels = zip(*data)`\\n\",\n",
+ " \"\"\"\n",
+ " \n",
+ " # sequences, labels = zip(*data)\n",
+ " samples = data\n",
+ " \n",
+ " y = torch.tensor(labels, dtype=torch.float)\n",
+ " \n",
+ " num_patients = len(sequences)\n",
+ " num_visits = [len(patient) for patient in sequences]\n",
+ " num_codes = [len(visit) for patient in sequences for visit in patient]\n",
+ "\n",
+ " max_num_visits = max(num_visits)\n",
+ " max_num_codes = max(num_codes)\n",
+ " \n",
+ " x = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.long)\n",
+ " rev_x = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.long)\n",
+ " masks = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.bool)\n",
+ " rev_masks = torch.zeros((num_patients, max_num_visits, max_num_codes), dtype=torch.bool)\n",
+ " for i_patient, patient in enumerate(sequences):\n",
+ " kMaxVisits = len(patient)\n",
+ " for j_visit, visit in enumerate(patient):\n",
+ " \"\"\"\n",
+ " TODO: update `x`, `rev_x`, `masks`, and `rev_masks`\n",
+ " \"\"\"\n",
+ " l = len(visit)\n",
+ " x[i_patient, j_visit, 0:l] = torch.LongTensor(visit)\n",
+ " rev_x[i_patient, kMaxVisits-1-j_visit, 0:l] = torch.LongTensor(visit)\n",
+ " masks[i_patient, j_visit, 0:l] = 1\n",
+ " rev_masks[i_patient, kMaxVisits-1-j_visit, 0:l] = 1\n",
+ " # print(f\"------ v: {kMaxVisits} ------\")\n",
+ " # print(x[i_patient, :, ])\n",
+ " # print(rev_x[i_patient, :, ])\n",
+ " # print(masks[i_patient, :, ])\n",
+ " # print(rev_masks[i_patient, :, ])\n",
+ " \n",
+ " return x, masks, rev_x, rev_masks, y"
+ ]
+ },
{
"cell_type": "code",
"execution_count": null,
@@ -2535,10 +2658,6 @@
"metadata": {},
"outputs": [],
"source": [
- "# Create the transform that will take each sample (visit) in the dataset\n",
- "# and convert the text description of the visit into a single embedding.\n",
- "bert_xform = BertTextEmbedTransform(None, EMBEDDING_DIM_)\n",
- "\n",
"# Mortality Task Datasets: set_task -> SampleEHRDataset(SampleBaseDataset)\n",
"# Add a transform to convert the visit into an embedding.\n",
"mortality_cemb_ds = mimic3base.set_task(task_fn=mortality_pred_task)\n",
@@ -2553,65 +2672,22 @@
" mort_cemb_train_ds,\n",
" batch_size=BATCH_SIZE_,\n",
" shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
+ " collate_fn=code_emb_per_patient_collate_function,\n",
")\n",
"mort_cemb_val_loader = DataLoader(\n",
" mort_cemb_val_ds,\n",
" batch_size=BATCH_SIZE_,\n",
" shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
+ " collate_fn=code_emb_per_patient_collate_function,\n",
")\n",
"mort_cemb_test_loader = DataLoader(\n",
" mort_cemb_test_ds,\n",
" batch_size=BATCH_SIZE_,\n",
" shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
+ " collate_fn=code_emb_per_patient_collate_function,\n",
")"
]
},
- {
- "cell_type": "code",
- "execution_count": null,
- "id": "9436cb7d-e302-460c-b35d-3025dff4b999",
- "metadata": {},
- "outputs": [],
- "source": [
- "# Create the transform that will take each sample (visit) in the dataset\n",
- "# and convert the text description of the visit into a single embedding.\n",
- "bert_xform = BertTextEmbedTransform(None, EMBEDDING_DIM_)\n",
- "\n",
- "# Mortality Task Datasets: set_task -> SampleEHRDataset(SampleBaseDataset)\n",
- "# Add a transform to convert the visit into an embedding.\n",
- "mortality_dataset = mimic3base.set_task(task_fn=mortality_pred_task)\n",
- "mortality_dataset = TextEmbedDataset(mortality_dataset, transform=bert_xform)\n",
- "mort_train_ds, mort_val_ds, mort_test_ds = split_by_patient(mortality_dataset, [0.8, 0.1, 0.1])\n",
- "\n",
- "# create dataloaders (torch.data.DataLoader)\n",
- "# mort_train_loader = get_dataloader(train_ds, batch_size=32, shuffle=True, collate_fn)\n",
- "# mort_val_loader = get_dataloader(val_ds, batch_size=32, shuffle=False)\n",
- "# mort_test_loader = get_dataloader(test_ds, batch_size=32, shuffle=False)\n",
- "\n",
- "mort_train_loader = DataLoader(\n",
- " mort_train_ds,\n",
- " batch_size=BATCH_SIZE_,\n",
- " shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
- ")\n",
- "mort_val_loader = DataLoader(\n",
- " mort_val_ds,\n",
- " batch_size=BATCH_SIZE_,\n",
- " shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
- ")\n",
- "mort_test_loader = DataLoader(\n",
- " mort_test_ds,\n",
- " batch_size=BATCH_SIZE_,\n",
- " shuffle=SHUFFLE_,\n",
- " collate_fn=bert_per_patient_collate_function,\n",
- ")\n",
- "\n"
- ]
- },
{
"cell_type": "markdown",
"id": "ebe2c2d9-c161-4b9d-8182-86384faea929",
@@ -2772,6 +2848,14 @@
"source": [
"## Train DescEmb"
]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "id": "fc9335e3-0da2-47f3-9a90-53531ad277b0",
+ "metadata": {},
+ "outputs": [],
+ "source": []
}
],
"metadata": {