QPR-12455 clean notebook

Author: Roland Lichters

Examples/Example_62/ore.ipynb

@@ -86,7 +86,7 @@
     }
    },
    "source": [
-    "## Run ORE \"Classic\", Bump Sensitivities, Single-Threaded"
+    "## Run ORE, Bump & Reval Sensitivities, Single-Threaded"
    ]
   },
   {
@@ -130,49 +130,6 @@
     "#utilities.writeList(ore.getReportNames())"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {},
-   "source": [
-    "## Run Again with ComputationGraph Enabled"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "params_cg = Parameters()\n",
-    "params_cg.fromFile(\"Input/ore_cg.xml\")\n",
-    "ore_cg = OREApp(params_cg)\n",
-    "ore_cg.run()\n",
-    "utilities.checkErrorsAndRunTime(ore_cg)\n",
-    "#utilities.writeList(ore_bump.getReportNames())"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# Compare Pricing Stats\n",
-    "utilities.match_pricingstats_12(ore, ore_cg, \"Speedup\", True)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "# Compare Sensitivities\n",
-    "# If we set pricing engine's ExternalDeviceCompatibilityMode to false, then get zero deviations from the reference run above\n",
-    "# if we set ExternalDeviceCompatibilityMode to true, then we see differences below due to different ordering of random variates in both runs\n",
-    "utilities.match_sensi_reports(ore, ore_cg, \"DeltaBump\", \"DeltaCG\", False)"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {
@@ -195,7 +152,13 @@
     }
    },
    "source": [
-    "## Run Again using AAD"
+    "## Run ORE using AAD\n",
+    "\n",
+    "Check the following is set in **pricingengine_ad.xml**\n",
+    "- **UseAD** is set to true\n",
+    "\n",
+    "Check the following is set in **sensitivity.xml**\n",
+    "- **ComputeGamma** is set to false: otherwise AD is deactivated, because it does not support gamma and cross gamma calculation so far\n"
    ]
   },
   {
@@ -239,7 +202,7 @@
    "outputs": [],
    "source": [
     "# Compare Pricing Stats\n",
-    "utilities.match_pricingstats_12(ore, ore_ad, \"Speedup\", False)"
+    "utilities.match_pricingstats_12(ore, ore_ad, \"Speedup\", True)"
    ]
   },
   {
@@ -271,34 +234,101 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## Discussion\n",
-    "Impact of:\n",
+    "### Discussion\n",
     "- Shift Scheme (EQ Spot)\n",
     "- Shift Size (Vega)\n",
     "- Lowering the BarrierLevel in EquityBarrierOption\n",
     "- Lowering the KnockOutLevel in EquityAccumulator\n",
     "- Lowering the KnockOutProfitAmount in FxTARF, introducing varying range strikes or leverages\n",
+    "- Todo: Add CAM"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Sensitivities: CPU vs GPU"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Run with ComputationGraph Enabled\n",
+    "\n",
+    "Check the following is set in **pricingengine_cg.xml**.\n",
+    "- **UseCG** is set to true\n",
+    "- **ExternalDeviceCompatibilityMode** is set to true\n",
+    "\n",
+    "The latter ensures that we use the same ordering of random variables here and in the subsequent runs using external devices."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "params_cg = Parameters()\n",
+    "params_cg.fromFile(\"Input/ore_cg.xml\")\n",
+    "ore_cg = OREApp(params_cg)\n",
+    "ore_cg.run()\n",
+    "utilities.checkErrorsAndRunTime(ore_cg)\n",
+    "#utilities.writeList(ore_bump.getReportNames())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Compare Pricing Stats\n",
+    "utilities.match_pricingstats_12(ore, ore_cg, \"Speedup\", False)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Compare Sensitivities\n",
     "\n",
-    "Todo:\n",
-    "- Add CAM"
+    "If we set pricing engine's ExternalDeviceCompatibilityMode to false here, then we should see zero deviations from the reference run above\n",
+    "\n",
+    "If we set ExternalDeviceCompatibilityMode to true though, then we see differences below due to different ordering of random variates in both runs,\n",
+    "However, this means we use the same ordering in the CG run as in the GPU run below.\n",
+    "\n",
+    "So if you want to check the former, the set ExternalDeviceCompatibilityMode=false first, rerun ore_cg and confirm perfect match of sensitivities."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Compare Sensitivities: This run vs reference \n",
+    "utilities.match_sensi_reports(ore, ore_cg, \"DeltaBump\", \"DeltaCG\", False)"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# Run again using the external GPU device\n",
+    "## Run again using the external GPU device\n",
     "\n",
     "Note that this functionality is work in progress.\n",
     "\n",
-    "Using an external device requires another few changes in pricingengine.xml's EngineParameters section:\n",
+    "Using an external device requires another few changes in **pricingengine_gpu.xml**'s EngineParameters section:\n",
     "- set **UseCG** to true\n",
     "- set **UseExternalComputeDevice** to true\n",
-    "- set **ExternalComputeDevice** to the device name (**OpenCL/Apple/Apple M2 Max** on my machine, a **38 core GPU** device)\n",
+    "- set **ExternalComputeDevice** to the device name, you can choose **BasicCpu/Default/Default** or (on my machine) **OpenCL/Apple/Apple M2 Max**\n",
     "\n",
     "Determine your available compute devices by running the QuantExt test suite with **quantext-test-suite --log_level=message --run_test=\"*/ComputeEnvironmentTest/testEnvironmentInit\"**\n",
     "\n",
-    "You should also see the device name **BasicCpu/Default/Default** which is the CPU and can be used for testing."
+    "**BasicCpu/Default/Default** refers to the CPU itself but treats it as an external device. This is useful for sanity checking the implementation - it should yield a perfect match in the sensitivity comparison below, but only moderate change in performance since we do not paralellize.\n",
+    "\n",
+    "**OpenCL/Apple/Apple M2 Max** is a **38 core GPU** device that should yield a noticeable performance improvement thanks to distributing the calculations across the GPU cores."
    ]
   },
   {