a few fixes

biomedia-mira · Feb 16, 2024 · 5668c05 · 5668c05
1 parent 9fa8a5e
commit 5668c05
Show file tree

Hide file tree

Showing 8 changed files with 144 additions and 820 deletions.
diff --git a/configs/data/chexpert.yaml b/configs/data/chexpert.yaml
@@ -1,8 +1,8 @@
 defaults:
   - base.yaml
   - _self_
-_target_: data_handling.padchest.CheXpertDataModule
-dataset: padchest
+_target_: data_handling.xray.CheXpertDataModule
+dataset: chexpert
 batch_size: 24
 num_workers: 12
 augmentations:

diff --git a/configs/data/padchest.yaml b/configs/data/padchest.yaml
@@ -1,7 +1,7 @@
 defaults:
   - base.yaml
   - _self_
-_target_: data_handling.padchest.PadChestDataModule
+_target_: data_handling.xray.PadChestDataModule
 dataset: padchest
 batch_size: 16
 num_workers: 14

diff --git a/evaluation/chexpert_pneumo.ipynb b/evaluation/chexpert_pneumo.ipynb
@@ -1,40 +1,27 @@
 {
  "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# CheXpert evaluation"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
     "import sys\n",
-    "import seaborn as sns\n",
-    "\n",
     "sys.path.append(\"/vol/biomedic3/mb121/causal-contrastive\")\n",
-    "\n",
-    "from sklearn.metrics import roc_auc_score, balanced_accuracy_score\n",
-    "\n",
-    "import numpy as np\n",
-    "import matplotlib.pyplot as plt\n",
-    "\n",
-    "from classification.classification_module import ClassificationModule\n",
-    "\n",
-    "from hydra import compose, initialize\n",
-    "import pandas as pd\n",
-    "from data_handling.padchest import CheXpertDataModule\n",
-    "\n",
-    "import matplotlib\n",
-    "from evaluation.helper_functions import (\n",
-    "    extract_train_label_prop,\n",
-    "    extract_run_type,\n",
-    "    extract_pretraining_type,\n",
-    "    extract_finetuning_type,\n",
-    "    run_inference,\n",
-    ")\n",
-    "\n",
-    "sns.set_theme(context=\"paper\", style=\"whitegrid\", font_scale=1.5)\n",
-    "matplotlib.rcParams[\"font.family\"] = \"serif\"\n",
     "import os\n",
-    "\n",
+    "import pandas as pd\n",
+    "from hydra import compose, initialize\n",
+    "from sklearn.metrics import roc_auc_score\n",
+    "from data_handling.xray import CheXpertDataModule\n",
+    "from classification.classification_module import ClassificationModule\n",
+    "from evaluation.helper_functions import run_inference\n",
     "os.chdir(\"/vol/biomedic3/mb121/causal-contrastive/evaluation\")"
    ]
   },
@@ -44,7 +31,15 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "model_dict_normal = {\n",
+    "# Mapping from human readable run name to Weights&Biases run_id. \n",
+    "\n",
+    "# Human readable name should be in format:\n",
+    "# for finetuning:\n",
+    "# {simclr/simclrcf/simclrcfaug}-{train_prop}-{seed}\n",
+    "# for linear probing\n",
+    "# {simclr/simclrcf/simclrcfaug}head-{train_prop}-{seed}\n",
+    "\n",
+    "model_dict_normal: dict[str, str] = {\n",
     "    \"simclr-1.0-33\": \"84lv0t6h\",\n",
     "    \"simclr-1.0-22\": \"yp8kgxkn\",\n",
     "    \"simclr-1.0-11\": \"wmbptk3z\",\n",
@@ -105,6 +100,9 @@
     "    \"simclrcfhead-0.25-11\": \"h92wu2up\",\n",
     "    \"simclrhead-0.1-33\": \"x7afc3ic\",\n",
     "    \"simclrhead-1.0-33\": \"gj75gyod\",\n",
+    "    \"supervised-0.25-11\": 'af029hmt',\n",
+    "    'supervised-0.25-22': 'r5rzknzo',\n",
+    "    'supervised-0.25-33': 'yms2a9pj'\n",
     "}"
    ]
   },
@@ -150,11 +148,6 @@
     "        res = {}\n",
     "        res[\"N_test\"] = [inference_results[\"targets\"].shape[0]]\n",
     "        res[\"Scanner\"] = [\"CheXpert\"]\n",
-    "        res[\"Bal_Acc\"] = [\n",
-    "            balanced_accuracy_score(\n",
-    "                inference_results[\"targets\"], np.argmax(inference_results[\"confs\"], 1)\n",
-    "            )\n",
-    "        ]\n",
     "        res[\"run_name\"] = run_name\n",
     "        res[\"ROC\"] = [\n",
     "            roc_auc_score(\n",
@@ -173,18 +166,7 @@
    "execution_count": null,
    "metadata": {},
    "outputs": [],
-   "source": [
-    "df[\"ctrain_label_prop\"] = df.run_name.apply(extract_train_label_prop)\n",
-    "df[\"group\"] = df.run_name.apply(extract_run_type)\n",
-    "df[\"Pretraining\"] = df.group.apply(lambda x: extract_pretraining_type(x))\n",
-    "df[\"Classifier\"] = df.group.apply(lambda x: extract_finetuning_type(x))\n",
-    "tmp = (\n",
-    "    df.groupby([\"Pretraining\", \"Classifier\", \"ctrain_label_prop\"])\n",
-    "    .run_name.unique()\n",
-    "    .apply(lambda x: len(x))\n",
-    ")\n",
-    "tmp.loc[tmp < 3]"
-   ]
+   "source": []
   }
  ],
  "metadata": {