Add files via upload

Author: SVC04

scripts/coherence_test.ipynb

@@ -0,0 +1,92 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "55c5916a-a77f-4497-9123-0f73b4018273",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from sgnlp.models.coherence_momentum import (\n",
+    "    CoherenceMomentumModel, \n",
+    "    CoherenceMomentumConfig, \n",
+    "    CoherenceMomentumPreprocessor\n",
+    ")\n",
+    "import pandas as pd\n",
+    "\n",
+    "# Load CoherenceMomentum model\n",
+    "\n",
+    "config = CoherenceMomentumConfig.from_pretrained(\n",
+    "    \"https://storage.googleapis.com/sgnlp-models/models/coherence_momentum/config.json\"\n",
+    ")\n",
+    "model = CoherenceMomentumModel.from_pretrained(\n",
+    "    \"https://storage.googleapis.com/sgnlp-models/models/coherence_momentum/pytorch_model.bin\",\n",
+    "    config=config\n",
+    ")\n",
+    "preprocessor = CoherenceMomentumPreprocessor(config.model_size, config.max_len)\n",
+    "\n",
+    "\n",
+    "# Function: compute coherence score for a text\n",
+    "\n",
+    "def get_coherence_score(text):\n",
+    "    inputs = preprocessor([text])\n",
+    "    return model.get_main_score(inputs[\"tokenized_texts\"]).item()\n",
+    "\n",
+    "\n",
+    "# Compute full vs segmented coherence for each article\n",
+    "\n",
+    "results = []\n",
+    "for i, (full_text, segments) in enumerate(zip(preprocessed_articles, preprocessed_segments)):\n",
+    "    reordered_text = \" \".join(segments)  # Reconstructed article from segments\n",
+    "    score_full = get_coherence_score(full_text)\n",
+    "    score_segments = get_coherence_score(reordered_text)\n",
+    "    drift = score_full - score_segments  \n",
+    "\n",
+    "    results.append({\n",
+    "        \"Article\": f\"A{i+1}\",\n",
+    "        \"Coherence Full\": round(score_full, 4),\n",
+    "        \"Coherence Reordered\": round(score_segments, 4),\n",
+    "        \"Drift\": round(drift, 4),\n",
+    "    })\n",
+    "\n",
+    "\n",
+    "# Save per-article results\n",
+    "\n",
+    "df_coherence = pd.DataFrame(results)\n",
+    "df_coherence.to_csv(\"coherence_results.csv\", index=False)\n",
+    "\n",
+    "\n",
+    "# Save summary stats (mean/std)\n",
+    "\n",
+    "df_summary = df_coherence.drop(\"Article\", axis=1).agg(['mean', 'std']).transpose().reset_index()\n",
+    "df_summary.columns = [\"Metric\", \"Mean\", \"Std Dev\"]\n",
+    "df_summary.to_csv(\"coherence_summary.csv\", index=False)\n",
+    "\n",
+    "# Print sample outputs\n",
+    "print(df_coherence.head())\n",
+    "print(df_summary)\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}