Add model loading instructions.

README.md

@@ -50,8 +50,7 @@ Clone [fairseq_signals](https://github.com/Jwoo5/fairseq-signals) and refer to t
 ### Data Preparation
 We implemented a flexible, end-to-end, multi-source data preprocessing pipeline. Please refer to it [here](https://github.com/Jwoo5/fairseq-signals/tree/master/scripts/preprocess/ecg).
 
-### Inference
-
+### Inference & Model Loading
 See our [inference tutorial notebook](inference_tutorial.ipynb)!
 
 ### Training

inference_tutorial.ipynb

@@ -492,6 +492,66 @@
     "# Visualize predicted and actual labels side-by-side\n",
     "pred_thresh_mapped[['predicted']].join(code_15_labels[['actual']], how='left')"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "d7c6b21a-2d1e-453e-ab8f-dc2015dd80a0",
+   "metadata": {},
+   "source": [
+    "# 5. Extra - Load models"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "ed6cece6-a6e0-46c5-b022-7bdeb9ce68ec",
+   "metadata": {},
+   "source": [
+    "Above, inference is ran through scripts using the hydra client. Loading the models separately is possible like so:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8afcc57b-cf99-4538-bac9-82f3381bcef4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import yaml\n",
+    "from fairseq_signals.models import build_model\n",
+    "\n",
+    "with open(\"ckpts/mimic_iv_ecg_physionet_pretrained.yaml\" ,\"r\") as f:\n",
+    "  pretrained_cfg = yaml.load(f, yaml.FullLoader)\n",
+    "\n",
+    "pretrained_model_cfg = pretrained_cfg[\"model\"]\n",
+    "pretrained_model = build_model(pretrained_model_cfg, task=None)\n",
+    "pretrained_model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ed23b1b2-3aae-401d-8a44-f9088a52c8c0",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import yaml\n",
+    "from fairseq_signals.models import build_model\n",
+    "\n",
+    "with open(\"ckpts/physionet_finetuned.yaml\" ,\"r\") as f:\n",
+    "  finetuned_cfg = yaml.load(f, yaml.FullLoader)\n",
+    "\n",
+    "finetuned_model_cfg = finetuned_cfg[\"model\"]\n",
+    "finetuned_model = build_model(finetuned_model_cfg, task=None)\n",
+    "finetuned_model"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "51c04185-5013-4d69-b542-768995a3b8ce",
+   "metadata": {},
+   "source": [
+    "If you are seeing `FileNotFoundError: [Errno 2] No such file or directory: '<REDACTED>'`, please ensure you have run the `Prepare checkpoints` cells above!"
+   ]
   }
  ],
  "metadata": {