decoder_visualization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle \n",
    "import yaml\n",
    "import pandas as pd\n",
    "from PrepareData import prepare_data\n",
    "\n",
    "\n",
    "import torch\n",
    "from torch import nn, optim, Tensor\n",
    "from torch.nn import functional as F\n",
    "import pickle \n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import rdkit\n",
    "from rdkit import Chem\n",
    "from rdkit.Chem import AllChem\n",
    "from rdkit.Chem import Draw\n",
    "import seaborn as sns\n",
    "from architecture import CLIP\n",
    "from tqdm import tqdm\n",
    "\n",
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "dtype = torch.float32\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from train_utils import load_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_deterministic(random_seed = 0):\n",
    "    torch.manual_seed(random_seed)\n",
    "    torch.cuda.manual_seed(random_seed)\n",
    "    torch.backends.cudnn.deterministic = True\n",
    "    torch.backends.cudnn.benchmark = False\n",
    "    np.random.seed(random_seed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "make_deterministic(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "config = yaml.safe_load(open('./checkpoints/RUN_unit_norm_fixed_decoder/config.yaml', 'r'))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best_epoch 475\n",
      "best_clip_epoch 663\n",
      "best_recon_epoch 176\n",
      "best_total_loss 0.057113549521853844\n",
      "best_clip_loss 0.010364494789692183\n",
      "best_recon_loss 0.04040767757161971\n"
     ]
    }
   ],
   "source": [
    "logs = pickle.load(open('./checkpoints/RUN_unit_norm_fixed_decoder/logs.pickle', 'rb'))\n",
    "for key in logs:\n",
    "    if \"best\" in key:\n",
    "        print(key, logs[key])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'checkpoints/RUN_unit_norm_fixed_decoder'"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "config['train']['checkpoint_dir']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "97e6da50b0f147ff842f8f2ae126faef",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7cc433020bd146f595b413ebdea0b2eb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "0it [00:00, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model = load_model(config['train']['checkpoint_dir'], type=\"best_total\")\n",
    "model.eval()\n",
    "dataloaders, max_charge, num_species = prepare_data(config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "from train_utils import decoder_performance\n",
    "from train_utils import top_scores, decoder_performance, distance_distribution\n",
    "\n",
    "def clip_performance(config, model, dataloaders, epoch):\n",
    "    # model.to(device)\n",
    "    model.eval()\n",
    "    max_charge = config['data']['max_charge']\n",
    "    num_species = config['data']['num_species']\n",
    "\n",
    "    molembeds = []\n",
    "    specembeds = []\n",
    "    with torch.no_grad():\n",
    "        for i, data in tqdm(enumerate(dataloaders['val'])):    \n",
    "            data = {k: v.to(device) for k, v in data.items()}\n",
    "            mol_latents, spec_latents, smile_preds, logit_scale, ids = model(data)\n",
    "            molembeds.append(mol_latents.detach().cpu())\n",
    "            specembeds.append(spec_latents.detach().cpu())\n",
    "        del mol_latents, spec_latents, smile_preds, logit_scale, ids\n",
    "\n",
    "    test_molembeds = torch.cat(molembeds, 0)\n",
    "    test_specembeds = torch.cat(specembeds, 0)\n",
    "    \n",
    "    molembeds = []\n",
    "    specembeds = []\n",
    "    with torch.no_grad():\n",
    "        for i, data in tqdm(enumerate(dataloaders['train'])):    \n",
    "            data = {k: v.to(device) for k, v in data.items()}\n",
    "            mol_latents, spec_latents, smile_preds, logit_scale, ids = model(data)\n",
    "            molembeds.append(mol_latents.detach().cpu())\n",
    "                # specembeds.append(spec_latents.detach().cpu())\n",
    "        del mol_latents, spec_latents, smile_preds, logit_scale, ids\n",
    "    \n",
    "    train_molembeds = torch.cat(molembeds, 0)\n",
    "    # train_specembeds = torch.cat(specembeds, 0)\n",
    "    \n",
    "    all_molembeds = torch.cat(( test_molembeds, train_molembeds), axis = 0)\n",
    "    del train_molembeds\n",
    "    \n",
    "    tops, scores = top_scores(test_specembeds, all_molembeds)\n",
    "    del all_molembeds\n",
    "    \n",
    "    for k, acc in zip(tops, scores):\n",
    "        # print(\"Full Top {} Spec\".format(k), acc)\n",
    "        print({\"Full Top {} Spec\".format(k): acc})\n",
    "        \n",
    "\n",
    "    tops, scores = top_scores(test_specembeds, test_molembeds )\n",
    "    for k, acc in zip(tops, scores):\n",
    "        # print(\"Test Top {} Spec\".format(k), acc)\n",
    "        print({\"Test Top {} Spec\".format(k): acc})\n",
    "\n",
    "    # decoder_performance(config, model, dataloaders, epoch)\n",
    "    plt.clf()\n",
    "    distance_distribution(test_molembeds, test_specembeds)\n",
    "    return test_molembeds, test_specembeds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "0it [00:00, ?it/s]/home2/kanakala.ganesh/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/parallel/_functions.py:68: UserWarning: Was asked to gather along dimension 0, but all input tensors were scalars; will instead unsqueeze and return a vector.\n",
      "  warnings.warn('Was asked to gather along dimension 0, but all '\n",
      "50it [00:34,  1.44it/s]\n",
      "250it [01:55,  2.16it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'Full Top 1 Spec': 0.005}\n",
      "{'Full Top 2 Spec': 0.005}\n",
      "{'Full Top 3 Spec': 0.005}\n",
      "{'Full Top 4 Spec': 0.005}\n",
      "{'Full Top 5 Spec': 0.005}\n",
      "{'Full Top 7 Spec': 0.005}\n",
      "{'Full Top 10 Spec': 0.015}\n",
      "{'Test Top 1 Spec': 0.01}\n",
      "{'Test Top 2 Spec': 0.015}\n",
      "{'Test Top 3 Spec': 0.03}\n",
      "{'Test Top 4 Spec': 0.04}\n",
      "{'Test Top 5 Spec': 0.04}\n",
      "{'Test Top 7 Spec': 0.04}\n",
      "{'Test Top 10 Spec': 0.055}\n"
     ]
    }
   ],
   "source": [
    "test_molembeds, test_specembeds = clip_performance(config,model,dataloaders, -1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "from train_utils import Sampler"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "PAD = 0\n",
    "UNK = 1\n",
    "EOS = 2\n",
    "SOS = 3\n",
    "MASK = 4\n",
    "\n",
    "class Sampler():\n",
    "    def __init__(self, model, vocab):\n",
    "        self.model = model\n",
    "        self.vocab = vocab\n",
    "        self.max_len = 40\n",
    "    def sample(self, embed, greedy_decode=False):\n",
    "        embed = embed.unsqueeze(0).to(device)\n",
    "        self.model.eval()\n",
    "        sample_tensor = torch.zeros((1,self.max_len), dtype=torch.int64).to(device)\n",
    "        sample_tensor[0,0] = SOS\n",
    "        with torch.no_grad():\n",
    "            for i in range(0,self.max_len-1):\n",
    "                tensor = sample_tensor[:,:i+1]\n",
    "                logits = self.model.forward(embed, tensor)[:,-1,:]\n",
    "                probabilities = F.softmax(logits, dim=1)\n",
    "                sampled_char = torch.multinomial(probabilities,1).item()\n",
    "                if greedy_decode:\n",
    "                    sampled_char = torch.argmax(probabilities)\n",
    "                    \n",
    "                sample_tensor[0,i+1] = sampled_char\n",
    "                if sampled_char == EOS:\n",
    "                    break\n",
    "            smiles = \"\"\n",
    "            chars = self.vocab.from_seq(sample_tensor[0])\n",
    "            for char in chars:\n",
    "                if char != \"<pad>\" and char != \"<eos>\" and char != \"<sos>\" and char != \"<unk>\":\n",
    "                    smiles += char\n",
    "            \n",
    "        return smiles\n",
    "  \n",
    "    def sample_multi(self, n, embed, greedy_decode=False):\n",
    "        smiles_list = []\n",
    "        for i in range(n):\n",
    "            smiles_list.append(self.sample(embed, greedy_decode))\n",
    "        return smiles_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calculate_decoder_accuracy( model, dataloaders, k=1):\n",
    "    pred_smiles_list = []\n",
    "    og_smiles_list = []\n",
    "    count = 0\n",
    "    sampler = Sampler(model.module.smiles_decoder, model.module.vocab)\n",
    "    with torch.no_grad():\n",
    "        for i, data in tqdm(enumerate(dataloaders['train'])):\n",
    "            data = {k: v.to(device) for k, v in data.items()}\n",
    "            spec_latents = model.module.forward_spec(data)\n",
    "            for spec, og in zip(spec_latents, data['smiles'] ):\n",
    "                ls = sampler.sample_multi(k,spec,greedy_decode=True)\n",
    "                generated_smiles = []\n",
    "                for smi in ls:\n",
    "                    try:\n",
    "                        generated_smiles.append(Chem.CanonSmiles(smi))\n",
    "                    except:\n",
    "                        pass\n",
    "                og_smile = \"\"\n",
    "                chars = model.module.vocab.from_seq(og)\n",
    "                for char in chars:\n",
    "                    if char != \"<pad>\" and char != \"<eos>\" and char != \"<sos>\" and char != \"<unk>\":\n",
    "                        og_smile += char\n",
    "                og_smile = Chem.CanonSmiles(og_smile)\n",
    "                \n",
    "                if og_smile in generated_smiles:\n",
    "                    count += 1\n",
    "                \n",
    "                og_smiles_list.append(og_smile)\n",
    "                pred_smiles_list.append(generated_smiles)\n",
    "            print(\"counts\", count)\n",
    "            print(\"No of Hits percentage : \",count /  len(og_smiles_list))\n",
    "\n",
    "        return count / len(og_smiles_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "sampler = Sampler( model.module.smiles_decoder, model.module.vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1it [00:20, 20.13s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "counts 237\n",
      "No of Hits percentage :  0.5925\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1it [00:23, 23.72s/it]\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[49], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m acc, pred_smiles_list, og_smiles_list \u001b[39m=\u001b[39m calculate_decoder_accuracy(model, dataloaders, k\u001b[39m=\u001b[39;49m\u001b[39m1\u001b[39;49m)\n",
      "Cell \u001b[0;32mIn[47], line 11\u001b[0m, in \u001b[0;36mcalculate_decoder_accuracy\u001b[0;34m(model, dataloaders, k)\u001b[0m\n\u001b[1;32m      9\u001b[0m spec_latents \u001b[39m=\u001b[39m model\u001b[39m.\u001b[39mmodule\u001b[39m.\u001b[39mforward_spec(data)\n\u001b[1;32m     10\u001b[0m \u001b[39mfor\u001b[39;00m spec, og \u001b[39min\u001b[39;00m \u001b[39mzip\u001b[39m(spec_latents, data[\u001b[39m'\u001b[39m\u001b[39msmiles\u001b[39m\u001b[39m'\u001b[39m] ):\n\u001b[0;32m---> 11\u001b[0m     ls \u001b[39m=\u001b[39m sampler\u001b[39m.\u001b[39;49msample_multi(k,spec,greedy_decode\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m)\n\u001b[1;32m     12\u001b[0m     generated_smiles \u001b[39m=\u001b[39m []\n\u001b[1;32m     13\u001b[0m     \u001b[39mfor\u001b[39;00m smi \u001b[39min\u001b[39;00m ls:\n",
      "Cell \u001b[0;32mIn[46], line 40\u001b[0m, in \u001b[0;36mSampler.sample_multi\u001b[0;34m(self, n, embed, greedy_decode)\u001b[0m\n\u001b[1;32m     38\u001b[0m smiles_list \u001b[39m=\u001b[39m []\n\u001b[1;32m     39\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(n):\n\u001b[0;32m---> 40\u001b[0m     smiles_list\u001b[39m.\u001b[39mappend(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49msample(embed, greedy_decode))\n\u001b[1;32m     41\u001b[0m \u001b[39mreturn\u001b[39;00m smiles_list\n",
      "Cell \u001b[0;32mIn[46], line 22\u001b[0m, in \u001b[0;36mSampler.sample\u001b[0;34m(self, embed, greedy_decode)\u001b[0m\n\u001b[1;32m     20\u001b[0m logits \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmodel\u001b[39m.\u001b[39mforward(embed, tensor)[:,\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m,:]\n\u001b[1;32m     21\u001b[0m probabilities \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39msoftmax(logits, dim\u001b[39m=\u001b[39m\u001b[39m1\u001b[39m)\n\u001b[0;32m---> 22\u001b[0m sampled_char \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39;49mmultinomial(probabilities,\u001b[39m1\u001b[39;49m)\u001b[39m.\u001b[39mitem()\n\u001b[1;32m     23\u001b[0m \u001b[39mif\u001b[39;00m greedy_decode:\n\u001b[1;32m     24\u001b[0m     sampled_char \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39margmax(probabilities)\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1it [00:58, 58.69s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2it [01:53, 56.60s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.62625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "3it [02:49, 55.99s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.6158333333333333\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "4it [03:44, 55.85s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.62625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "5it [04:40, 55.90s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.6175\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "6it [05:37, 56.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.61\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "7it [06:32, 55.96s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.6064285714285714\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "8it [07:28, 55.88s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5971875\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "9it [08:25, 56.07s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5955555555555555\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "10it [09:25, 57.39s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.59275\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "11it [10:21, 56.94s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5915909090909091\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "12it [11:16, 56.31s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.58875\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "13it [12:10, 55.84s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5884615384615385\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "14it [13:06, 55.87s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5891071428571428\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "15it [14:02, 55.73s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5893333333333334\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "16it [14:57, 55.57s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.59203125\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "17it [15:53, 55.83s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5948529411764706\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "18it [16:49, 55.90s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5919444444444445\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "19it [17:46, 56.02s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5921052631578947\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "20it [18:41, 55.88s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.595125\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "21it [19:41, 57.02s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5951190476190477\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "22it [20:38, 56.97s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5943181818181819\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "23it [21:34, 56.65s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5954347826086956\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "24it [22:30, 56.40s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5957291666666666\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "25it [23:26, 56.36s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5961\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "26it [24:22, 56.19s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5971153846153846\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "27it [25:17, 56.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5952777777777778\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "28it [26:13, 56.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5964285714285714\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "29it [27:09, 56.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of HitS percentage :  0.5952586206896552\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "29it [27:16, 56.43s/it]\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[16], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m acc, pred_smiles_list, og_smiles_list \u001b[39m=\u001b[39m calculate_decoder_accuracy(model, dataloaders, k\u001b[39m=\u001b[39;49m\u001b[39m3\u001b[39;49m)\n",
      "Cell \u001b[0;32mIn[13], line 11\u001b[0m, in \u001b[0;36mcalculate_decoder_accuracy\u001b[0;34m(model, dataloaders, k)\u001b[0m\n\u001b[1;32m      9\u001b[0m spec_latents \u001b[39m=\u001b[39m model\u001b[39m.\u001b[39mmodule\u001b[39m.\u001b[39mforward_spec(data)\n\u001b[1;32m     10\u001b[0m \u001b[39mfor\u001b[39;00m spec, og \u001b[39min\u001b[39;00m \u001b[39mzip\u001b[39m(spec_latents, data[\u001b[39m'\u001b[39m\u001b[39msmiles\u001b[39m\u001b[39m'\u001b[39m] ):\n\u001b[0;32m---> 11\u001b[0m     ls \u001b[39m=\u001b[39m sampler\u001b[39m.\u001b[39;49msample_multi(k,spec,greedy_decode\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m)\n\u001b[1;32m     12\u001b[0m     generated_smiles \u001b[39m=\u001b[39m []\n\u001b[1;32m     13\u001b[0m     \u001b[39mfor\u001b[39;00m smi \u001b[39min\u001b[39;00m ls:\n",
      "Cell \u001b[0;32mIn[12], line 40\u001b[0m, in \u001b[0;36mSampler.sample_multi\u001b[0;34m(self, n, embed, greedy_decode)\u001b[0m\n\u001b[1;32m     38\u001b[0m smiles_list \u001b[39m=\u001b[39m []\n\u001b[1;32m     39\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(n):\n\u001b[0;32m---> 40\u001b[0m     smiles_list\u001b[39m.\u001b[39mappend(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49msample(embed, greedy_decode))\n\u001b[1;32m     41\u001b[0m \u001b[39mreturn\u001b[39;00m smiles_list\n",
      "Cell \u001b[0;32mIn[12], line 20\u001b[0m, in \u001b[0;36mSampler.sample\u001b[0;34m(self, embed, greedy_decode)\u001b[0m\n\u001b[1;32m     18\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(\u001b[39m0\u001b[39m,\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmax_len\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m):\n\u001b[1;32m     19\u001b[0m     tensor \u001b[39m=\u001b[39m sample_tensor[:,:i\u001b[39m+\u001b[39m\u001b[39m1\u001b[39m]\n\u001b[0;32m---> 20\u001b[0m     logits \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mmodel\u001b[39m.\u001b[39;49mforward(embed, tensor)[:,\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m,:]\n\u001b[1;32m     21\u001b[0m     probabilities \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39msoftmax(logits, dim\u001b[39m=\u001b[39m\u001b[39m1\u001b[39m)\n\u001b[1;32m     22\u001b[0m     sampled_char \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39mmultinomial(probabilities,\u001b[39m1\u001b[39m)\u001b[39m.\u001b[39mitem()\n",
      "File \u001b[0;32m~/CLIP_PART_1/models/decoder.py:100\u001b[0m, in \u001b[0;36mLatentToMol.forward\u001b[0;34m(self, spec, smi, tgt_mask, tgt_padding_mask)\u001b[0m\n\u001b[1;32m     97\u001b[0m spec \u001b[39m=\u001b[39m spec\u001b[39m.\u001b[39munsqueeze(\u001b[39m1\u001b[39m)\n\u001b[1;32m     98\u001b[0m \u001b[39m# print(smi.shape)\u001b[39;00m\n\u001b[1;32m     99\u001b[0m \u001b[39m# print(smi.dtype)\u001b[39;00m\n\u001b[0;32m--> 100\u001b[0m smi \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49membed(smi) \n\u001b[1;32m    101\u001b[0m smi \u001b[39m=\u001b[39m smi[:,\u001b[39m1\u001b[39m:,:]\n\u001b[1;32m    102\u001b[0m smi \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mpe(smi)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/sparse.py:161\u001b[0m, in \u001b[0;36mEmbedding.forward\u001b[0;34m(self, input)\u001b[0m\n\u001b[1;32m    159\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mforward\u001b[39m(\u001b[39mself\u001b[39m, \u001b[39minput\u001b[39m: Tensor) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m Tensor:\n\u001b[1;32m    160\u001b[0m     \u001b[39mreturn\u001b[39;00m F\u001b[39m.\u001b[39membedding(\n\u001b[0;32m--> 161\u001b[0m         \u001b[39minput\u001b[39m, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mweight, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mpadding_idx, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmax_norm,\n\u001b[1;32m    162\u001b[0m         \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnorm_type, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mscale_grad_by_freq, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39msparse)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1256\u001b[0m, in \u001b[0;36mModule.__getattr__\u001b[0;34m(self, name)\u001b[0m\n\u001b[1;32m   1253\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39m'\u001b[39m\u001b[39m_is_full_backward_hook\u001b[39m\u001b[39m'\u001b[39m \u001b[39mnot\u001b[39;00m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m:\n\u001b[1;32m   1254\u001b[0m         \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_is_full_backward_hook \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m-> 1256\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m__getattr__\u001b[39m(\u001b[39mself\u001b[39m, name: \u001b[39mstr\u001b[39m) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m Union[Tensor, \u001b[39m'\u001b[39m\u001b[39mModule\u001b[39m\u001b[39m'\u001b[39m]:\n\u001b[1;32m   1257\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39m'\u001b[39m\u001b[39m_parameters\u001b[39m\u001b[39m'\u001b[39m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m:\n\u001b[1;32m   1258\u001b[0m         _parameters \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m[\u001b[39m'\u001b[39m\u001b[39m_parameters\u001b[39m\u001b[39m'\u001b[39m]\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "20it [16:00, 48.01s/it]\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[41], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m acc, pred_smiles_list, og_smiles_list \u001b[39m=\u001b[39m calculate_decoder_accuracy(model, dataloaders, k\u001b[39m=\u001b[39;49m\u001b[39m5\u001b[39;49m)\n",
      "Cell \u001b[0;32mIn[38], line 10\u001b[0m, in \u001b[0;36mcalculate_decoder_accuracy\u001b[0;34m(model, dataloaders, k)\u001b[0m\n\u001b[1;32m      8\u001b[0m spec_latents \u001b[39m=\u001b[39m model\u001b[39m.\u001b[39mmodule\u001b[39m.\u001b[39mforward_spec(data)\n\u001b[1;32m      9\u001b[0m \u001b[39mfor\u001b[39;00m spec, og \u001b[39min\u001b[39;00m \u001b[39mzip\u001b[39m(spec_latents, data[\u001b[39m'\u001b[39m\u001b[39msmiles\u001b[39m\u001b[39m'\u001b[39m] ):\n\u001b[0;32m---> 10\u001b[0m     ls \u001b[39m=\u001b[39m sampler\u001b[39m.\u001b[39;49msample_multi(k,spec)\n\u001b[1;32m     11\u001b[0m     generated_smiles \u001b[39m=\u001b[39m []\n\u001b[1;32m     12\u001b[0m     \u001b[39mfor\u001b[39;00m smi \u001b[39min\u001b[39;00m ls:\n",
      "File \u001b[0;32m~/CLIP_PART_1/train_utils.py:408\u001b[0m, in \u001b[0;36mSampler.sample_multi\u001b[0;34m(self, n, embed)\u001b[0m\n\u001b[1;32m    406\u001b[0m smiles_list \u001b[39m=\u001b[39m []\n\u001b[1;32m    407\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(n):\n\u001b[0;32m--> 408\u001b[0m     smiles_list\u001b[39m.\u001b[39mappend(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49msample(embed))\n\u001b[1;32m    409\u001b[0m \u001b[39mreturn\u001b[39;00m smiles_list\n",
      "File \u001b[0;32m~/CLIP_PART_1/train_utils.py:391\u001b[0m, in \u001b[0;36mSampler.sample\u001b[0;34m(self, embed)\u001b[0m\n\u001b[1;32m    389\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(\u001b[39m0\u001b[39m,\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmax_len\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m):\n\u001b[1;32m    390\u001b[0m     tensor \u001b[39m=\u001b[39m sample_tensor[:,:i\u001b[39m+\u001b[39m\u001b[39m1\u001b[39m]\n\u001b[0;32m--> 391\u001b[0m     logits \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mmodel\u001b[39m.\u001b[39;49mforward(embed, tensor)[:,\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m,:]\n\u001b[1;32m    392\u001b[0m     probabilities \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39msoftmax(logits, dim\u001b[39m=\u001b[39m\u001b[39m1\u001b[39m)\n\u001b[1;32m    393\u001b[0m     sampled_char \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39mmultinomial(probabilities,\u001b[39m1\u001b[39m)\u001b[39m.\u001b[39mitem()\n",
      "File \u001b[0;32m~/CLIP_PART_1/models/decoder.py:109\u001b[0m, in \u001b[0;36mLatentToMol.forward\u001b[0;34m(self, spec, smi, tgt_mask, tgt_padding_mask)\u001b[0m\n\u001b[1;32m    105\u001b[0m \u001b[39m# smi = self.pe(smi) # [batch, seqlen, hidden]\u001b[39;00m\n\u001b[1;32m    107\u001b[0m mask \u001b[39m=\u001b[39m set_up_causal_mask(x\u001b[39m.\u001b[39mshape[\u001b[39m1\u001b[39m])\u001b[39m.\u001b[39mto(smi\u001b[39m.\u001b[39mdevice)\n\u001b[0;32m--> 109\u001b[0m x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mtrfmencoder(\n\u001b[1;32m    110\u001b[0m     src\u001b[39m=\u001b[39;49mx,\n\u001b[1;32m    111\u001b[0m     mask \u001b[39m=\u001b[39;49m mask\n\u001b[1;32m    112\u001b[0m )\n\u001b[1;32m    113\u001b[0m x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mln_f(x)\n\u001b[1;32m    114\u001b[0m out \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mclassifier(x)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:280\u001b[0m, in \u001b[0;36mTransformerEncoder.forward\u001b[0;34m(self, src, mask, src_key_padding_mask)\u001b[0m\n\u001b[1;32m    277\u001b[0m         src_key_padding_mask_for_layers \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[1;32m    279\u001b[0m \u001b[39mfor\u001b[39;00m mod \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mlayers:\n\u001b[0;32m--> 280\u001b[0m     output \u001b[39m=\u001b[39m mod(output, src_mask\u001b[39m=\u001b[39;49mmask, src_key_padding_mask\u001b[39m=\u001b[39;49msrc_key_padding_mask_for_layers)\n\u001b[1;32m    282\u001b[0m \u001b[39mif\u001b[39;00m convert_to_nested:\n\u001b[1;32m    283\u001b[0m     output \u001b[39m=\u001b[39m output\u001b[39m.\u001b[39mto_padded_tensor(\u001b[39m0.\u001b[39m)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:535\u001b[0m, in \u001b[0;36mTransformerEncoderLayer.forward\u001b[0;34m(self, src, src_mask, src_key_padding_mask)\u001b[0m\n\u001b[1;32m    533\u001b[0m x \u001b[39m=\u001b[39m src\n\u001b[1;32m    534\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnorm_first:\n\u001b[0;32m--> 535\u001b[0m     x \u001b[39m=\u001b[39m x \u001b[39m+\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_sa_block(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mnorm1(x), src_mask, src_key_padding_mask)\n\u001b[1;32m    536\u001b[0m     x \u001b[39m=\u001b[39m x \u001b[39m+\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_ff_block(\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnorm2(x))\n\u001b[1;32m    537\u001b[0m \u001b[39melse\u001b[39;00m:\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:546\u001b[0m, in \u001b[0;36mTransformerEncoderLayer._sa_block\u001b[0;34m(self, x, attn_mask, key_padding_mask)\u001b[0m\n\u001b[1;32m    544\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m_sa_block\u001b[39m(\u001b[39mself\u001b[39m, x: Tensor,\n\u001b[1;32m    545\u001b[0m               attn_mask: Optional[Tensor], key_padding_mask: Optional[Tensor]) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m Tensor:\n\u001b[0;32m--> 546\u001b[0m     x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mself_attn(x, x, x,\n\u001b[1;32m    547\u001b[0m                        attn_mask\u001b[39m=\u001b[39;49mattn_mask,\n\u001b[1;32m    548\u001b[0m                        key_padding_mask\u001b[39m=\u001b[39;49mkey_padding_mask,\n\u001b[1;32m    549\u001b[0m                        need_weights\u001b[39m=\u001b[39;49m\u001b[39mFalse\u001b[39;49;00m)[\u001b[39m0\u001b[39m]\n\u001b[1;32m    550\u001b[0m     \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mdropout1(x)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/activation.py:1167\u001b[0m, in \u001b[0;36mMultiheadAttention.forward\u001b[0;34m(self, query, key, value, key_padding_mask, need_weights, attn_mask, average_attn_weights)\u001b[0m\n\u001b[1;32m   1156\u001b[0m     attn_output, attn_output_weights \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39mmulti_head_attention_forward(\n\u001b[1;32m   1157\u001b[0m         query, key, value, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39membed_dim, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnum_heads,\n\u001b[1;32m   1158\u001b[0m         \u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_weight, \u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_bias,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m   1164\u001b[0m         q_proj_weight\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mq_proj_weight, k_proj_weight\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mk_proj_weight,\n\u001b[1;32m   1165\u001b[0m         v_proj_weight\u001b[39m=\u001b[39m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mv_proj_weight, average_attn_weights\u001b[39m=\u001b[39maverage_attn_weights)\n\u001b[1;32m   1166\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m-> 1167\u001b[0m     attn_output, attn_output_weights \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39;49mmulti_head_attention_forward(\n\u001b[1;32m   1168\u001b[0m         query, key, value, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49membed_dim, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mnum_heads,\n\u001b[1;32m   1169\u001b[0m         \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49min_proj_weight, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49min_proj_bias,\n\u001b[1;32m   1170\u001b[0m         \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mbias_k, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mbias_v, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49madd_zero_attn,\n\u001b[1;32m   1171\u001b[0m         \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mdropout, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mout_proj\u001b[39m.\u001b[39;49mweight, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mout_proj\u001b[39m.\u001b[39;49mbias,\n\u001b[1;32m   1172\u001b[0m         training\u001b[39m=\u001b[39;49m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mtraining,\n\u001b[1;32m   1173\u001b[0m         key_padding_mask\u001b[39m=\u001b[39;49mkey_padding_mask, need_weights\u001b[39m=\u001b[39;49mneed_weights,\n\u001b[1;32m   1174\u001b[0m         attn_mask\u001b[39m=\u001b[39;49mattn_mask, average_attn_weights\u001b[39m=\u001b[39;49maverage_attn_weights)\n\u001b[1;32m   1175\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mbatch_first \u001b[39mand\u001b[39;00m is_batched:\n\u001b[1;32m   1176\u001b[0m     \u001b[39mreturn\u001b[39;00m attn_output\u001b[39m.\u001b[39mtranspose(\u001b[39m1\u001b[39m, \u001b[39m0\u001b[39m), attn_output_weights\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/functional.py:5095\u001b[0m, in \u001b[0;36mmulti_head_attention_forward\u001b[0;34m(query, key, value, embed_dim_to_check, num_heads, in_proj_weight, in_proj_bias, bias_k, bias_v, add_zero_attn, dropout_p, out_proj_weight, out_proj_bias, training, key_padding_mask, need_weights, attn_mask, use_separate_proj_weight, q_proj_weight, k_proj_weight, v_proj_weight, static_k, static_v, average_attn_weights)\u001b[0m\n\u001b[1;32m   5090\u001b[0m     \u001b[39massert\u001b[39;00m bias_v \u001b[39mis\u001b[39;00m \u001b[39mNone\u001b[39;00m\n\u001b[1;32m   5092\u001b[0m \u001b[39m#\u001b[39;00m\n\u001b[1;32m   5093\u001b[0m \u001b[39m# reshape q, k, v for multihead attention and make em batch first\u001b[39;00m\n\u001b[1;32m   5094\u001b[0m \u001b[39m#\u001b[39;00m\n\u001b[0;32m-> 5095\u001b[0m q \u001b[39m=\u001b[39m q\u001b[39m.\u001b[39;49mcontiguous()\u001b[39m.\u001b[39mview(tgt_len, bsz \u001b[39m*\u001b[39m num_heads, head_dim)\u001b[39m.\u001b[39mtranspose(\u001b[39m0\u001b[39m, \u001b[39m1\u001b[39m)\n\u001b[1;32m   5096\u001b[0m \u001b[39mif\u001b[39;00m static_k \u001b[39mis\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n\u001b[1;32m   5097\u001b[0m     k \u001b[39m=\u001b[39m k\u001b[39m.\u001b[39mcontiguous()\u001b[39m.\u001b[39mview(k\u001b[39m.\u001b[39mshape[\u001b[39m0\u001b[39m], bsz \u001b[39m*\u001b[39m num_heads, head_dim)\u001b[39m.\u001b[39mtranspose(\u001b[39m0\u001b[39m, \u001b[39m1\u001b[39m)\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "31it [05:11, 10.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No of Hits :  4598\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Sampler():\n",
    "    def __init__(self, model, vocab):\n",
    "        self.model = model\n",
    "        self.vocab = vocab\n",
    "        self.max_len = 40\n",
    "    def sample(self, embed, greedy_decode=False):\n",
    "        embed = embed.unsqueeze(0).to(device)\n",
    "        self.model.eval()\n",
    "        sample_tensor = torch.zeros((1,self.max_len), dtype=torch.int64).to(device)\n",
    "        sample_tensor[0,0] = SOS\n",
    "        with torch.no_grad():\n",
    "            for i in range(0,self.max_len-1):\n",
    "                tensor = sample_tensor[:,:i+1]\n",
    "                logits = self.model.forward(embed, tensor)[:,-1,:]\n",
    "                probabilities = F.softmax(logits, dim=1)\n",
    "                sampled_char = torch.multinomial(probabilities,1).item()\n",
    "                if greedy_decode:\n",
    "                    sampled_char = torch.argmax(probabilities)\n",
    "                    \n",
    "                sample_tensor[0,i+1] = sampled_char\n",
    "                if sampled_char == EOS:\n",
    "                    break\n",
    "            smiles = \"\"\n",
    "            chars = self.vocab.from_seq(sample_tensor[0])\n",
    "            for char in chars:\n",
    "                if char != \"<pad>\" and char != \"<eos>\" and char != \"<sos>\" and char != \"<unk>\":\n",
    "                    smiles += char\n",
    "            \n",
    "        return smiles\n",
    "  \n",
    "    def sample_multi(self, n, embed, greedy_decode=False):\n",
    "        smiles_list = []\n",
    "        for i in range(n):\n",
    "            smiles_list.append(self.sample(embed, greedy_decode))\n",
    "        return smiles_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2it [00:22, 11.25s/it]\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[43], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m acc, pred_smiles_list, og_smiles_list \u001b[39m=\u001b[39m calculate_decoder_accuracy(model, dataloaders, k\u001b[39m=\u001b[39;49m\u001b[39m1\u001b[39;49m)\n",
      "Cell \u001b[0;32mIn[38], line 10\u001b[0m, in \u001b[0;36mcalculate_decoder_accuracy\u001b[0;34m(model, dataloaders, k)\u001b[0m\n\u001b[1;32m      8\u001b[0m spec_latents \u001b[39m=\u001b[39m model\u001b[39m.\u001b[39mmodule\u001b[39m.\u001b[39mforward_spec(data)\n\u001b[1;32m      9\u001b[0m \u001b[39mfor\u001b[39;00m spec, og \u001b[39min\u001b[39;00m \u001b[39mzip\u001b[39m(spec_latents, data[\u001b[39m'\u001b[39m\u001b[39msmiles\u001b[39m\u001b[39m'\u001b[39m] ):\n\u001b[0;32m---> 10\u001b[0m     ls \u001b[39m=\u001b[39m sampler\u001b[39m.\u001b[39;49msample_multi(k,spec)\n\u001b[1;32m     11\u001b[0m     generated_smiles \u001b[39m=\u001b[39m []\n\u001b[1;32m     12\u001b[0m     \u001b[39mfor\u001b[39;00m smi \u001b[39min\u001b[39;00m ls:\n",
      "File \u001b[0;32m~/CLIP_PART_1/train_utils.py:408\u001b[0m, in \u001b[0;36mSampler.sample_multi\u001b[0;34m(self, n, embed)\u001b[0m\n\u001b[1;32m    406\u001b[0m smiles_list \u001b[39m=\u001b[39m []\n\u001b[1;32m    407\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(n):\n\u001b[0;32m--> 408\u001b[0m     smiles_list\u001b[39m.\u001b[39mappend(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49msample(embed))\n\u001b[1;32m    409\u001b[0m \u001b[39mreturn\u001b[39;00m smiles_list\n",
      "File \u001b[0;32m~/CLIP_PART_1/train_utils.py:391\u001b[0m, in \u001b[0;36mSampler.sample\u001b[0;34m(self, embed)\u001b[0m\n\u001b[1;32m    389\u001b[0m \u001b[39mfor\u001b[39;00m i \u001b[39min\u001b[39;00m \u001b[39mrange\u001b[39m(\u001b[39m0\u001b[39m,\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mmax_len\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m):\n\u001b[1;32m    390\u001b[0m     tensor \u001b[39m=\u001b[39m sample_tensor[:,:i\u001b[39m+\u001b[39m\u001b[39m1\u001b[39m]\n\u001b[0;32m--> 391\u001b[0m     logits \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mmodel\u001b[39m.\u001b[39;49mforward(embed, tensor)[:,\u001b[39m-\u001b[39m\u001b[39m1\u001b[39m,:]\n\u001b[1;32m    392\u001b[0m     probabilities \u001b[39m=\u001b[39m F\u001b[39m.\u001b[39msoftmax(logits, dim\u001b[39m=\u001b[39m\u001b[39m1\u001b[39m)\n\u001b[1;32m    393\u001b[0m     sampled_char \u001b[39m=\u001b[39m torch\u001b[39m.\u001b[39mmultinomial(probabilities,\u001b[39m1\u001b[39m)\u001b[39m.\u001b[39mitem()\n",
      "File \u001b[0;32m~/CLIP_PART_1/models/decoder.py:109\u001b[0m, in \u001b[0;36mLatentToMol.forward\u001b[0;34m(self, spec, smi, tgt_mask, tgt_padding_mask)\u001b[0m\n\u001b[1;32m    105\u001b[0m \u001b[39m# smi = self.pe(smi) # [batch, seqlen, hidden]\u001b[39;00m\n\u001b[1;32m    107\u001b[0m mask \u001b[39m=\u001b[39m set_up_causal_mask(x\u001b[39m.\u001b[39mshape[\u001b[39m1\u001b[39m])\u001b[39m.\u001b[39mto(smi\u001b[39m.\u001b[39mdevice)\n\u001b[0;32m--> 109\u001b[0m x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mtrfmencoder(\n\u001b[1;32m    110\u001b[0m     src\u001b[39m=\u001b[39;49mx,\n\u001b[1;32m    111\u001b[0m     mask \u001b[39m=\u001b[39;49m mask\n\u001b[1;32m    112\u001b[0m )\n\u001b[1;32m    113\u001b[0m x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mln_f(x)\n\u001b[1;32m    114\u001b[0m out \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mclassifier(x)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:280\u001b[0m, in \u001b[0;36mTransformerEncoder.forward\u001b[0;34m(self, src, mask, src_key_padding_mask)\u001b[0m\n\u001b[1;32m    277\u001b[0m         src_key_padding_mask_for_layers \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[1;32m    279\u001b[0m \u001b[39mfor\u001b[39;00m mod \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mlayers:\n\u001b[0;32m--> 280\u001b[0m     output \u001b[39m=\u001b[39m mod(output, src_mask\u001b[39m=\u001b[39;49mmask, src_key_padding_mask\u001b[39m=\u001b[39;49msrc_key_padding_mask_for_layers)\n\u001b[1;32m    282\u001b[0m \u001b[39mif\u001b[39;00m convert_to_nested:\n\u001b[1;32m    283\u001b[0m     output \u001b[39m=\u001b[39m output\u001b[39m.\u001b[39mto_padded_tensor(\u001b[39m0.\u001b[39m)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:535\u001b[0m, in \u001b[0;36mTransformerEncoderLayer.forward\u001b[0;34m(self, src, src_mask, src_key_padding_mask)\u001b[0m\n\u001b[1;32m    533\u001b[0m x \u001b[39m=\u001b[39m src\n\u001b[1;32m    534\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnorm_first:\n\u001b[0;32m--> 535\u001b[0m     x \u001b[39m=\u001b[39m x \u001b[39m+\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_sa_block(\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mnorm1(x), src_mask, src_key_padding_mask)\n\u001b[1;32m    536\u001b[0m     x \u001b[39m=\u001b[39m x \u001b[39m+\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_ff_block(\u001b[39mself\u001b[39m\u001b[39m.\u001b[39mnorm2(x))\n\u001b[1;32m    537\u001b[0m \u001b[39melse\u001b[39;00m:\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/transformer.py:546\u001b[0m, in \u001b[0;36mTransformerEncoderLayer._sa_block\u001b[0;34m(self, x, attn_mask, key_padding_mask)\u001b[0m\n\u001b[1;32m    544\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m_sa_block\u001b[39m(\u001b[39mself\u001b[39m, x: Tensor,\n\u001b[1;32m    545\u001b[0m               attn_mask: Optional[Tensor], key_padding_mask: Optional[Tensor]) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m Tensor:\n\u001b[0;32m--> 546\u001b[0m     x \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mself_attn(x, x, x,\n\u001b[1;32m    547\u001b[0m                        attn_mask\u001b[39m=\u001b[39;49mattn_mask,\n\u001b[1;32m    548\u001b[0m                        key_padding_mask\u001b[39m=\u001b[39;49mkey_padding_mask,\n\u001b[1;32m    549\u001b[0m                        need_weights\u001b[39m=\u001b[39;49m\u001b[39mFalse\u001b[39;49;00m)[\u001b[39m0\u001b[39m]\n\u001b[1;32m    550\u001b[0m     \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mdropout1(x)\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1194\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *input, **kwargs)\u001b[0m\n\u001b[1;32m   1190\u001b[0m \u001b[39m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1191\u001b[0m \u001b[39m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1192\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m (\u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_backward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_hooks \u001b[39mor\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_forward_pre_hooks \u001b[39mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1193\u001b[0m         \u001b[39mor\u001b[39;00m _global_forward_hooks \u001b[39mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1194\u001b[0m     \u001b[39mreturn\u001b[39;00m forward_call(\u001b[39m*\u001b[39;49m\u001b[39minput\u001b[39;49m, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m   1195\u001b[0m \u001b[39m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1196\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[39m=\u001b[39m [], []\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/activation.py:1077\u001b[0m, in \u001b[0;36mMultiheadAttention.forward\u001b[0;34m(self, query, key, value, key_padding_mask, need_weights, attn_mask, average_attn_weights)\u001b[0m\n\u001b[1;32m   1072\u001b[0m \u001b[39melif\u001b[39;00m query \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m key \u001b[39mor\u001b[39;00m key \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m value:\n\u001b[1;32m   1073\u001b[0m     \u001b[39m# When lifting this restriction, don't forget to either\u001b[39;00m\n\u001b[1;32m   1074\u001b[0m     \u001b[39m# enforce that the dtypes all match or test cases where\u001b[39;00m\n\u001b[1;32m   1075\u001b[0m     \u001b[39m# they don't!\u001b[39;00m\n\u001b[1;32m   1076\u001b[0m     why_not_fast_path \u001b[39m=\u001b[39m \u001b[39m\"\u001b[39m\u001b[39mnon-self attention was used (query, key, and value are not the same Tensor)\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[0;32m-> 1077\u001b[0m \u001b[39melif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_bias \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m \u001b[39mand\u001b[39;00m query\u001b[39m.\u001b[39mdtype \u001b[39m!=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49min_proj_bias\u001b[39m.\u001b[39mdtype:\n\u001b[1;32m   1078\u001b[0m     why_not_fast_path \u001b[39m=\u001b[39m \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mdtypes of query (\u001b[39m\u001b[39m{\u001b[39;00mquery\u001b[39m.\u001b[39mdtype\u001b[39m}\u001b[39;00m\u001b[39m) and self.in_proj_bias (\u001b[39m\u001b[39m{\u001b[39;00m\u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_bias\u001b[39m.\u001b[39mdtype\u001b[39m}\u001b[39;00m\u001b[39m) don\u001b[39m\u001b[39m'\u001b[39m\u001b[39mt match\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m   1079\u001b[0m \u001b[39melif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_weight \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m \u001b[39mand\u001b[39;00m query\u001b[39m.\u001b[39mdtype \u001b[39m!=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39min_proj_weight\u001b[39m.\u001b[39mdtype:\n\u001b[1;32m   1080\u001b[0m     \u001b[39m# this case will fail anyway, but at least they'll get a useful error message.\u001b[39;00m\n",
      "File \u001b[0;32m~/miniconda3/envs/sbdd-env/lib/python3.10/site-packages/torch/nn/modules/module.py:1256\u001b[0m, in \u001b[0;36mModule.__getattr__\u001b[0;34m(self, name)\u001b[0m\n\u001b[1;32m   1253\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39m'\u001b[39m\u001b[39m_is_full_backward_hook\u001b[39m\u001b[39m'\u001b[39m \u001b[39mnot\u001b[39;00m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m:\n\u001b[1;32m   1254\u001b[0m         \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_is_full_backward_hook \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m-> 1256\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m__getattr__\u001b[39m(\u001b[39mself\u001b[39m, name: \u001b[39mstr\u001b[39m) \u001b[39m-\u001b[39m\u001b[39m>\u001b[39m Union[Tensor, \u001b[39m'\u001b[39m\u001b[39mModule\u001b[39m\u001b[39m'\u001b[39m]:\n\u001b[1;32m   1257\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39m'\u001b[39m\u001b[39m_parameters\u001b[39m\u001b[39m'\u001b[39m \u001b[39min\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m:\n\u001b[1;32m   1258\u001b[0m         _parameters \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m\u001b[39m__dict__\u001b[39m[\u001b[39m'\u001b[39m\u001b[39m_parameters\u001b[39m\u001b[39m'\u001b[39m]\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "acc, pred_smiles_list, og_smiles_list = calculate_decoder_accuracy(model, dataloaders, k=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Percentage valid 0.9855143229166666\n"
     ]
    }
   ],
   "source": [
    "parsed_mols = np.array([Chem.MolFromSmiles(s) for s in smiles_list])\n",
    "print(\"Percentage valid\", (parsed_mols != None).sum()/len(parsed_mols))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.3761393229166667\n"
     ]
    }
   ],
   "source": [
    "count = 0\n",
    "for og_smile, pred_smile in zip(og_smiles_list, smiles_list):\n",
    "    og = Chem.CanonSmiles(og_smile)\n",
    "    try:\n",
    "        pred= Chem.CanonSmiles(pred_smile)\n",
    "        if og == pred:\n",
    "            count += 1\n",
    "    except:\n",
    "        pass\n",
    "print(count / len(og_smiles_list))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.PngImagePlugin.PngImageFile image mode=RGB size=2200x200>"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Draw.MolsToGridImage(np.array([Chem.MolFromSmiles(smi) for smi in og_smiles_list])[5000:5010], molsPerRow=11, returnPNG=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.PngImagePlugin.PngImageFile image mode=RGB size=2200x200>"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Draw.MolsToGridImage(np.array([Chem.MolFromSmiles(smi) for smi in smiles_list])[5000:5010], molsPerRow=11, returnPNG=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "14"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(og_smiles)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "sbdd-env",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.8"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}