Decal_VSP.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 126,
   "id": "8dc604cc-e521-47f0-b2ee-74a764c95116",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import numpy as np\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "import random\n",
    "import os\n",
    "import shutil\n",
    "from keci_r_MRR import Keci_exp\n",
    "from matplotlib.backends.backend_pdf import PdfPages\n",
    "import matplotlib.pyplot as plt\n",
    "random.seed(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "23a5fda3-6e66-4b60-9d87-05bdfd6da9a8",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "300\n"
     ]
    }
   ],
   "source": [
    "#Generate left out data used for prediction (This has been done turn by turn for each dataset)\n",
    "\n",
    "values_to_include =[(4,1,2)] #[(0,0,4), (0,4,4), (1,2,0), (4,1,2), (4,2,1), (0,1,0), (0,0,3)]\n",
    "\n",
    "pred_data = {}\n",
    "occurrences_count = {value: 0 for value in values_to_include}\n",
    "\n",
    "for key, value in data_dict.items():\n",
    "    if value in values_to_include and occurrences_count[value] < 300:\n",
    "        pred_data[key] = value\n",
    "        occurrences_count[value] += 1\n",
    "\n",
    "# Remove the extracted elements from the original dictionary\n",
    "for key in pred_data:\n",
    "    del data_dict[key]\n",
    "\n",
    "print(len(pred_data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "ef16b8a6-24aa-4ead-a732-07cfe0b4791d",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "#Generate test data\n",
    "\n",
    "random_keys = random.sample(list(data_dict.keys()), 200)\n",
    "test_data = {key: data_dict.pop(key) for key in random_keys}\n",
    "print(len(test_data))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b43ea650-51d4-430c-bb12-4eb59fc475b8",
   "metadata": {},
   "source": [
    "# Modelling"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb79eaa1-cfcc-4385-b743-51cb4ff4dfbe",
   "metadata": {},
   "source": [
    "## LSTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "ab3fd043-ed25-4d51-93c6-ac7768177c0d",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_2230923/3573386890.py:44: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [1/150], Loss: 753.6534\n",
      "Epoch [2/150], Loss: 668.4742\n",
      "Epoch [3/150], Loss: 606.5065\n",
      "Epoch [4/150], Loss: 563.6023\n",
      "Epoch [5/150], Loss: 526.4786\n",
      "Epoch [6/150], Loss: 490.1487\n",
      "Epoch [7/150], Loss: 453.0331\n",
      "Epoch [8/150], Loss: 415.1502\n",
      "Epoch [9/150], Loss: 376.9977\n",
      "Epoch [10/150], Loss: 339.7404\n",
      "Epoch [11/150], Loss: 304.6809\n",
      "Epoch [12/150], Loss: 272.4497\n",
      "Epoch [13/150], Loss: 242.6075\n",
      "Epoch [14/150], Loss: 215.2803\n",
      "Epoch [15/150], Loss: 190.8946\n",
      "Epoch [16/150], Loss: 169.6428\n",
      "Epoch [17/150], Loss: 151.0203\n",
      "Epoch [18/150], Loss: 135.0970\n",
      "Epoch [19/150], Loss: 121.4626\n",
      "Epoch [20/150], Loss: 109.8998\n",
      "Epoch [21/150], Loss: 100.0712\n",
      "Epoch [22/150], Loss: 91.5758\n",
      "Epoch [23/150], Loss: 84.1502\n",
      "Epoch [24/150], Loss: 77.6849\n",
      "Epoch [25/150], Loss: 71.9474\n",
      "Epoch [26/150], Loss: 66.7828\n",
      "Epoch [27/150], Loss: 62.1701\n",
      "Epoch [28/150], Loss: 57.9547\n",
      "Epoch [29/150], Loss: 54.1030\n",
      "Epoch [30/150], Loss: 50.5783\n",
      "Epoch [31/150], Loss: 47.3404\n",
      "Epoch [32/150], Loss: 44.3534\n",
      "Epoch [33/150], Loss: 41.6056\n",
      "Epoch [34/150], Loss: 39.0558\n",
      "Epoch [35/150], Loss: 36.7046\n",
      "Epoch [36/150], Loss: 34.5118\n",
      "Epoch [37/150], Loss: 32.4816\n",
      "Epoch [38/150], Loss: 30.5894\n",
      "Epoch [39/150], Loss: 28.8262\n",
      "Epoch [40/150], Loss: 27.1715\n",
      "Epoch [41/150], Loss: 25.6217\n",
      "Epoch [42/150], Loss: 24.1665\n",
      "Epoch [43/150], Loss: 22.7930\n",
      "Epoch [44/150], Loss: 21.4914\n",
      "Epoch [45/150], Loss: 20.2669\n",
      "Epoch [46/150], Loss: 19.1002\n",
      "Epoch [47/150], Loss: 17.9944\n",
      "Epoch [48/150], Loss: 16.9419\n",
      "Epoch [49/150], Loss: 15.9400\n",
      "Epoch [50/150], Loss: 14.9859\n",
      "Epoch [51/150], Loss: 14.0767\n",
      "Epoch [52/150], Loss: 13.2080\n",
      "Epoch [53/150], Loss: 12.3815\n",
      "Epoch [54/150], Loss: 11.5904\n",
      "Epoch [55/150], Loss: 10.8386\n",
      "Epoch [56/150], Loss: 10.1204\n",
      "Epoch [57/150], Loss: 9.4412\n",
      "Epoch [58/150], Loss: 8.7994\n",
      "Epoch [59/150], Loss: 8.1973\n",
      "Epoch [60/150], Loss: 7.6343\n",
      "Epoch [61/150], Loss: 7.1108\n",
      "Epoch [62/150], Loss: 6.6258\n",
      "Epoch [63/150], Loss: 6.1768\n",
      "Epoch [64/150], Loss: 5.7647\n",
      "Epoch [65/150], Loss: 5.3868\n",
      "Epoch [66/150], Loss: 5.0417\n",
      "Epoch [67/150], Loss: 4.7279\n",
      "Epoch [68/150], Loss: 4.4426\n",
      "Epoch [69/150], Loss: 4.1845\n",
      "Epoch [70/150], Loss: 3.9509\n",
      "Epoch [71/150], Loss: 3.7393\n",
      "Epoch [72/150], Loss: 3.5479\n",
      "Epoch [73/150], Loss: 3.3738\n",
      "Epoch [74/150], Loss: 3.2157\n",
      "Epoch [75/150], Loss: 3.0717\n",
      "Epoch [76/150], Loss: 2.9398\n",
      "Epoch [77/150], Loss: 2.8190\n",
      "Epoch [78/150], Loss: 2.7079\n",
      "Epoch [79/150], Loss: 2.6056\n",
      "Epoch [80/150], Loss: 2.5110\n",
      "Epoch [81/150], Loss: 2.4232\n",
      "Epoch [82/150], Loss: 2.3415\n",
      "Epoch [83/150], Loss: 2.2655\n",
      "Epoch [84/150], Loss: 2.1943\n",
      "Epoch [85/150], Loss: 2.1278\n",
      "Epoch [86/150], Loss: 2.0654\n",
      "Epoch [87/150], Loss: 2.0066\n",
      "Epoch [88/150], Loss: 1.9513\n",
      "Epoch [89/150], Loss: 1.8991\n",
      "Epoch [90/150], Loss: 1.8497\n",
      "Epoch [91/150], Loss: 1.8028\n",
      "Epoch [92/150], Loss: 1.7584\n",
      "Epoch [93/150], Loss: 1.7162\n",
      "Epoch [94/150], Loss: 1.6760\n",
      "Epoch [95/150], Loss: 1.6378\n",
      "Epoch [96/150], Loss: 1.6013\n",
      "Epoch [97/150], Loss: 1.5665\n",
      "Epoch [98/150], Loss: 1.5332\n",
      "Epoch [99/150], Loss: 1.5013\n",
      "Epoch [100/150], Loss: 1.4708\n",
      "Epoch [101/150], Loss: 1.4415\n",
      "Epoch [102/150], Loss: 1.4133\n",
      "Epoch [103/150], Loss: 1.3863\n",
      "Epoch [104/150], Loss: 1.3603\n",
      "Epoch [105/150], Loss: 1.3353\n",
      "Epoch [106/150], Loss: 1.3112\n",
      "Epoch [107/150], Loss: 1.2880\n",
      "Epoch [108/150], Loss: 1.2657\n",
      "Epoch [109/150], Loss: 1.2440\n",
      "Epoch [110/150], Loss: 1.2232\n",
      "Epoch [111/150], Loss: 1.2030\n",
      "Epoch [112/150], Loss: 1.1835\n",
      "Epoch [113/150], Loss: 1.1646\n",
      "Epoch [114/150], Loss: 1.1463\n",
      "Epoch [115/150], Loss: 1.1286\n",
      "Epoch [116/150], Loss: 1.1114\n",
      "Epoch [117/150], Loss: 1.0947\n",
      "Epoch [118/150], Loss: 1.0786\n",
      "Epoch [119/150], Loss: 1.0629\n",
      "Epoch [120/150], Loss: 1.0476\n",
      "Epoch [121/150], Loss: 1.0328\n",
      "Epoch [122/150], Loss: 1.0184\n",
      "Epoch [123/150], Loss: 1.0044\n",
      "Epoch [124/150], Loss: 0.9908\n",
      "Epoch [125/150], Loss: 0.9776\n",
      "Epoch [126/150], Loss: 0.9647\n",
      "Epoch [127/150], Loss: 0.9521\n",
      "Epoch [128/150], Loss: 0.9399\n",
      "Epoch [129/150], Loss: 0.9280\n",
      "Epoch [130/150], Loss: 0.9163\n",
      "Epoch [131/150], Loss: 0.9050\n",
      "Epoch [132/150], Loss: 0.8939\n",
      "Epoch [133/150], Loss: 0.8831\n",
      "Epoch [134/150], Loss: 0.8726\n",
      "Epoch [135/150], Loss: 0.8623\n",
      "Epoch [136/150], Loss: 0.8522\n",
      "Epoch [137/150], Loss: 0.8424\n",
      "Epoch [138/150], Loss: 0.8328\n",
      "Epoch [139/150], Loss: 0.8234\n",
      "Epoch [140/150], Loss: 0.8142\n",
      "Epoch [141/150], Loss: 0.8052\n",
      "Epoch [142/150], Loss: 0.7964\n",
      "Epoch [143/150], Loss: 0.7878\n",
      "Epoch [144/150], Loss: 0.7794\n",
      "Epoch [145/150], Loss: 0.7712\n",
      "Epoch [146/150], Loss: 0.7631\n",
      "Epoch [147/150], Loss: 0.7552\n",
      "Epoch [148/150], Loss: 0.7474\n",
      "Epoch [149/150], Loss: 0.7398\n",
      "Epoch [150/150], Loss: 0.7324\n"
     ]
    }
   ],
   "source": [
    "class MyMultiClassClassifier_lstm(nn.Module):\n",
    "    def __init__(self, num_classes):\n",
    "        super(MyMultiClassClassifier_lstm, self).__init__()\n",
    "        self.lstm = nn.LSTM(5000,256,num_layers=2)\n",
    "        \n",
    "        self.fc3_p = nn.Linear(256, 3*num_classes)\n",
    "        # self.fc3_q = nn.Linear(256, num_classes)\n",
    "        # self.fc3_r = nn.Linear(256, num_classes)\n",
    "\n",
    "    def forward(self, x):\n",
    "        output1, (hn1, cn1) = self.lstm(x.reshape(1,48,5000))\n",
    "        output1 = output1.sum(1)\n",
    "        output = torch.relu(output1)\n",
    "        \n",
    "        output = self.fc3_p(output)\n",
    "\n",
    "        return output.reshape(-1,5,3)\n",
    "\n",
    "num_classes = 5\n",
    "input_size = 5000 * 48\n",
    "\n",
    "multi_class_classifier_lstm = MyMultiClassClassifier_lstm(num_classes)\n",
    "\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.SGD(multi_class_classifier_lstm.parameters(), lr=0.001)\n",
    "\n",
    "loss_lstm = []\n",
    "num_epochs = 150\n",
    "\n",
    "for epoch in range(num_epochs):\n",
    "    total_loss = 0  # to monitor the loss\n",
    "\n",
    "    for D_i, target_triple in data_dict.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target_p, target_q, target_r = target_triple\n",
    "        \n",
    "        p_q_r = multi_class_classifier_lstm(D_i_flat)\n",
    "       \n",
    "        loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n",
    "        \n",
    "        total_loss +=loss.item()\n",
    "        \n",
    "        \n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "      \n",
    "        \n",
    "\n",
    "    if (epoch + 1) % 1 == 0:\n",
    "        print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {total_loss:.4f}')\n",
    "    loss_lstm.append(total_loss)\n",
    "        \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "id": "928531a3-0cb4-4f8a-bde7-94941244e8b6",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "150"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(loss_lstm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "id": "7c9ff78b-f585-45e5-90c2-82b250fe1f9c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 0, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 1), Actual: (0, 0, 3)\n",
      "Predicted: (1, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 4, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "the model is able to predict with 44.5 percent\n"
     ]
    }
   ],
   "source": [
    "        \n",
    "# Test the model with train data first\n",
    "# data = data_dict\n",
    "\n",
    "data = test_data\n",
    "#data = pred_data\n",
    "with torch.no_grad():\n",
    "    s = 0 \n",
    "    for D_i, target in data.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target = torch.tensor(target)\n",
    "        outputs = multi_class_classifier_lstm(D_i_flat)\n",
    "    \n",
    "        p,q,r = outputs.argmax(1).squeeze(0)\n",
    "        \n",
    "        p_q_r = (p.item(),q.item(),r.item())\n",
    "        \n",
    "        if p_q_r == tuple(target.numpy()):\n",
    "            s += 1 \n",
    "            \n",
    "        print(f'Predicted: {p_q_r}, Actual: {tuple(target.numpy())}')\n",
    "print(f'the model is able to predict with { s/(len(data))*100} percent')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "97a78e65-97b6-4644-9062-e25d747029d6",
   "metadata": {},
   "source": [
    "## GRU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "id": "0ffdd797-bcd2-4048-8580-25a9e818f2ad",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_2230923/4221610534.py:39: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [1/150], Loss: 441.9413\n",
      "Epoch [2/150], Loss: 374.5903\n",
      "Epoch [3/150], Loss: 307.4259\n",
      "Epoch [4/150], Loss: 250.7930\n",
      "Epoch [5/150], Loss: 203.4949\n",
      "Epoch [6/150], Loss: 166.1220\n",
      "Epoch [7/150], Loss: 136.9144\n",
      "Epoch [8/150], Loss: 114.7943\n",
      "Epoch [9/150], Loss: 98.2594\n",
      "Epoch [10/150], Loss: 85.5097\n",
      "Epoch [11/150], Loss: 75.3745\n",
      "Epoch [12/150], Loss: 67.1343\n",
      "Epoch [13/150], Loss: 60.3429\n",
      "Epoch [14/150], Loss: 54.6755\n",
      "Epoch [15/150], Loss: 49.8317\n",
      "Epoch [16/150], Loss: 45.5790\n",
      "Epoch [17/150], Loss: 41.8163\n",
      "Epoch [18/150], Loss: 38.4762\n",
      "Epoch [19/150], Loss: 35.4721\n",
      "Epoch [20/150], Loss: 32.7683\n",
      "Epoch [21/150], Loss: 30.3389\n",
      "Epoch [22/150], Loss: 28.0900\n",
      "Epoch [23/150], Loss: 26.0468\n",
      "Epoch [24/150], Loss: 24.1656\n",
      "Epoch [25/150], Loss: 22.4410\n",
      "Epoch [26/150], Loss: 20.8776\n",
      "Epoch [27/150], Loss: 19.4340\n",
      "Epoch [28/150], Loss: 18.1080\n",
      "Epoch [29/150], Loss: 16.9081\n",
      "Epoch [30/150], Loss: 15.7891\n",
      "Epoch [31/150], Loss: 14.7580\n",
      "Epoch [32/150], Loss: 13.7967\n",
      "Epoch [33/150], Loss: 12.9066\n",
      "Epoch [34/150], Loss: 12.0732\n",
      "Epoch [35/150], Loss: 11.3049\n",
      "Epoch [36/150], Loss: 10.5774\n",
      "Epoch [37/150], Loss: 9.8950\n",
      "Epoch [38/150], Loss: 9.2477\n",
      "Epoch [39/150], Loss: 8.6313\n",
      "Epoch [40/150], Loss: 8.0497\n",
      "Epoch [41/150], Loss: 7.4977\n",
      "Epoch [42/150], Loss: 6.9760\n",
      "Epoch [43/150], Loss: 6.4760\n",
      "Epoch [44/150], Loss: 6.0018\n",
      "Epoch [45/150], Loss: 5.5497\n",
      "Epoch [46/150], Loss: 5.1181\n",
      "Epoch [47/150], Loss: 4.7114\n",
      "Epoch [48/150], Loss: 4.3274\n",
      "Epoch [49/150], Loss: 3.9681\n",
      "Epoch [50/150], Loss: 3.6374\n",
      "Epoch [51/150], Loss: 3.3352\n",
      "Epoch [52/150], Loss: 3.0595\n",
      "Epoch [53/150], Loss: 2.8110\n",
      "Epoch [54/150], Loss: 2.5870\n",
      "Epoch [55/150], Loss: 2.3864\n",
      "Epoch [56/150], Loss: 2.2063\n",
      "Epoch [57/150], Loss: 2.0461\n",
      "Epoch [58/150], Loss: 1.9028\n",
      "Epoch [59/150], Loss: 1.7759\n",
      "Epoch [60/150], Loss: 1.6635\n",
      "Epoch [61/150], Loss: 1.5635\n",
      "Epoch [62/150], Loss: 1.4746\n",
      "Epoch [63/150], Loss: 1.3957\n",
      "Epoch [64/150], Loss: 1.3249\n",
      "Epoch [65/150], Loss: 1.2614\n",
      "Epoch [66/150], Loss: 1.2041\n",
      "Epoch [67/150], Loss: 1.1521\n",
      "Epoch [68/150], Loss: 1.1048\n",
      "Epoch [69/150], Loss: 1.0617\n",
      "Epoch [70/150], Loss: 1.0219\n",
      "Epoch [71/150], Loss: 0.9854\n",
      "Epoch [72/150], Loss: 0.9516\n",
      "Epoch [73/150], Loss: 0.9204\n",
      "Epoch [74/150], Loss: 0.8912\n",
      "Epoch [75/150], Loss: 0.8641\n",
      "Epoch [76/150], Loss: 0.8387\n",
      "Epoch [77/150], Loss: 0.8150\n",
      "Epoch [78/150], Loss: 0.7926\n",
      "Epoch [79/150], Loss: 0.7716\n",
      "Epoch [80/150], Loss: 0.7517\n",
      "Epoch [81/150], Loss: 0.7330\n",
      "Epoch [82/150], Loss: 0.7151\n",
      "Epoch [83/150], Loss: 0.6982\n",
      "Epoch [84/150], Loss: 0.6823\n",
      "Epoch [85/150], Loss: 0.6670\n",
      "Epoch [86/150], Loss: 0.6525\n",
      "Epoch [87/150], Loss: 0.6386\n",
      "Epoch [88/150], Loss: 0.6254\n",
      "Epoch [89/150], Loss: 0.6128\n",
      "Epoch [90/150], Loss: 0.6007\n",
      "Epoch [91/150], Loss: 0.5890\n",
      "Epoch [92/150], Loss: 0.5779\n",
      "Epoch [93/150], Loss: 0.5672\n",
      "Epoch [94/150], Loss: 0.5569\n",
      "Epoch [95/150], Loss: 0.5470\n",
      "Epoch [96/150], Loss: 0.5375\n",
      "Epoch [97/150], Loss: 0.5283\n",
      "Epoch [98/150], Loss: 0.5195\n",
      "Epoch [99/150], Loss: 0.5109\n",
      "Epoch [100/150], Loss: 0.5026\n",
      "Epoch [101/150], Loss: 0.4947\n",
      "Epoch [102/150], Loss: 0.4869\n",
      "Epoch [103/150], Loss: 0.4794\n",
      "Epoch [104/150], Loss: 0.4722\n",
      "Epoch [105/150], Loss: 0.4652\n",
      "Epoch [106/150], Loss: 0.4584\n",
      "Epoch [107/150], Loss: 0.4518\n",
      "Epoch [108/150], Loss: 0.4454\n",
      "Epoch [109/150], Loss: 0.4392\n",
      "Epoch [110/150], Loss: 0.4332\n",
      "Epoch [111/150], Loss: 0.4273\n",
      "Epoch [112/150], Loss: 0.4216\n",
      "Epoch [113/150], Loss: 0.4160\n",
      "Epoch [114/150], Loss: 0.4106\n",
      "Epoch [115/150], Loss: 0.4054\n",
      "Epoch [116/150], Loss: 0.4002\n",
      "Epoch [117/150], Loss: 0.3953\n",
      "Epoch [118/150], Loss: 0.3904\n",
      "Epoch [119/150], Loss: 0.3857\n",
      "Epoch [120/150], Loss: 0.3810\n",
      "Epoch [121/150], Loss: 0.3765\n",
      "Epoch [122/150], Loss: 0.3721\n",
      "Epoch [123/150], Loss: 0.3678\n",
      "Epoch [124/150], Loss: 0.3636\n",
      "Epoch [125/150], Loss: 0.3595\n",
      "Epoch [126/150], Loss: 0.3555\n",
      "Epoch [127/150], Loss: 0.3516\n",
      "Epoch [128/150], Loss: 0.3477\n",
      "Epoch [129/150], Loss: 0.3440\n",
      "Epoch [130/150], Loss: 0.3403\n",
      "Epoch [131/150], Loss: 0.3367\n",
      "Epoch [132/150], Loss: 0.3332\n",
      "Epoch [133/150], Loss: 0.3298\n",
      "Epoch [134/150], Loss: 0.3264\n",
      "Epoch [135/150], Loss: 0.3231\n",
      "Epoch [136/150], Loss: 0.3199\n",
      "Epoch [137/150], Loss: 0.3167\n",
      "Epoch [138/150], Loss: 0.3136\n",
      "Epoch [139/150], Loss: 0.3106\n",
      "Epoch [140/150], Loss: 0.3076\n",
      "Epoch [141/150], Loss: 0.3046\n",
      "Epoch [142/150], Loss: 0.3018\n",
      "Epoch [143/150], Loss: 0.2990\n",
      "Epoch [144/150], Loss: 0.2962\n",
      "Epoch [145/150], Loss: 0.2935\n",
      "Epoch [146/150], Loss: 0.2908\n",
      "Epoch [147/150], Loss: 0.2882\n",
      "Epoch [148/150], Loss: 0.2856\n",
      "Epoch [149/150], Loss: 0.2831\n",
      "Epoch [150/150], Loss: 0.2806\n"
     ]
    }
   ],
   "source": [
    "class MyMultiClassClassifier_gru(nn.Module):\n",
    "    def __init__(self, num_classes):\n",
    "        super(MyMultiClassClassifier_gru, self).__init__()\n",
    "        self.gru = nn.GRU(5000,256,num_layers=2)\n",
    "        \n",
    "        self.fc3_p = nn.Linear(256, 3*num_classes)\n",
    "    def forward(self, x):\n",
    "        \n",
    "        output1, hn1 = self.gru(x.reshape(1,48,5000))\n",
    "        output1 = output1.sum(1)\n",
    "        output = torch.relu(output1)\n",
    "        \n",
    "        output = self.fc3_p(output)\n",
    "\n",
    "        return output.reshape(-1,5,3)\n",
    "\n",
    "num_classes = 5\n",
    "input_size = 5000 * 48\n",
    "\n",
    "multi_class_classifier_gru = MyMultiClassClassifier_gru(num_classes)\n",
    "\n",
    "loss_gru = []\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.SGD(multi_class_classifier_gru.parameters(), lr=0.001)\n",
    "\n",
    "\n",
    "num_epochs = 150\n",
    "\n",
    "for epoch in range(num_epochs):\n",
    "    total_loss = 0  # to monitor the loss\n",
    "\n",
    "    for D_i, target_triple in data_dict.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target_p, target_q, target_r = target_triple\n",
    "        \n",
    "        p_q_r = multi_class_classifier_gru(D_i_flat)\n",
    "        \n",
    "       \n",
    "        loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n",
    "        \n",
    "        total_loss +=loss.item()\n",
    "        \n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "      \n",
    "        \n",
    "\n",
    "    if (epoch + 1) % 1 == 0:\n",
    "        print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {total_loss:.4f}')\n",
    "        loss_gru.append(total_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "id": "678c7be2-5c88-493f-8f1a-c4942aa77765",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted: (0, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 1), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (1, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 0, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (1, 4, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "the model is able to predict with 42.0 percent\n"
     ]
    }
   ],
   "source": [
    "        \n",
    "# Test the model with train data first\n",
    "# data = data_dict\n",
    "\n",
    "data = test_data\n",
    "#data = pred_data\n",
    "with torch.no_grad():\n",
    "    s = 0 \n",
    "    for D_i, target in data.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target = torch.tensor(target)\n",
    "        outputs = multi_class_classifier_gru(D_i_flat)\n",
    "    \n",
    "        p,q,r = outputs.argmax(1).squeeze(0)\n",
    "        \n",
    "        p_q_r = (p.item(),q.item(),r.item())\n",
    "        \n",
    "        if p_q_r == tuple(target.numpy()):\n",
    "            s += 1 \n",
    "            \n",
    "        print(f'Predicted: {p_q_r}, Actual: {tuple(target.numpy())}')\n",
    "print(f'the model is able to predict with { s/(len(data))*100} percent')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a5146e8a-4ec8-4d46-bfad-43fadf44828f",
   "metadata": {},
   "source": [
    "# Concatenate to predict simultaneously"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "55454174-6d4d-4344-b323-61084fe07a89",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_2230923/679499654.py:47: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [1/150], Loss: 380.5985\n",
      "Epoch [2/150], Loss: 346.7325\n",
      "Epoch [3/150], Loss: 299.4430\n",
      "Epoch [4/150], Loss: 256.5110\n",
      "Epoch [5/150], Loss: 215.9307\n",
      "Epoch [6/150], Loss: 180.7755\n",
      "Epoch [7/150], Loss: 150.8835\n",
      "Epoch [8/150], Loss: 127.2010\n",
      "Epoch [9/150], Loss: 108.7470\n",
      "Epoch [10/150], Loss: 94.9260\n",
      "Epoch [17/150], Loss: 46.1978\n",
      "Epoch [18/150], Loss: 42.4529\n",
      "Epoch [19/150], Loss: 39.0816\n",
      "Epoch [20/150], Loss: 36.0850\n",
      "Epoch [21/150], Loss: 33.3356\n",
      "Epoch [22/150], Loss: 30.8277\n",
      "Epoch [23/150], Loss: 28.5177\n",
      "Epoch [24/150], Loss: 26.4124\n",
      "Epoch [25/150], Loss: 24.4864\n",
      "Epoch [26/150], Loss: 22.7019\n",
      "Epoch [27/150], Loss: 21.0697\n",
      "Epoch [28/150], Loss: 19.5587\n",
      "Epoch [29/150], Loss: 18.1777\n",
      "Epoch [30/150], Loss: 16.9095\n",
      "Epoch [31/150], Loss: 15.7488\n",
      "Epoch [32/150], Loss: 14.6761\n",
      "Epoch [33/150], Loss: 13.6854\n",
      "Epoch [34/150], Loss: 12.7656\n",
      "Epoch [35/150], Loss: 11.9121\n",
      "Epoch [36/150], Loss: 11.1211\n",
      "Epoch [37/150], Loss: 10.3761\n",
      "Epoch [38/150], Loss: 9.6827\n",
      "Epoch [39/150], Loss: 9.0274\n",
      "Epoch [40/150], Loss: 8.4160\n",
      "Epoch [41/150], Loss: 7.8309\n",
      "Epoch [42/150], Loss: 7.2846\n",
      "Epoch [43/150], Loss: 6.7636\n",
      "Epoch [44/150], Loss: 6.2740\n",
      "Epoch [45/150], Loss: 5.8070\n",
      "Epoch [46/150], Loss: 5.3677\n",
      "Epoch [47/150], Loss: 4.9502\n",
      "Epoch [48/150], Loss: 4.5606\n",
      "Epoch [49/150], Loss: 4.1965\n",
      "Epoch [50/150], Loss: 3.8577\n",
      "Epoch [51/150], Loss: 3.5444\n",
      "Epoch [52/150], Loss: 3.2581\n",
      "Epoch [53/150], Loss: 2.9993\n",
      "Epoch [54/150], Loss: 2.7655\n",
      "Epoch [55/150], Loss: 2.5550\n",
      "Epoch [56/150], Loss: 2.3661\n",
      "Epoch [57/150], Loss: 2.1968\n",
      "Epoch [58/150], Loss: 2.0445\n",
      "Epoch [59/150], Loss: 1.9079\n",
      "Epoch [60/150], Loss: 1.7854\n",
      "Epoch [61/150], Loss: 1.6754\n",
      "Epoch [62/150], Loss: 1.5769\n",
      "Epoch [63/150], Loss: 1.4884\n",
      "Epoch [64/150], Loss: 1.4088\n",
      "Epoch [65/150], Loss: 1.3371\n",
      "Epoch [66/150], Loss: 1.2726\n",
      "Epoch [67/150], Loss: 1.2141\n",
      "Epoch [68/150], Loss: 1.1610\n",
      "Epoch [69/150], Loss: 1.1124\n",
      "Epoch [70/150], Loss: 1.0680\n",
      "Epoch [71/150], Loss: 1.0273\n",
      "Epoch [72/150], Loss: 0.9898\n",
      "Epoch [73/150], Loss: 0.9550\n",
      "Epoch [74/150], Loss: 0.9228\n",
      "Epoch [75/150], Loss: 0.8929\n",
      "Epoch [76/150], Loss: 0.8651\n",
      "Epoch [77/150], Loss: 0.8390\n",
      "Epoch [78/150], Loss: 0.8146\n",
      "Epoch [79/150], Loss: 0.7917\n",
      "Epoch [80/150], Loss: 0.7701\n",
      "Epoch [81/150], Loss: 0.7498\n",
      "Epoch [82/150], Loss: 0.7306\n",
      "Epoch [83/150], Loss: 0.7124\n",
      "Epoch [84/150], Loss: 0.6952\n",
      "Epoch [85/150], Loss: 0.6788\n",
      "Epoch [86/150], Loss: 0.6633\n",
      "Epoch [87/150], Loss: 0.6485\n",
      "Epoch [88/150], Loss: 0.6344\n",
      "Epoch [89/150], Loss: 0.6209\n",
      "Epoch [90/150], Loss: 0.6080\n",
      "Epoch [91/150], Loss: 0.5957\n",
      "Epoch [92/150], Loss: 0.5839\n",
      "Epoch [93/150], Loss: 0.5726\n",
      "Epoch [94/150], Loss: 0.5617\n",
      "Epoch [95/150], Loss: 0.5513\n",
      "Epoch [96/150], Loss: 0.5412\n",
      "Epoch [97/150], Loss: 0.5316\n",
      "Epoch [98/150], Loss: 0.5223\n",
      "Epoch [99/150], Loss: 0.5133\n",
      "Epoch [100/150], Loss: 0.5047\n",
      "Epoch [101/150], Loss: 0.4963\n",
      "Epoch [102/150], Loss: 0.4882\n",
      "Epoch [103/150], Loss: 0.4804\n",
      "Epoch [104/150], Loss: 0.4729\n",
      "Epoch [105/150], Loss: 0.4656\n",
      "Epoch [106/150], Loss: 0.4585\n",
      "Epoch [107/150], Loss: 0.4517\n",
      "Epoch [108/150], Loss: 0.4450\n",
      "Epoch [109/150], Loss: 0.4386\n",
      "Epoch [110/150], Loss: 0.4323\n",
      "Epoch [111/150], Loss: 0.4263\n",
      "Epoch [112/150], Loss: 0.4204\n",
      "Epoch [113/150], Loss: 0.4147\n",
      "Epoch [114/150], Loss: 0.4091\n",
      "Epoch [115/150], Loss: 0.4037\n",
      "Epoch [116/150], Loss: 0.3984\n",
      "Epoch [117/150], Loss: 0.3933\n",
      "Epoch [118/150], Loss: 0.3883\n",
      "Epoch [119/150], Loss: 0.3834\n",
      "Epoch [120/150], Loss: 0.3787\n",
      "Epoch [121/150], Loss: 0.3740\n",
      "Epoch [122/150], Loss: 0.3695\n",
      "Epoch [123/150], Loss: 0.3651\n",
      "Epoch [124/150], Loss: 0.3608\n",
      "Epoch [125/150], Loss: 0.3566\n",
      "Epoch [126/150], Loss: 0.3526\n",
      "Epoch [127/150], Loss: 0.3486\n",
      "Epoch [128/150], Loss: 0.3447\n",
      "Epoch [129/150], Loss: 0.3408\n",
      "Epoch [130/150], Loss: 0.3371\n",
      "Epoch [131/150], Loss: 0.3335\n",
      "Epoch [132/150], Loss: 0.3299\n",
      "Epoch [133/150], Loss: 0.3264\n",
      "Epoch [134/150], Loss: 0.3230\n",
      "Epoch [135/150], Loss: 0.3196\n",
      "Epoch [136/150], Loss: 0.3164\n",
      "Epoch [137/150], Loss: 0.3132\n",
      "Epoch [138/150], Loss: 0.3100\n",
      "Epoch [139/150], Loss: 0.3069\n",
      "Epoch [140/150], Loss: 0.3039\n",
      "Epoch [141/150], Loss: 0.3010\n",
      "Epoch [142/150], Loss: 0.2980\n",
      "Epoch [143/150], Loss: 0.2952\n",
      "Epoch [144/150], Loss: 0.2924\n",
      "Epoch [145/150], Loss: 0.2897\n",
      "Epoch [146/150], Loss: 0.2870\n",
      "Epoch [147/150], Loss: 0.2844\n",
      "Epoch [148/150], Loss: 0.2818\n",
      "Epoch [149/150], Loss: 0.2792\n",
      "Epoch [150/150], Loss: 0.2767\n"
     ]
    }
   ],
   "source": [
    "class MyMultiClassClassifier(nn.Module):\n",
    "    def __init__(self, num_classes):\n",
    "        super(MyMultiClassClassifier, self).__init__()\n",
    "        self.gru = nn.GRU(5000,256,num_layers=2)\n",
    "        self.lstm = nn.LSTM(5000,256,num_layers=2)\n",
    "        # self.fc2 = nn.Linear(64, 32)\n",
    "        self.fc3_p = nn.Linear(2*256, 3*num_classes)\n",
    "        \n",
    "\n",
    "    def forward(self, x):\n",
    "        output1, hn1 = self.gru(x.reshape(1,48,5000))\n",
    "        output2, (hn2, cn2) = self.lstm(x.reshape(1,48,5000))\n",
    "        output1 = output1.sum(1)\n",
    "        output2 = output2.sum(1)\n",
    "        output = torch.concatenate([output1,output2], 1)\n",
    "        output = torch.relu(output)\n",
    "        \n",
    "        #x = torch.relu(self.fc2(x))\n",
    "        output = self.fc3_p(output)\n",
    "       \n",
    "        return output.reshape(-1,5,3)\n",
    "\n",
    "# Assuming 5 classes (0 to 4 for each of p, q, r)\n",
    "num_classes = 5\n",
    "input_size = 5000 * 48    #(Nx3d dimension matrix)\n",
    "\n",
    "multi_class_classifier_model = MyMultiClassClassifier(num_classes)\n",
    "\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "\n",
    "\n",
    "optimizer = optim.SGD(multi_class_classifier_model.parameters(), lr=0.001)\n",
    "num_epochs = 150\n",
    "loss_concat = []\n",
    "for epoch in range(num_epochs):\n",
    "    total_loss = 0  # to monitor the loss\n",
    "\n",
    "    for D_i, target_triple in data_dict.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target_p, target_q, target_r = target_triple\n",
    "\n",
    "       \n",
    "        p_q_r = multi_class_classifier_model(D_i_flat)\n",
    "\n",
    "       \n",
    "        loss = criterion(p_q_r, torch.tensor(torch.tensor([target_p,target_q,target_r]), dtype=torch.long).unsqueeze(0))\n",
    "        \n",
    "        total_loss +=loss.item()\n",
    "        \n",
    "        \n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "    if (epoch + 1) % 1 == 0:\n",
    "        print(f'Epoch [{epoch+1}/{num_epochs}], Loss: {total_loss:.4f}')\n",
    "        loss_concat.append(total_loss)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "id": "d86b22b6-d2a6-4197-8a80-78111c6bb6be",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted: (0, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (1, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (1, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 2, 1), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 0, 1), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 3), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (4, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (4, 2, 1)\n",
      "Predicted: (0, 0, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 4), Actual: (4, 2, 1)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 4, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 0), Actual: (0, 0, 3)\n",
      "Predicted: (1, 2, 0), Actual: (4, 2, 1)\n",
      "Predicted: (0, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 3)\n",
      "Predicted: (0, 0, 0), Actual: (0, 0, 4)\n",
      "Predicted: (0, 0, 3), Actual: (0, 0, 3)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (0, 4, 4), Actual: (4, 2, 1)\n",
      "Predicted: (1, 2, 0), Actual: (1, 2, 0)\n",
      "Predicted: (1, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (1, 2, 0)\n",
      "Predicted: (1, 2, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 0), Actual: (0, 4, 4)\n",
      "Predicted: (0, 1, 3), Actual: (0, 1, 0)\n",
      "Predicted: (0, 2, 1), Actual: (0, 4, 4)\n",
      "Predicted: (0, 0, 4), Actual: (0, 0, 4)\n",
      "Predicted: (0, 4, 4), Actual: (0, 4, 4)\n",
      "Predicted: (0, 2, 4), Actual: (1, 2, 0)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "Predicted: (0, 1, 0), Actual: (0, 1, 0)\n",
      "Predicted: (4, 2, 1), Actual: (1, 2, 0)\n",
      "the model is able to predict with 35.5 percent\n"
     ]
    }
   ],
   "source": [
    "        \n",
    "# Test the model with train data first\n",
    "# data = data_dict\n",
    "\n",
    "data = test_data\n",
    "# data = pred_data\n",
    "with torch.no_grad():\n",
    "    s = 0 \n",
    "    for D_i, target in data.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target = torch.tensor(target)\n",
    "        outputs = multi_class_classifier_model(D_i_flat)\n",
    "    \n",
    "        p,q,r = outputs.argmax(1).squeeze(0)\n",
    "        \n",
    "        p_q_r = (p.item(),q.item(),r.item())\n",
    "        \n",
    "        if p_q_r == tuple(target.numpy()):\n",
    "            s += 1 \n",
    "            \n",
    "        print(f'Predicted: {p_q_r}, Actual: {tuple(target.numpy())}')\n",
    "print(f'the model is able to predict with { s/(len(data))*100} percent')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dee7a2d7-1c10-4171-8446-1604a1fe5f29",
   "metadata": {},
   "source": [
    "# Assembling LSTM, GRU and Concat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "ba415ddd-53b1-442f-a3ad-5a07f92d9bce",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the model is able to predict with 45.0 percent\n"
     ]
    }
   ],
   "source": [
    "# Test the model with train data first\n",
    "# data = data_dict\n",
    "data = test_data\n",
    "#data = pred_data\n",
    "\n",
    "#define the weights for the assembling\n",
    "a = 0.75\n",
    "b = 0.2\n",
    "c = 1-a-b\n",
    "\n",
    "ss = 0\n",
    "score, dataset = (4,1,2), 'FB15k' #left out dataset \n",
    "with torch.no_grad():\n",
    "    s = 0 \n",
    "    i = 0\n",
    "    m = []\n",
    "    pred = []\n",
    "    for D_i, target in data.items():\n",
    "        D_i_flat = D_i.view(-1)\n",
    "        target = torch.tensor(target)\n",
    "        outputs1 = multi_class_classifier_lstm(D_i_flat)\n",
    "        outputs2 = multi_class_classifier_gru(D_i_flat)\n",
    "        outputs3 = multi_class_classifier_model(D_i_flat)\n",
    "        \n",
    "        outputs1 = outputs1.squeeze(0)\n",
    "        outputs2 = outputs2.squeeze(0)\n",
    "        outputs3 = outputs3.reshape(5,3)\n",
    "        \n",
    "        \n",
    "        p1,q1,r1 = outputs1[:,0], outputs1[:,1], outputs1[:,2]\n",
    "        p2,q2,r2 = outputs2[:,0], outputs2[:,1], outputs2[:,2]\n",
    "        \n",
    "        p3,q3,r3 = outputs3[:,0], outputs3[:,1], outputs3[:,2]\n",
    "        \n",
    "        p = a*p1+b*p2 + c*p3 \n",
    "        q = a*q1+b*q2 + c*q3\n",
    "        r = a*r1+b*r2 + c*r3\n",
    "        \n",
    "        p= torch.softmax(p,0).argmax(0)\n",
    "        q= torch.softmax(q,0).argmax(0)\n",
    "        r= torch.softmax(r,0).argmax(0)\n",
    "        \n",
    "        \n",
    "        p_q_r = (p.item(),q.item(),r.item())\n",
    "        \n",
    "        if tuple(target.numpy()) == score:\n",
    "            pred.append(p_q_r)\n",
    "        \n",
    "        if ((p_q_r == score) and (tuple(target.numpy()) == score)):\n",
    "            ss+=1\n",
    "            \n",
    "        # if ((p_q_r != score) and (tuple(target.numpy()) == score)):\n",
    "        #     m.append(p_q_r) #extract false positive triples (eg: the model predicted (0,0,1) instead of (0,1,0))\n",
    "    \n",
    "        \n",
    "        if p_q_r == tuple(target.numpy()):\n",
    "            s += 1 \n",
    "            \n",
    "        # print(f'Predicted: {p_q_r}, Actual: {tuple(target.numpy())}')\n",
    "print(f'the model is able to predict with { s/(len(data))*100} percent')\n",
    "# print(ss) \n",
    "# print(f'm_{dataset} = {m}')\n",
    "#print(f'pred_{dataset} = {pred}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "id": "42152c11-dbef-406e-a430-b170bd3dde94",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 400x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(4, 3))\n",
    "\n",
    "plt.plot(loss_concat,'--b', linewidth = 2.5)\n",
    "plt.plot(loss_gru,'--g', linewidth = 2.5)\n",
    "plt.plot(loss_lstm,'--r', linewidth = 2.5)\n",
    "plt.legend(['Concat','GRU', 'LSTM'], fontsize = 20)\n",
    "plt.xlabel('Number of epochs', fontsize = 20)\n",
    "plt.ylabel('Loss', fontsize = 26)\n",
    "plt.grid()\n",
    "\n",
    "plt.savefig('AAAAA.pdf', bbox_inches='tight')  # Specify the filename\n",
    "plt.show()\n",
    "\n",
    "\n",
    "# with PdfPages('AAAAA.pdf') as pdf:\n",
    "#     pdf.savefig(AAAAA)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "id": "11cadfa0-032c-4d35-9071-04d91c135585",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x150 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "l_UMLS = [(0, 0, 0, 0.760275910907946), (0, 0, 1, 0.862229000778265), (0, 0, 2, 0.847344843367232), (0, 0, 3, 0.8615240687227784), (0, 0, 4, 0.8860335825703147), (0, 1, 0, 0.8410411483389052), (0, 1, 1, 0.8422898680155887), (0, 1, 2, 0.8532997026992168), (0, 1, 3, 0.8533437777189904), (0, 1, 4, 0.8616491071838286), (0, 2, 0, 0.8680325898919706), (0, 2, 1, 0.8465739244125556), (0, 2, 2, 0.8563098935692324), (0, 2, 3, 0.8587118141565244), (0, 2, 4, 0.8463189189085363), (0, 3, 0, 0.8310852915363337), (0, 3, 1, 0.8627462986062845), (0, 3, 2, 0.858209767013325), (0, 3, 3, 0.84411741381356), (0, 3, 4, 0.8684288363271512), (0, 4, 0, 0.8565114944949549), (0, 4, 1, 0.8430583166781902), (0, 4, 2, 0.875933626157362), (0, 4, 3, 0.8660626183259119), (0, 4, 4, 0.7904911584156828), (1, 0, 0, 0.8345858331319158), (1, 0, 1, 0.8290638372948682), (1, 0, 2, 0.8496088095698351), (1, 0, 3, 0.8800515113404744), (1, 0, 4, 0.853838021450744), (1, 1, 0, 0.8269957636749112), (1, 1, 1, 0.8281741731662684), (1, 1, 2, 0.81574670852512), (1, 1, 3, 0.8324084878716717), (1, 1, 4, 0.874802619867001), (1, 2, 0, 0.8499241303024414), (1, 2, 1, 0.866974985175783), (1, 2, 2, 0.8565911952863915), (1, 2, 3, 0.8618783165241926), (1, 2, 4, 0.8634113413469774), (1, 3, 0, 0.8325477324861398), (1, 3, 1, 0.8596351796394212), (1, 3, 2, 0.8375923840734821), (1, 3, 3, 0.8466595498924637), (1, 3, 4, 0.7741863025302355), (1, 4, 0, 0.8744075533193798), (1, 4, 1, 0.8703413388748277), (1, 4, 2, 0.8619258829088289), (1, 4, 3, 0.7950820845148011), (1, 4, 4, 0.7439202043700484), (2, 0, 0, 0.859884782711585), (2, 0, 1, 0.8496479237637672), (2, 0, 2, 0.8500634261022122), (2, 0, 3, 0.8532528142433634), (2, 0, 4, 0.8457164155408384), (2, 1, 0, 0.8441325324156124), (2, 1, 1, 0.8516628833672049), (2, 1, 2, 0.8494311311737645), (2, 1, 3, 0.8533905816860125), (2, 1, 4, 0.8354785976845163), (2, 2, 0, 0.836528997281592), (2, 2, 1, 0.8431284502243321), (2, 2, 2, 0.8701203427951945), (2, 2, 3, 0.8694574610408072), (2, 2, 4, 0.7597119518681906), (2, 3, 0, 0.8356354230786348), (2, 3, 1, 0.8728155173706473), (2, 3, 2, 0.8306614110656458), (2, 3, 3, 0.7747271950210397), (2, 3, 4, 0.7710263955271583), (2, 4, 0, 0.8458425625913123), (2, 4, 1, 0.8753927614986484), (2, 4, 2, 0.7410682402719259), (2, 4, 3, 0.7815668833690336), (2, 4, 4, 0.8454593924948682), (3, 0, 0, 0.8528226518496842), (3, 0, 1, 0.8550812487146976), (3, 0, 2, 0.8691285799303342), (3, 0, 3, 0.8658256468267105), (3, 0, 4, 0.8689137556761258), (3, 1, 0, 0.8631341107685665), (3, 1, 1, 0.8498495405705275), (3, 1, 2, 0.8660271148163033), (3, 1, 3, 0.8526676097939123), (3, 1, 4, 0.7877072228025567), (3, 2, 0, 0.8613507793502249), (3, 2, 1, 0.8728821257692595), (3, 2, 2, 0.870755188149104), (3, 2, 3, 0.7965719327002163), (3, 2, 4, 0.8104317816306914), (3, 3, 0, 0.8555549184957711), (3, 3, 1, 0.8612468074934252), (3, 3, 2, 0.7633279594833156), (3, 3, 3, 0.7768915432332466), (3, 3, 4, 0.8475603753936408), (3, 4, 0, 0.8567752320488278), (3, 4, 1, 0.8062792608581694), (3, 4, 2, 0.7219243657735646), (3, 4, 3, 0.7783389398382022), (3, 4, 4, 0.7935538900078376), (4, 0, 0, 0.8511859319742767), (4, 0, 1, 0.8747517455780935), (4, 0, 2, 0.8754589560872349), (4, 0, 3, 0.8633193145860292), (4, 0, 4, 0.803811778806831), (4, 1, 0, 0.852159864952638), (4, 1, 1, 0.8494864040097024), (4, 1, 2, 0.8617039263980638), (4, 1, 3, 0.772779998033765), (4, 1, 4, 0.7556050703219983), (4, 2, 0, 0.8666358913276776), (4, 2, 1, 0.8510576740640554), (4, 2, 2, 0.7714989563495686), (4, 2, 3, 0.7950203638063116), (4, 2, 4, 0.8439636579177954), (4, 3, 0, 0.8548472313561037), (4, 3, 1, 0.8128656557862864), (4, 3, 2, 0.7835019026348693), (4, 3, 3, 0.7941074211758173), (4, 3, 4, 0.7929811658297832), (4, 4, 0, 0.7909921023757506), (4, 4, 1, 0.8550057470675408), (4, 4, 2, 0.8003844173583224), (4, 4, 3, 0.7791112046470429), (4, 4, 4, 0.7919423348487613)]\n",
    "\n",
    "#plot MRR values of UMLS\n",
    "l = l_UMLS\n",
    "mrr_values = [item[-1] for item in l]\n",
    "triples = [(item[0], item[1], item[2]) for item in l]\n",
    "x_labels = range(0,len(mrr_values))#[str(triple) for triple in triples]\n",
    "\n",
    "\n",
    "plt.figure(figsize=(10,1.5))\n",
    "\n",
    "plt.plot(x_labels,mrr_values, '.-b', linewidth=1.5)\n",
    "\n",
    "# ax.set_xlim(0, data.shape[1])\n",
    "# ax.set_ylim(0, data.shape[0])\n",
    "\n",
    "plt.gca().yaxis.set_major_formatter(plt.matplotlib.ticker.StrMethodFormatter(\"{x:,.2f}\"))\n",
    "plt.ylabel('MRR', fontsize=26)\n",
    "#plt.title('MRR Values for UMLS')\n",
    "plt.grid()\n",
    "plt.savefig('exhaus_UMLS.pdf',bbox_inches='tight')\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (anaconda)",
   "language": "python",
   "name": "python3-anaconda"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}