docs/tutorials/05_stability_over_time.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "06120253-65b9-48b5-8925-4565a96933d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import glob\n",
    "import itertools\n",
    "\n",
    "from pathlib import Path\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from qiskit import transpile, QuantumCircuit\n",
    "from qiskit.providers.fake_provider import FakeLima\n",
    "from qiskit.primitives import Estimator, BackendEstimator\n",
    "from qiskit.circuit.random import random_circuit\n",
    "from qiskit.quantum_info import SparsePauliOp\n",
    "\n",
    "import torch\n",
    "from torch.optim import Adam\n",
    "from torch.optim.lr_scheduler import ReduceLROnPlateau\n",
    "from torch.nn.functional import dropout\n",
    "\n",
    "from torch_geometric.nn import (\n",
    "    GCNConv, \n",
    "    TransformerConv, \n",
    "    GATv2Conv, \n",
    "    global_mean_pool, \n",
    "    Linear, \n",
    "    ChebConv, \n",
    "    SAGEConv,\n",
    "    ASAPooling,\n",
    "    dense_diff_pool,\n",
    "    avg_pool_neighbor_x\n",
    ")\n",
    "from torch_geometric.data import Data\n",
    "from torch_geometric.loader import DataLoader\n",
    "from torch_geometric.utils import to_dense_adj, to_dense_batch\n",
    "\n",
    "from tqdm.notebook import tqdm_notebook\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "from blackwater.data.loaders.exp_val import CircuitGraphExpValMitigationDataset\n",
    "from blackwater.data.generators.exp_val import exp_value_generator\n",
    "from blackwater.data.utils import generate_random_pauli_sum_op\n",
    "from blackwater.library.ngem.estimator import ngem\n",
    "from blackwater.data.utils import (\n",
    "    generate_random_pauli_sum_op,\n",
    "    create_estimator_meas_data,\n",
    "    circuit_to_graph_data_json,\n",
    "    get_backend_properties_v1,\n",
    "    encode_pauli_sum_op,\n",
    "    create_meas_data_from_estimators\n",
    ")\n",
    "from blackwater.data.generators.exp_val import ExpValueEntry\n",
    "from blackwater.metrics.improvement_factor import improvement_factor, Trial, Problem\n",
    "\n",
    "from qiskit_aer.primitives import Estimator as AerEstimator\n",
    "from qiskit_aer.noise import NoiseModel\n",
    "import qiskit_aer.noise as noise\n",
    "\n",
    "from qiskit.providers.models import BackendProperties\n",
    "\n",
    "from pprint import pprint\n",
    "import copy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5c379f17-89b8-49f4-b164-27896833c6a8",
   "metadata": {},
   "outputs": [],
   "source": [
    "backend = FakeLima()\n",
    "properties = get_backend_properties_v1(backend)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "0a4bd2e8-1f85-4ba1-8eac-075ace69ac32",
   "metadata": {},
   "outputs": [],
   "source": [
    "# 0. generate data\n",
    "# 1. create orinal noise model (nm0)\n",
    "# 2. train model on nm0 with 10k circuits and get MSE\n",
    "# 3. wiggle noise model (nm1) and check MSE \n",
    "# 4. train model from scratch with 1k, 2k, 3k, 4k, ... and get MSE\n",
    "# 5. plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1815e9fa-0b9a-4391-83ab-ead63e2c92cf",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "1597daed-94d9-4840-928c-c81e255e5995",
   "metadata": {},
   "source": [
    "### Generate data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3c397d0a-1a27-4817-80a2-ceead8d9e907",
   "metadata": {},
   "outputs": [],
   "source": [
    "# # circuits data\n",
    "\n",
    "# N_FILES = 100\n",
    "# N_ENTRIES_PER_FILE = 100\n",
    "\n",
    "# for i in tqdm_notebook(range(N_FILES)):\n",
    "#     with open(f\"./data/stability_over_time_experiment/circuits/qasm_circuits_{i}.txt\", \"w\") as f:\n",
    "#         for _ in range(N_ENTRIES_PER_FILE):\n",
    "#             circuit = transpile(\n",
    "#                 random_circuit(5, 1, measure=True), \n",
    "#                 basis_gates=backend.configuration().basis_gates,\n",
    "#                 optimization_level=0\n",
    "#             )\n",
    "            \n",
    "#             f.write(circuit.qasm())\n",
    "#             f.write(\"=====\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "bb2d69e1-8cfa-446f-8a58-4c58a4706c89",
   "metadata": {},
   "outputs": [],
   "source": [
    "# build estimator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a8e84638-db4c-411a-a5aa-d8afb4acdecd",
   "metadata": {},
   "outputs": [],
   "source": [
    "import requests"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "0e1a14ac-5f7d-42aa-b5b7-e6e196fe5e26",
   "metadata": {},
   "outputs": [],
   "source": [
    "lima_props = json.loads(\n",
    "    requests.get(\n",
    "        \"https://raw.githubusercontent.com/Qiskit/qiskit-terra/3241978340f534b76e4cf0e5e25ed16e2af5cf78/qiskit/providers/fake_provider/backends/lima/props_lima.json\"\n",
    "    ).text\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "a08f0016-f773-4b03-83dc-49dd3494f87e",
   "metadata": {},
   "outputs": [],
   "source": [
    "backend_original = FakeLima()\n",
    "backend_tweaked = FakeLima()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "0dff50a6-982b-4cba-9dce-940a0123249b",
   "metadata": {},
   "outputs": [],
   "source": [
    "backend_original.properties = lambda: BackendProperties.from_dict(lima_props)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "c75126b1-fac9-4804-b420-8bdfb2a6ce22",
   "metadata": {},
   "outputs": [],
   "source": [
    "aer_estimator = AerEstimator()\n",
    "estimator_original = BackendEstimator(backend_original, skip_transpilation=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "ca72e055-9351-4627-b60a-7b3e05d7fab2",
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001B[0;31m---------------------------------------------------------------------------\u001B[0m",
      "\u001B[0;31mKeyboardInterrupt\u001B[0m                         Traceback (most recent call last)",
      "Input \u001B[0;32mIn [7]\u001B[0m, in \u001B[0;36m<cell line: 35>\u001B[0;34m()\u001B[0m\n\u001B[1;32m     31\u001B[0m         new_gates\u001B[38;5;241m.\u001B[39mappend(new_gate)\n\u001B[1;32m     33\u001B[0m     \u001B[38;5;28;01mreturn\u001B[39;00m new_qubits, new_gates\n\u001B[0;32m---> 35\u001B[0m \u001B[43mchange_properties\u001B[49m\u001B[43m(\u001B[49m\u001B[43mlima_props\u001B[49m\u001B[43m)\u001B[49m\n",
      "Input \u001B[0;32mIn [7]\u001B[0m, in \u001B[0;36mchange_properties\u001B[0;34m(lima_props, lower_bound, upper_bound)\u001B[0m\n\u001B[1;32m     12\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m param \u001B[38;5;129;01min\u001B[39;00m qubit:\n\u001B[1;32m     13\u001B[0m     new_param \u001B[38;5;241m=\u001B[39m param\u001B[38;5;241m.\u001B[39mcopy()\n\u001B[0;32m---> 14\u001B[0m     new_param[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mvalue\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[43mparam\u001B[49m[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mvalue\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m*\u001B[39m CHANGE\n\u001B[1;32m     15\u001B[0m     new_qubit\u001B[38;5;241m.\u001B[39mappend(new_param)\n\u001B[1;32m     17\u001B[0m new_qubits\u001B[38;5;241m.\u001B[39mappend(new_qubit)\n",
      "Input \u001B[0;32mIn [7]\u001B[0m, in \u001B[0;36mchange_properties\u001B[0;34m(lima_props, lower_bound, upper_bound)\u001B[0m\n\u001B[1;32m     12\u001B[0m \u001B[38;5;28;01mfor\u001B[39;00m param \u001B[38;5;129;01min\u001B[39;00m qubit:\n\u001B[1;32m     13\u001B[0m     new_param \u001B[38;5;241m=\u001B[39m param\u001B[38;5;241m.\u001B[39mcopy()\n\u001B[0;32m---> 14\u001B[0m     new_param[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mvalue\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m=\u001B[39m \u001B[43mparam\u001B[49m[\u001B[38;5;124m\"\u001B[39m\u001B[38;5;124mvalue\u001B[39m\u001B[38;5;124m\"\u001B[39m] \u001B[38;5;241m*\u001B[39m CHANGE\n\u001B[1;32m     15\u001B[0m     new_qubit\u001B[38;5;241m.\u001B[39mappend(new_param)\n\u001B[1;32m     17\u001B[0m new_qubits\u001B[38;5;241m.\u001B[39mappend(new_qubit)\n",
      "File \u001B[0;32m_pydevd_bundle/pydevd_cython_darwin_310_64.pyx:1179\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_darwin_310_64.SafeCallWrapper.__call__\u001B[0;34m()\u001B[0m\n",
      "File \u001B[0;32m_pydevd_bundle/pydevd_cython_darwin_310_64.pyx:620\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_darwin_310_64.PyDBFrame.trace_dispatch\u001B[0;34m()\u001B[0m\n",
      "File \u001B[0;32m_pydevd_bundle/pydevd_cython_darwin_310_64.pyx:929\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_darwin_310_64.PyDBFrame.trace_dispatch\u001B[0;34m()\u001B[0m\n",
      "File \u001B[0;32m_pydevd_bundle/pydevd_cython_darwin_310_64.pyx:920\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_darwin_310_64.PyDBFrame.trace_dispatch\u001B[0;34m()\u001B[0m\n",
      "File \u001B[0;32m_pydevd_bundle/pydevd_cython_darwin_310_64.pyx:317\u001B[0m, in \u001B[0;36m_pydevd_bundle.pydevd_cython_darwin_310_64.PyDBFrame.do_wait_suspend\u001B[0;34m()\u001B[0m\n",
      "File \u001B[0;32m/Applications/PyCharm.app/Contents/plugins/python/helpers/pydev/pydevd.py:1160\u001B[0m, in \u001B[0;36mPyDB.do_wait_suspend\u001B[0;34m(self, thread, frame, event, arg, send_suspend_message, is_unhandled_exception)\u001B[0m\n\u001B[1;32m   1157\u001B[0m         from_this_thread\u001B[38;5;241m.\u001B[39mappend(frame_id)\n\u001B[1;32m   1159\u001B[0m \u001B[38;5;28;01mwith\u001B[39;00m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_threads_suspended_single_notification\u001B[38;5;241m.\u001B[39mnotify_thread_suspended(thread_id, stop_reason):\n\u001B[0;32m-> 1160\u001B[0m     \u001B[38;5;28;43mself\u001B[39;49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43m_do_wait_suspend\u001B[49m\u001B[43m(\u001B[49m\u001B[43mthread\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mframe\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mevent\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43marg\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43msuspend_type\u001B[49m\u001B[43m,\u001B[49m\u001B[43m \u001B[49m\u001B[43mfrom_this_thread\u001B[49m\u001B[43m)\u001B[49m\n",
      "File \u001B[0;32m/Applications/PyCharm.app/Contents/plugins/python/helpers/pydev/pydevd.py:1175\u001B[0m, in \u001B[0;36mPyDB._do_wait_suspend\u001B[0;34m(self, thread, frame, event, arg, suspend_type, from_this_thread)\u001B[0m\n\u001B[1;32m   1172\u001B[0m             \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39m_call_mpl_hook()\n\u001B[1;32m   1174\u001B[0m         \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mprocess_internal_commands()\n\u001B[0;32m-> 1175\u001B[0m         \u001B[43mtime\u001B[49m\u001B[38;5;241;43m.\u001B[39;49m\u001B[43msleep\u001B[49m\u001B[43m(\u001B[49m\u001B[38;5;241;43m0.01\u001B[39;49m\u001B[43m)\u001B[49m\n\u001B[1;32m   1177\u001B[0m \u001B[38;5;28mself\u001B[39m\u001B[38;5;241m.\u001B[39mcancel_async_evaluation(get_current_thread_id(thread), \u001B[38;5;28mstr\u001B[39m(\u001B[38;5;28mid\u001B[39m(frame)))\n\u001B[1;32m   1179\u001B[0m \u001B[38;5;66;03m# process any stepping instructions\u001B[39;00m\n",
      "\u001B[0;31mKeyboardInterrupt\u001B[0m: "
     ]
    }
   ],
   "source": [
    "def change_properties(lima_props, lower_bound: float = 1.0, upper_bound: float = 3.0):\n",
    "    \"\"\"Changes qubit and gate properties by given bounds.\n",
    "    \n",
    "    Returns:\n",
    "        New qubit and gate properties\n",
    "    \"\"\"\n",
    "    CHANGE = np.random.uniform(lower_bound, upper_bound)\n",
    "    \n",
    "    new_qubits = []\n",
    "    for qubit in lima_props[\"qubits\"]:\n",
    "        new_qubit = []\n",
    "        for param in qubit:\n",
    "            new_param = param.copy()\n",
    "            new_param[\"value\"] = param[\"value\"] * CHANGE\n",
    "            new_qubit.append(new_param)\n",
    "\n",
    "        new_qubits.append(new_qubit)\n",
    "\n",
    "    new_gates = []\n",
    "    for gate in lima_props[\"gates\"]:\n",
    "        new_gate = gate.copy()\n",
    "\n",
    "        new_parameters = []\n",
    "        for param in gate[\"parameters\"]:\n",
    "            new_param = param.copy()\n",
    "            new_param[\"value\"] = param[\"value\"] * CHANGE\n",
    "            new_parameters.append(new_param)\n",
    "\n",
    "        new_gate[\"parameters\"] = new_parameters\n",
    "\n",
    "        new_gates.append(new_gate)\n",
    "        \n",
    "    return new_qubits, new_gates\n",
    "\n",
    "change_properties(lima_props)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "88ad6be0-72d7-4104-ac97-e1668663fce4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": "[{'qubits': [0],\n  'gate': 'id',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00019195510390342677},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'id0'},\n {'qubits': [1],\n  'gate': 'id',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00033064681663890665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'id1'},\n {'qubits': [2],\n  'gate': 'id',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0001807726361160903},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'id2'},\n {'qubits': [3],\n  'gate': 'id',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0003925363414603665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'id3'},\n {'qubits': [4],\n  'gate': 'id',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0006789531879417325},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'id4'},\n {'qubits': [0],\n  'gate': 'rz',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 0}],\n  'name': 'rz0'},\n {'qubits': [1],\n  'gate': 'rz',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 0}],\n  'name': 'rz1'},\n {'qubits': [2],\n  'gate': 'rz',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 0}],\n  'name': 'rz2'},\n {'qubits': [3],\n  'gate': 'rz',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 0}],\n  'name': 'rz3'},\n {'qubits': [4],\n  'gate': 'rz',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 0}],\n  'name': 'rz4'},\n {'qubits': [0],\n  'gate': 'sx',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00019195510390342677},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'sx0'},\n {'qubits': [1],\n  'gate': 'sx',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00033064681663890665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'sx1'},\n {'qubits': [2],\n  'gate': 'sx',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0001807726361160903},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'sx2'},\n {'qubits': [3],\n  'gate': 'sx',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0003925363414603665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'sx3'},\n {'qubits': [4],\n  'gate': 'sx',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0006789531879417325},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'sx4'},\n {'qubits': [0],\n  'gate': 'x',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00019195510390342677},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'x0'},\n {'qubits': [1],\n  'gate': 'x',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.00033064681663890665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'x1'},\n {'qubits': [2],\n  'gate': 'x',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0001807726361160903},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'x2'},\n {'qubits': [3],\n  'gate': 'x',\n  'parameters': [{'date': '2021-03-15T00:11:29-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0003925363414603665},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'x3'},\n {'qubits': [4],\n  'gate': 'x',\n  'parameters': [{'date': '2021-03-15T00:13:23-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.0006789531879417325},\n   {'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 35.55555555555556}],\n  'name': 'x4'},\n {'qubits': [4, 3],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.02043493012013503},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 483.55555555555554}],\n  'name': 'cx4_3'},\n {'qubits': [3, 4],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.02043493012013503},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 519.1111111111111}],\n  'name': 'cx3_4'},\n {'qubits': [0, 1],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:30:08-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.008339674869236618},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 305.77777777777777}],\n  'name': 'cx0_1'},\n {'qubits': [1, 0],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:30:08-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.008339674869236618},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 341.3333333333333}],\n  'name': 'cx1_0'},\n {'qubits': [3, 1],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:24:22-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.009799775939734023},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 462.2222222222222}],\n  'name': 'cx3_1'},\n {'qubits': [1, 3],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:24:22-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.009799775939734023},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 497.77777777777777}],\n  'name': 'cx1_3'},\n {'qubits': [2, 1],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:19:06-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.005961828308163608},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 298.66666666666663}],\n  'name': 'cx2_1'},\n {'qubits': [1, 2],\n  'gate': 'cx',\n  'parameters': [{'date': '2021-03-15T00:19:06-04:00',\n    'name': 'gate_error',\n    'unit': '',\n    'value': 0.005961828308163608},\n   {'date': '2021-03-12T00:36:03-05:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 334.22222222222223}],\n  'name': 'cx1_2'},\n {'qubits': [0],\n  'gate': 'reset',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 5742.222222222222}],\n  'name': 'reset0'},\n {'qubits': [1],\n  'gate': 'reset',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 5742.222222222222}],\n  'name': 'reset1'},\n {'qubits': [2],\n  'gate': 'reset',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 5742.222222222222}],\n  'name': 'reset2'},\n {'qubits': [3],\n  'gate': 'reset',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 5742.222222222222}],\n  'name': 'reset3'},\n {'qubits': [4],\n  'gate': 'reset',\n  'parameters': [{'date': '2021-03-15T00:36:03-04:00',\n    'name': 'gate_length',\n    'unit': 'ns',\n    'value': 5742.222222222222}],\n  'name': 'reset4'}]"
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lima_props['gates']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "82aa6561-b899-4a33-a398-69d08d57d776",
   "metadata": {},
   "outputs": [],
   "source": [
    "tweaked_lima_props = lima_props.copy()\n",
    "\n",
    "new_qubits, new_gates =  change_properties(lima_props, 1.0, 3.0)\n",
    "\n",
    "tweaked_lima_props[\"gates\"] = new_gates\n",
    "tweaked_lima_props[\"qubits\"] = new_qubits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "f0f6a308-14c7-4915-8ae6-fdc6707cf31e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "d613347f-97a9-4313-98bc-c001d8561b1f",
   "metadata": {},
   "outputs": [],
   "source": [
    "shifts_range = range(1, 11, 3)\n",
    "\n",
    "obs = SparsePauliOp([\"ZZZZZ\"])\n",
    "\n",
    "tweaked_estimators = []\n",
    "tweaked_backends = []\n",
    "\n",
    "for shift in shifts_range:\n",
    "    tweaked_lima_props = lima_props.copy()\n",
    "\n",
    "    new_qubits, new_gates =  change_properties(lima_props, shift, shift + 2)\n",
    "\n",
    "    tweaked_lima_props[\"gates\"] = new_gates\n",
    "    tweaked_lima_props[\"qubits\"] = new_qubits\n",
    "    \n",
    "    backend_tweaked.properties = lambda: BackendProperties.from_dict(tweaked_lima_props)\n",
    "    \n",
    "    estimator_tweaked = BackendEstimator(backend_tweaked, skip_transpilation=True)\n",
    "    \n",
    "    tweaked_estimators.append(estimator_tweaked)\n",
    "    tweaked_backends.append(backend_tweaked)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "fadfe94c-61ac-4ae2-a988-0eaa330394e3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "86f43276ae8c405ba1f43565ec97f668",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/20 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "N_FILES = 20\n",
    "\n",
    "for i in tqdm_notebook(range(N_FILES)):\n",
    "    with open(f\"./data/stability_over_time_experiment/circuits/qasm_circuits_{i}.txt\", \"r\") as f:\n",
    "        qasm_strings = f.read().split(\"=====\")[:-1]\n",
    "\n",
    "        with open(f\"./data/stability_over_time_experiment/measurements/results_{i}.txt\", \"w\") as fr:\n",
    "            for qasm in qasm_strings:\n",
    "                circuit = QuantumCircuit.from_qasm_str(qasm)\n",
    "\n",
    "                estimators = [aer_estimator, estimator_original] + tweaked_estimators\n",
    "                \n",
    "                ideal_exp_vals = create_meas_data_from_estimators(\n",
    "                    circuits=circuit, \n",
    "                    observables=obs, \n",
    "                    estimators=estimators,\n",
    "                    shots=10000\n",
    "                )\n",
    "\n",
    "                fr.write(\",\".join([str(x) for x in ideal_exp_vals]))\n",
    "                fr.write(\"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "d70097a9-b7ca-4098-ab2c-a0444a77b32b",
   "metadata": {},
   "outputs": [],
   "source": [
    "n_shifts = len(list(shifts_range))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2bb4c214-62f0-428b-9e90-0c6bdfffe120",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c2d500ea-85e8-4ff8-a57c-8fee3d8cbde5",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "a8256356-cf91-4e00-8c06-92aa5fad3677",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0389df2a38b547c88b0d098bad3bf2fd",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/20 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "observable = encode_pauli_sum_op(SparsePauliOp(\"ZZZZZ\"))\n",
    "\n",
    "for i in tqdm_notebook(range(N_FILES)):\n",
    "    circuits = []\n",
    "    with open(f\"./data/stability_over_time_experiment/circuits/qasm_circuits_{i}.txt\", \"r\") as fc:\n",
    "        qasm_strings = fc.read().split(\"=====\")[:-1]\n",
    "        circuits = [QuantumCircuit.from_qasm_str(qasm)\n",
    "                    for qasm in qasm_strings]\n",
    "\n",
    "    measurements = []\n",
    "                                       \n",
    "    df = pd.read_csv(f\"./data/stability_over_time_experiment/measurements/results_{i}.txt\", header=None)\n",
    "    df.columns = [\"ideal\", \"noise\"] + [f\"shift_{x}\" for x in range(n_shifts)]\n",
    "\n",
    "    for row in df.iterrows():\n",
    "        measurements.append([\n",
    "            row[1][\"ideal\"],\n",
    "            row[1][\"noise\"]\n",
    "        ] + [\n",
    "            row[1][f\"shift_{x}\"] for x in range(n_shifts)\n",
    "        ])    \n",
    "                       \n",
    "    original_entries = []\n",
    "    tweaked_entries = {\n",
    "        x: [] for x in range(n_shifts)\n",
    "    }\n",
    "\n",
    "    for mments, circuit in zip(measurements, circuits):\n",
    "        ideal_ev, original_ev, tweaked_meas = mments[0], mments[1], mments[2:]\n",
    "         \n",
    "        original_graph_data = circuit_to_graph_data_json(\n",
    "            circuit=circuit,\n",
    "            properties=get_backend_properties_v1(backend_original),\n",
    "            use_qubit_features=True,\n",
    "            use_gate_features=True,\n",
    "        )\n",
    "        original_entry = ExpValueEntry(\n",
    "            circuit_graph=original_graph_data,\n",
    "            observable=observable,\n",
    "            ideal_exp_value=ideal_ev,\n",
    "            noisy_exp_values=[original_ev],\n",
    "            circuit_depth=circuit.depth()\n",
    "        ) \n",
    "        \n",
    "        original_entries.append(original_entry.to_dict())\n",
    "        \n",
    "        for idx, (tweaked_ev, tweaked_bcknd) in enumerate(zip(tweaked_meas, tweaked_backends)):\n",
    "            tweaked_graph_data = circuit_to_graph_data_json(\n",
    "                circuit=circuit,\n",
    "                properties=get_backend_properties_v1(tweaked_bcknd),\n",
    "                use_qubit_features=True,\n",
    "                use_gate_features=True,\n",
    "            )\n",
    "\n",
    "            tweaked_entry = ExpValueEntry(\n",
    "                circuit_graph=tweaked_graph_data,\n",
    "                observable=observable,\n",
    "                ideal_exp_value=ideal_ev,\n",
    "                noisy_exp_values=[tweaked_ev],\n",
    "                circuit_depth=circuit.depth()\n",
    "            )\n",
    "            \n",
    "            tweaked_entries[idx].append(tweaked_entry.to_dict())\n",
    "             \n",
    "    save_file_path = f\"./data/stability_over_time_experiment/pyg_data/original/{i}.json\"\n",
    "    with open(save_file_path, \"w\") as json_file:\n",
    "        json.dump(original_entries, json_file)\n",
    "             \n",
    "    for idx, entries in tweaked_entries.items():\n",
    "        folder = f\"./data/stability_over_time_experiment/pyg_data/tweaked_{idx}\"\n",
    "        Path(folder).mkdir(parents=True, exist_ok=True)\n",
    "        save_file_path = f\"{folder}/{i}.json\"\n",
    "        \n",
    "        with open(save_file_path, \"w\") as json_file:\n",
    "            json.dump(entries, json_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e5fe54b5-f276-468e-9b8d-0bf12e51e235",
   "metadata": {},
   "source": [
    "###  Train networks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "id": "79272cff-f027-4cc6-b0f9-9cf54bca64f8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DataBatch(x=[286, 22], edge_index=[2, 585], edge_attr=[299, 3], y=[10, 1], observable=[10, 1, 21], circuit_depth=[10, 1], noisy_0=[10, 1], batch=[286], ptr=[11])\n"
     ]
    }
   ],
   "source": [
    "max_files_to_use = 20\n",
    "train_size = 15\n",
    "\n",
    "data_paths = glob.glob(\"./data/stability_over_time_experiment/pyg_data/original/**.json\")[:max_files_to_use]\n",
    "\n",
    "BATCH_SIZE = 10\n",
    "\n",
    "train_loader = DataLoader(\n",
    "    CircuitGraphExpValMitigationDataset(\n",
    "        data_paths[:train_size],\n",
    "    ),\n",
    "    batch_size=BATCH_SIZE,\n",
    "    shuffle=True\n",
    ")\n",
    "\n",
    "val_loader = DataLoader(\n",
    "    CircuitGraphExpValMitigationDataset(\n",
    "        data_paths[train_size:],\n",
    "    ),\n",
    "    batch_size=BATCH_SIZE,\n",
    "    shuffle=False\n",
    ")\n",
    "\n",
    "for data in train_loader:\n",
    "    print(data)\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "86787fad-794a-4aa3-b2b4-661968124fb8",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 188,
   "id": "68a5331d-745c-4233-8084-d583be4ad850",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ExpValCircuitGraphModel(torch.nn.Module):\n",
    "    def __init__(\n",
    "        self,\n",
    "        n_qubits: int,\n",
    "        num_node_features: int,\n",
    "        hidden_channels: int\n",
    "    ):\n",
    "        super().__init__()\n",
    "        self.conv1 = GCNConv(num_node_features, hidden_channels)\n",
    "        self.conv2 = GCNConv(hidden_channels, hidden_channels)\n",
    "        self.conv3 = GCNConv(hidden_channels, 1)\n",
    "        \n",
    "        self.cheb_conv1 = ChebConv(num_node_features, hidden_channels, K=3)\n",
    "        self.cheb_conv2 = ChebConv(hidden_channels, 1, K=2)\n",
    "\n",
    "        self.sage_conv1 = SAGEConv(num_node_features, hidden_channels)\n",
    "        self.sage_conv2 = SAGEConv(hidden_channels, 1)\n",
    "        \n",
    "        self.obs_seq = torch.nn.Sequential(\n",
    "            Linear(n_qubits * 4 + 1, hidden_channels),\n",
    "            torch.nn.Dropout(0.2),\n",
    "            Linear(hidden_channels, 1)\n",
    "        )\n",
    "        \n",
    "        self.body_seq = torch.nn.Sequential(\n",
    "            Linear(6, hidden_channels),\n",
    "            Linear(hidden_channels, 1)\n",
    "        )\n",
    "\n",
    "    def forward(self, \n",
    "                exp_value, observable, \n",
    "                circuit_depth, nodes, \n",
    "                edge_index, batch):\n",
    "        # GCN\n",
    "        graph = self.conv1(nodes, edge_index).relu()\n",
    "        graph = dropout(graph, training=self.training, p=0.1)\n",
    "        graph = self.conv2(graph, edge_index).relu()\n",
    "        graph = dropout(graph, training=self.training, p=0.1)\n",
    "        graph = self.conv3(graph, edge_index)\n",
    "        graph = global_mean_pool(graph, batch)\n",
    "        \n",
    "        # Cheb\n",
    "        cheb = self.cheb_conv1(nodes, edge_index).relu()\n",
    "        cheb = dropout(cheb, training=self.training, p=0.2)\n",
    "        cheb = self.cheb_conv2(cheb, edge_index)\n",
    "        cheb = global_mean_pool(cheb, batch)\n",
    "        \n",
    "        # Sage\n",
    "        sage = self.sage_conv1(nodes, edge_index).relu()\n",
    "        sage = dropout(sage, training=self.training, p=0.2)\n",
    "        sage = self.sage_conv2(sage, edge_index)\n",
    "        sage = global_mean_pool(sage, batch)\n",
    "        \n",
    "        obs = self.obs_seq(observable)\n",
    "        obs = torch.mean(obs, dim=1)\n",
    "        \n",
    "        merge = torch.cat((\n",
    "            graph, \n",
    "            cheb,\n",
    "            sage,\n",
    "            obs,\n",
    "            circuit_depth,\n",
    "            exp_value\n",
    "        ), dim=1)\n",
    "\n",
    "        return self.body_seq(merge)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "id": "a38593ce-25d2-455b-adc0-730f4231eb31",
   "metadata": {},
   "outputs": [],
   "source": [
    "GLOBAL_LR = 0.0001\n",
    "N_EPOCHS = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 229,
   "id": "168f6fd4-8e60-440a-b81e-e90dfd585a89",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = ExpValCircuitGraphModel(\n",
    "    n_qubits=5,\n",
    "    num_node_features=22, \n",
    "    hidden_channels=10\n",
    ")\n",
    "optimizer = Adam(model.parameters(), lr=GLOBAL_LR)\n",
    "criterion = torch.nn.MSELoss()\n",
    "scheduler = ReduceLROnPlateau(optimizer, \n",
    "                              'min', \n",
    "                              factor=0.5, \n",
    "                              patience=20, \n",
    "                              verbose=True, \n",
    "                              min_lr=0.000001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 230,
   "id": "b829b644-0866-4fef-b53d-cb292c6dddf1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ca06891c7c374924ae982b4f1c0b2fbe",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Model training:   0%|          | 0/20 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "min_valid_loss = np.inf\n",
    "\n",
    "train_losses = []\n",
    "val_losses = []\n",
    "\n",
    "progress = tqdm_notebook(range(N_EPOCHS*2), desc='Model training', leave=True)\n",
    "for epoch in progress:\n",
    "    train_loss = 0.0\n",
    "    model.train()\n",
    "    for i, data in enumerate(train_loader):\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        out = model(\n",
    "            data.noisy_0, \n",
    "            data.observable, \n",
    "            data.circuit_depth,\n",
    "            data.x, \n",
    "            data.edge_index, \n",
    "            data.batch\n",
    "        )\n",
    "        loss = criterion(out, data.y)        \n",
    "        \n",
    "        train_loss += loss.item()\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "                \n",
    "    valid_loss = 0.0\n",
    "    model.eval()\n",
    "    for i, data in enumerate(val_loader):\n",
    "        out = model(data.noisy_0, data.observable, data.circuit_depth, data.x, data.edge_index, data.batch)\n",
    "        loss = criterion(out, data.y)        \n",
    "\n",
    "        valid_loss += loss.item()\n",
    "\n",
    "    scheduler.step(valid_loss)\n",
    "\n",
    "    if epoch >= 1:    \n",
    "        train_losses.append(train_loss / len(train_loader))\n",
    "        val_losses.append(valid_loss / len(val_loader))\n",
    "\n",
    "        progress.set_description(f\"{round(train_losses[-1], 5)}, {round(val_losses[-1], 5)}\")\n",
    "        progress.refresh()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 231,
   "id": "d2213ca3-0b1e-4e1b-9f6b-561fae5a89f9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(train_losses, label=\"train_loss\")\n",
    "plt.plot(val_losses, label=\"val_loss\")\n",
    "# plt.yscale('log')\n",
    "\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 194,
   "id": "5d5ba75c-9547-4662-8d99-f3a85090c693",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5287716485733209"
      ]
     },
     "execution_count": 194,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "final_error = val_losses[-1]\n",
    "final_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 195,
   "id": "8ea3ac17-379d-4bd2-864d-e5d4b5179ddc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# plt.plot(val_losses)\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e7a27d39-08de-48d6-9ea8-aa1ed61ae84c",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 196,
   "id": "d13725ae-9168-43e8-9489-0d90c08cec5e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def post_train_model(model, train_loader, val_loader):\n",
    "    optimizer = Adam(model.parameters(), lr=GLOBAL_LR)\n",
    "    criterion = torch.nn.MSELoss()\n",
    "    scheduler = ReduceLROnPlateau(optimizer, \n",
    "                                  'min', \n",
    "                                  factor=0.5, \n",
    "                                  patience=20, \n",
    "                                  verbose=True, \n",
    "                                  min_lr=0.000001)\n",
    "    \n",
    "    min_valid_loss = np.inf\n",
    "\n",
    "    train_losses = []\n",
    "    val_losses = []\n",
    "\n",
    "    progress = tqdm_notebook(range(N_EPOCHS), desc='Model training', leave=True)\n",
    "    for epoch in progress:\n",
    "        train_loss = 0.0\n",
    "        model.train()\n",
    "        for i, data in enumerate(train_loader):\n",
    "            optimizer.zero_grad()\n",
    "\n",
    "            out = model(\n",
    "                data.noisy_0, \n",
    "                data.observable, \n",
    "                data.circuit_depth,\n",
    "                data.x, \n",
    "                data.edge_index, \n",
    "                data.batch\n",
    "            )\n",
    "            loss = criterion(out, data.y)        \n",
    "\n",
    "            train_loss += loss.item()\n",
    "\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "\n",
    "        valid_loss = 0.0\n",
    "        model.eval()\n",
    "        for i, data in enumerate(val_loader):\n",
    "            out = model(data.noisy_0, data.observable, data.circuit_depth, data.x, data.edge_index, data.batch)\n",
    "            loss = criterion(out, data.y)        \n",
    "\n",
    "            valid_loss += loss.item()\n",
    "\n",
    "        scheduler.step(valid_loss)\n",
    "\n",
    "        if epoch >= 1:    \n",
    "            train_losses.append(train_loss / len(train_loader))\n",
    "            val_losses.append(valid_loss / len(val_loader))\n",
    "\n",
    "            progress.set_description(f\"{round(train_losses[-1], 5)}, {round(val_losses[-1], 5)}\")\n",
    "            progress.refresh()\n",
    "    \n",
    "    return val_losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2ae98c35-b1a9-4735-846a-0cc7293664ff",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "id": "c04b89eb-abda-4ee5-bfa0-ed5e4049ea5e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "aaec19d9c63a44b5a6960bccc148b505",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/4 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "debf27ce33f14001b6e7e0028e28fd35",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Model training:   0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5486e3fe629f4a5aa0170cff1d379902",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Model training:   0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7c3d19ad81e041d1adf6a7d2cd35e8f2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Model training:   0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d294da57f835434ba0b4cc3ce58446d0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Model training:   0%|          | 0/10 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "tweaked_losses = []\n",
    "\n",
    "for idx in tqdm_notebook(range(n_shifts)):\n",
    "    tweaked_data_paths = glob.glob(f\"./data/stability_over_time_experiment/pyg_data/tweaked_{idx}/**.json\")[:max_files_to_use]\n",
    "    \n",
    "    tweaked_train_loader = DataLoader(\n",
    "        CircuitGraphExpValMitigationDataset(\n",
    "            tweaked_data_paths[:train_size],\n",
    "        ),\n",
    "        batch_size=BATCH_SIZE,\n",
    "        shuffle=True\n",
    "    )\n",
    "\n",
    "    tweaked_val_loader = DataLoader(\n",
    "        CircuitGraphExpValMitigationDataset(\n",
    "            tweaked_data_paths[train_size:],\n",
    "        ),\n",
    "        batch_size=BATCH_SIZE,\n",
    "        shuffle=False\n",
    "    )\n",
    "    \n",
    "    model_copy = copy.deepcopy(model)\n",
    "\n",
    "    res = post_train_model(model_copy, tweaked_train_loader, tweaked_val_loader)\n",
    "    \n",
    "    tweaked_losses.append(res)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ec93ab68-c0f6-4120-92ec-9fe007b893a4",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "id": "478095b3-5466-4db4-ae36-22ec27b9ff11",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_from_scratch = ExpValCircuitGraphModel(\n",
    "    n_qubits=5,\n",
    "    num_node_features=22, \n",
    "    hidden_channels=10\n",
    ")\n",
    "optimizer = Adam(model_from_scratch.parameters(), lr=GLOBAL_LR)\n",
    "criterion = torch.nn.MSELoss()\n",
    "scheduler = ReduceLROnPlateau(optimizer, \n",
    "                              'min', \n",
    "                              factor=0.5, \n",
    "                              patience=20, \n",
    "                              verbose=True, \n",
    "                              min_lr=0.000001)\n",
    "\n",
    "min_valid_loss = np.inf\n",
    "\n",
    "train_losses = []\n",
    "val_losses = []\n",
    "\n",
    "progress = tqdm_notebook(range(N_EPOCHS), desc='Model training', leave=True)\n",
    "for epoch in progress:\n",
    "    train_loss = 0.0\n",
    "    model_from_scratch.train()\n",
    "    for i, data in enumerate(train_loader):\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        out = model_from_scratch(\n",
    "            data.noisy_0, \n",
    "            data.observable, \n",
    "            data.circuit_depth,\n",
    "            data.x, \n",
    "            data.edge_index, \n",
    "            data.batch\n",
    "        )\n",
    "        loss = criterion(out, data.y)        \n",
    "        \n",
    "        train_loss += loss.item()\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "                \n",
    "    valid_loss = 0.0\n",
    "    model_from_scratch.eval()\n",
    "    for i, data in enumerate(val_loader):\n",
    "        out = model_from_scratch(data.noisy_0, data.observable, data.circuit_depth, data.x, data.edge_index, data.batch)\n",
    "        loss = criterion(out, data.y)        \n",
    "\n",
    "        valid_loss += loss.item()\n",
    "\n",
    "    scheduler.step(valid_loss)\n",
    "\n",
    "    if epoch >= 1:    \n",
    "        train_losses.append(train_loss / len(train_loader))\n",
    "        val_losses.append(valid_loss / len(val_loader))\n",
    "        \n",
    "        if tweaked_losses[-1][-1] > val_losses[-1]:\n",
    "            break\n",
    "\n",
    "        progress.set_description(f\"{round(train_losses[-1], 5)}, {round(val_losses[-1], 5)}\")\n",
    "        progress.refresh()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 303,
   "id": "b82740ff-83e5-42c0-b997-10a8057fd186",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[9.450379936085191,\n",
       " 4.392636608522992,\n",
       " 2.9582744728687196,\n",
       " 2.1951593038647674,\n",
       " 1.6635156352852665,\n",
       " 1.299276543495267,\n",
       " 1.025503741378008,\n",
       " 0.8324147053236185,\n",
       " 0.6842359814879506,\n",
       " 0.573568127980066,\n",
       " 0.49732134345658996,\n",
       " 0.441268450229667,\n",
       " 0.4012047807144564,\n",
       " 0.370573227072871,\n",
       " 0.3477846480386202,\n",
       " 0.32782895509586774,\n",
       " 0.3111303766799528,\n",
       " 0.2966482018315515,\n",
       " 0.282635310292244]"
      ]
     },
     "execution_count": 303,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 300,
   "id": "b9d297ef-6a57-483b-a920-014427fbf36a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(val_losses, label=\"Train from scratch\")\n",
    "\n",
    "for idx, tweaked_loss in enumerate(tweaked_losses):\n",
    "    plt.plot(tweaked_loss, label=f\"Turning from {shifts_range[idx]*10}% parameters shift\", color=(0.5 - 0.1*idx, 0.5, 0.5))\n",
    "\n",
    "plt.axhline(y=tweaked_losses[0][-1], color='r', linestyle='-', lw=1)\n",
    "\n",
    "plt.xlabel(\"Epochs\")\n",
    "plt.ylabel(\"Val loss\")\n",
    "plt.legend()\n",
    "# plt.yscale(\"log\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 306,
   "id": "be662922-af3d-4822-be64-8ad8b9a463ab",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(\"./results/stability_results.json\", \"w\") as f:\n",
    "    json.dump({\n",
    "        \"from_scratch\": val_losses,\n",
    "        \"tuned\": tweaked_losses\n",
    "    }, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4b80893d-718c-452d-b2ad-94ad56d888e2",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1661ddd6-36cc-43c6-997e-498dae3830af",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4e1805aa-79ea-4357-ad5f-d18e70888fa4",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dfb11947-a0ce-4a0a-a70d-e7a301b72302",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}