1_training.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "d34ee45f-fb11-480f-a48d-17119a599e6a",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "import torch\n",
    "from torch.utils.data import DataLoader\n",
    "from tqdm import tqdm, trange\n",
    "from moleval.metrics.metrics import GetMetrics\n",
    "from rdkit import Chem, RDLogger\n",
    "RDLogger.DisableLog('rdApp.*')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "b05f04db-59f3-4034-a6c6-1429a36d39ef",
   "metadata": {},
   "outputs": [],
   "source": [
    "TASKNAME = 'gdb13'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "88f82ed7-be7c-4e82-b4e1-e79015240dee",
   "metadata": {},
   "outputs": [],
   "source": [
    "output_dir = os.path.join('results', TASKNAME)\n",
    "if not os.path.exists(output_dir):\n",
    "    os.mkdir(output_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "88361cea-a1b7-49ad-b1f4-10441194c621",
   "metadata": {},
   "outputs": [],
   "source": [
    "use_cuda = torch.cuda.is_available()\n",
    "device = torch.device(\"cuda\" if use_cuda else \"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "89d3a46e-8ff3-499f-b62b-40f0bad7c70d",
   "metadata": {},
   "outputs": [],
   "source": [
    "from molbit.dataset import SmilesDataset, SmilesEncoder\n",
    "from molbit.gumbel import GumbelSmilesVAE\n",
    "from molbit.vae import CyclicalAnnealingScheduler"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b31c46a1-e47b-477d-b4a6-86d8d5773637",
   "metadata": {},
   "source": [
    "# Data Loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "55653871-27c9-4b0a-972e-32c165ae86f4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of SMILES (raw): 1000000\n",
      "Maximum of seqlen: 37\n",
      "Vocabulary size (including 3 special tokens): 23\n"
     ]
    }
   ],
   "source": [
    "dataset = SmilesDataset(os.path.join('data', 'gdb13.csv'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "2a45c6d2-a6dc-4d86-bb55-69dd3d4b299c",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset.save_char2idx(os.path.join(output_dir, 'vocabs.csv'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "50fa3da3-6ed7-41e1-a19a-ac10485cb90d",
   "metadata": {},
   "source": [
    "# Model Initialization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "039658ee-962e-480c-9d6c-b20dd3bb7503",
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = GumbelSmilesVAE(vocab_size=dataset.vocab_size,\n",
    "                            embedding_size=16,\n",
    "                            hidden_size=256,\n",
    "                            latent_size=64,\n",
    "                            sos_idx=dataset.sos_idx,\n",
    "                            eos_idx=dataset.eos_idx,\n",
    "                            pad_idx=dataset.pad_idx,\n",
    "                            device=device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e750eeee-8aa6-4138-8314-831a04c4f3c5",
   "metadata": {},
   "outputs": [],
   "source": [
    "generator.save_configs(os.path.join(output_dir, 'configs.csv'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7852051a-d38c-498d-bf44-e3b73bd2370c",
   "metadata": {},
   "source": [
    "# Model Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "93556bbc-5e6a-4bef-b8be-ba9d1445c2fe",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0/62500 (0%)] loss_vae: 3.138  loss_rec: 3.138  loss_kld: 0.001  beta: 0.000\n",
      "[1000/62500 (2%)] loss_vae: 1.096  loss_rec: 1.096  loss_kld: 0.693  beta: 0.000\n",
      "[2000/62500 (3%)] loss_vae: 1.050  loss_rec: 1.050  loss_kld: 0.693  beta: 0.000\n",
      "[3000/62500 (5%)] loss_vae: 1.037  loss_rec: 1.037  loss_kld: 0.693  beta: 0.000\n",
      "[4000/62500 (6%)] loss_vae: 1.045  loss_rec: 1.036  loss_kld: 0.693  beta: 0.012\n",
      "[5000/62500 (8%)] loss_vae: 0.855  loss_rec: 0.780  loss_kld: 0.536  beta: 0.140\n",
      "[6000/62500 (10%)] loss_vae: 0.758  loss_rec: 0.639  loss_kld: 0.444  beta: 0.268\n",
      "[7000/62500 (11%)] loss_vae: 0.737  loss_rec: 0.579  loss_kld: 0.398  beta: 0.396\n",
      "[8000/62500 (13%)] loss_vae: 0.802  loss_rec: 0.595  loss_kld: 0.394  beta: 0.524\n",
      "[9000/62500 (14%)] loss_vae: 0.818  loss_rec: 0.578  loss_kld: 0.368  beta: 0.652\n",
      "[10000/62500 (16%)] loss_vae: 0.762  loss_rec: 0.501  loss_kld: 0.334  beta: 0.780\n",
      "[11000/62500 (18%)] loss_vae: 0.820  loss_rec: 0.543  loss_kld: 0.305  beta: 0.908\n",
      "[12000/62500 (19%)] loss_vae: 0.828  loss_rec: 0.543  loss_kld: 0.285  beta: 1.000\n",
      "[13000/62500 (21%)] loss_vae: 0.843  loss_rec: 0.558  loss_kld: 0.284  beta: 1.000\n",
      "[14000/62500 (22%)] loss_vae: 0.874  loss_rec: 0.586  loss_kld: 0.288  beta: 1.000\n",
      "[15000/62500 (24%)] loss_vae: 0.799  loss_rec: 0.514  loss_kld: 0.285  beta: 1.000\n",
      "[16000/62500 (26%)] loss_vae: 0.474  loss_rec: 0.474  loss_kld: 0.472  beta: 0.000\n",
      "[17000/62500 (27%)] loss_vae: 0.386  loss_rec: 0.386  loss_kld: 0.548  beta: 0.000\n",
      "[18000/62500 (29%)] loss_vae: 0.396  loss_rec: 0.396  loss_kld: 0.575  beta: 0.000\n",
      "[19000/62500 (30%)] loss_vae: 0.351  loss_rec: 0.351  loss_kld: 0.573  beta: 0.000\n",
      "[20000/62500 (32%)] loss_vae: 0.367  loss_rec: 0.331  loss_kld: 0.596  beta: 0.060\n",
      "[21000/62500 (34%)] loss_vae: 0.438  loss_rec: 0.328  loss_kld: 0.590  beta: 0.188\n",
      "[22000/62500 (35%)] loss_vae: 0.570  loss_rec: 0.381  loss_kld: 0.599  beta: 0.316\n",
      "[23000/62500 (37%)] loss_vae: 0.579  loss_rec: 0.323  loss_kld: 0.579  beta: 0.444\n",
      "[24000/62500 (38%)] loss_vae: 0.668  loss_rec: 0.346  loss_kld: 0.562  beta: 0.572\n",
      "[25000/62500 (40%)] loss_vae: 0.689  loss_rec: 0.317  loss_kld: 0.531  beta: 0.700\n",
      "[26000/62500 (42%)] loss_vae: 0.730  loss_rec: 0.333  loss_kld: 0.479  beta: 0.828\n",
      "[27000/62500 (43%)] loss_vae: 0.757  loss_rec: 0.355  loss_kld: 0.421  beta: 0.956\n",
      "[28000/62500 (45%)] loss_vae: 0.746  loss_rec: 0.363  loss_kld: 0.383  beta: 1.000\n",
      "[29000/62500 (46%)] loss_vae: 0.778  loss_rec: 0.400  loss_kld: 0.378  beta: 1.000\n",
      "[30000/62500 (48%)] loss_vae: 0.770  loss_rec: 0.400  loss_kld: 0.371  beta: 1.000\n",
      "[31000/62500 (50%)] loss_vae: 0.736  loss_rec: 0.362  loss_kld: 0.374  beta: 1.000\n",
      "[32000/62500 (51%)] loss_vae: 0.296  loss_rec: 0.296  loss_kld: 0.561  beta: 0.000\n",
      "[33000/62500 (53%)] loss_vae: 0.282  loss_rec: 0.282  loss_kld: 0.568  beta: 0.000\n",
      "[34000/62500 (54%)] loss_vae: 0.250  loss_rec: 0.249  loss_kld: 0.593  beta: 0.000\n",
      "[35000/62500 (56%)] loss_vae: 0.248  loss_rec: 0.248  loss_kld: 0.600  beta: 0.000\n",
      "[36000/62500 (58%)] loss_vae: 0.320  loss_rec: 0.255  loss_kld: 0.604  beta: 0.108\n",
      "[37000/62500 (59%)] loss_vae: 0.401  loss_rec: 0.258  loss_kld: 0.604  beta: 0.236\n",
      "[38000/62500 (61%)] loss_vae: 0.442  loss_rec: 0.228  loss_kld: 0.588  beta: 0.364\n",
      "[39000/62500 (62%)] loss_vae: 0.567  loss_rec: 0.278  loss_kld: 0.587  beta: 0.492\n",
      "[40000/62500 (64%)] loss_vae: 0.555  loss_rec: 0.204  loss_kld: 0.567  beta: 0.620\n",
      "[41000/62500 (66%)] loss_vae: 0.640  loss_rec: 0.249  loss_kld: 0.523  beta: 0.748\n",
      "[42000/62500 (67%)] loss_vae: 0.684  loss_rec: 0.253  loss_kld: 0.491  beta: 0.876\n",
      "[43000/62500 (69%)] loss_vae: 0.697  loss_rec: 0.250  loss_kld: 0.447  beta: 1.000\n",
      "[44000/62500 (70%)] loss_vae: 0.732  loss_rec: 0.309  loss_kld: 0.423  beta: 1.000\n",
      "[45000/62500 (72%)] loss_vae: 0.700  loss_rec: 0.285  loss_kld: 0.414  beta: 1.000\n",
      "[46000/62500 (74%)] loss_vae: 0.689  loss_rec: 0.274  loss_kld: 0.415  beta: 1.000\n",
      "[47000/62500 (75%)] loss_vae: 0.226  loss_rec: 0.226  loss_kld: 0.510  beta: 0.000\n",
      "[48000/62500 (77%)] loss_vae: 0.192  loss_rec: 0.192  loss_kld: 0.578  beta: 0.000\n",
      "[49000/62500 (78%)] loss_vae: 0.214  loss_rec: 0.214  loss_kld: 0.593  beta: 0.000\n",
      "[50000/62500 (80%)] loss_vae: 0.198  loss_rec: 0.197  loss_kld: 0.594  beta: 0.000\n",
      "[51000/62500 (82%)] loss_vae: 0.195  loss_rec: 0.178  loss_kld: 0.610  beta: 0.028\n",
      "[52000/62500 (83%)] loss_vae: 0.288  loss_rec: 0.193  loss_kld: 0.607  beta: 0.156\n",
      "[53000/62500 (85%)] loss_vae: 0.369  loss_rec: 0.200  loss_kld: 0.597  beta: 0.284\n",
      "[54000/62500 (86%)] loss_vae: 0.471  loss_rec: 0.228  loss_kld: 0.590  beta: 0.412\n",
      "[55000/62500 (88%)] loss_vae: 0.485  loss_rec: 0.166  loss_kld: 0.590  beta: 0.540\n",
      "[56000/62500 (90%)] loss_vae: 0.584  loss_rec: 0.211  loss_kld: 0.558  beta: 0.668\n",
      "[57000/62500 (91%)] loss_vae: 0.597  loss_rec: 0.193  loss_kld: 0.508  beta: 0.796\n",
      "[58000/62500 (93%)] loss_vae: 0.682  loss_rec: 0.237  loss_kld: 0.481  beta: 0.924\n",
      "[59000/62500 (94%)] loss_vae: 0.678  loss_rec: 0.226  loss_kld: 0.452  beta: 1.000\n",
      "[60000/62500 (96%)] loss_vae: 0.623  loss_rec: 0.176  loss_kld: 0.448  beta: 1.000\n",
      "[61000/62500 (98%)] loss_vae: 0.676  loss_rec: 0.252  loss_kld: 0.425  beta: 1.000\n",
      "[62000/62500 (99%)] loss_vae: 0.632  loss_rec: 0.195  loss_kld: 0.437  beta: 1.000\n"
     ]
    }
   ],
   "source": [
    "history = []\n",
    "\n",
    "loader = DataLoader(dataset, batch_size=32, shuffle=True)\n",
    "num_batch = len(loader)\n",
    "num_epoch = 2\n",
    "num_steps = num_epoch * num_batch\n",
    "scheduler = CyclicalAnnealingScheduler(num_steps, M=4) # M: number of cycles\n",
    "\n",
    "for i in range(num_epoch):\n",
    "    for j, batch in enumerate(loader):\n",
    "        ## kld scheduling\n",
    "        step = i*num_batch + j\n",
    "        beta = scheduler(step)\n",
    "        \n",
    "        ## input tensor\n",
    "        X_lgt = batch['length'].to(device)\n",
    "        X_smi = dataset.encode(batch['smiles'], X_lgt.max()).to(device)\n",
    "        \n",
    "        ## fit\n",
    "        loss_vae, loss_rec, loss_kld = generator.partial_fit(X_smi, X_lgt, beta)\n",
    "        history.append({'loss_vae':loss_vae, 'loss_rec':loss_rec, 'loss_kld':loss_kld, 'beta':beta})\n",
    "    \n",
    "        ## logging\n",
    "        if step % 1000 == 0:\n",
    "            print(f'[{step}/{num_steps} ({step/num_steps*100:.0f}%)] loss_vae: {loss_vae:.3f}  loss_rec: {loss_rec:.3f}  loss_kld: {loss_kld:.3f}  beta: {beta:.3f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "2fbe0178-32a9-4725-9135-f873436a29ce",
   "metadata": {},
   "outputs": [],
   "source": [
    "generator.save_model(os.path.join(output_dir, 'ckpt.pt'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "176de8d0-23a2-4618-affc-ef14e90a8282",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_history = pd.DataFrame(history)\n",
    "df_history.to_csv(os.path.join(output_dir, 'history.csv'), index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "fecb94cf-bb81-400d-be77-e779b910a85b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'steps')"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.set_theme(style='whitegrid')\n",
    "ax = sns.lineplot(x=df_history.index, y=df_history['loss_vae'], err_style=None)\n",
    "ax = sns.lineplot(x=df_history.index, y=df_history['loss_rec'], err_style=None, ax=ax)\n",
    "ax = sns.lineplot(x=df_history.index, y=df_history['loss_kld'], err_style=None, ax=ax)\n",
    "\n",
    "ax.legend(['loss_vae', 'loss_rec', 'loss_kld'])\n",
    "ax.set_ylabel('loss')\n",
    "ax.set_xlabel('steps')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "57731914-14ce-41d2-8d78-5f79172f226f",
   "metadata": {},
   "source": [
    "# Random Sampling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "9d06f05d-44ac-4adc-a6f0-ed0a4f31a195",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Vocabulary size (including 3 special tokens): 23\n"
     ]
    }
   ],
   "source": [
    "tokenizer = SmilesEncoder((os.path.join(output_dir, 'vocabs.csv')))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "f872c14c-eaba-418c-9546-bea294deb3ab",
   "metadata": {},
   "outputs": [],
   "source": [
    "new_generator = GumbelSmilesVAE(vocab_size=tokenizer.vocab_size,\n",
    "                                embedding_size=16,\n",
    "                                hidden_size=256,\n",
    "                                latent_size=64,\n",
    "                                sos_idx=tokenizer.sos_idx,\n",
    "                                eos_idx=tokenizer.eos_idx,\n",
    "                                pad_idx=tokenizer.pad_idx,\n",
    "                                device=device)\n",
    "\n",
    "new_generator.load_model(os.path.join(output_dir, 'ckpt.pt'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "50c7b00a-f383-4844-af33-7cb0cf9346fd",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10000/10000 [00:52<00:00, 191.78it/s]\n"
     ]
    }
   ],
   "source": [
    "rng = np.random.RandomState(2025)\n",
    "latent_vectors = (rng.randn(10000,64) > 0).astype(np.float32)\n",
    "\n",
    "generated = []\n",
    "with torch.no_grad():\n",
    "    for z in tqdm(latent_vectors):\n",
    "        z = torch.tensor(z, device=device)\n",
    "        smi = new_generator.inference(z)\n",
    "        smi = tokenizer.decode([smi])[0]\n",
    "        generated.append(smi)\n",
    "\n",
    "df_gen = pd.DataFrame(generated, columns=['smiles'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5a0edd27-840a-4381-a4eb-950941d36c7b",
   "metadata": {},
   "source": [
    "# Validity Check"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "bac89499-e14c-4af8-a91f-0c245d3d17c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_gen.loc[:,'canonical'] = ''\n",
    "df_gen.loc[:,'valid'] = 0\n",
    "df_gen = df_gen.astype({'valid':int})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "2b15fb6d-8ae2-4ce6-8d34-f69075feb7d0",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10000/10000 [00:01<00:00, 5301.12it/s]\n"
     ]
    }
   ],
   "source": [
    "for i in trange(len(df_gen)):\n",
    "    smi = df_gen.loc[i,'smiles']\n",
    "    mol = Chem.MolFromSmiles(smi)\n",
    "    if mol:\n",
    "        df_gen.loc[i,'canonical'] = Chem.MolToSmiles(mol, kekuleSmiles=True)\n",
    "        df_gen.loc[i,'valid'] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "f04ce1ef-ad72-44d2-ac87-a36479d3f44e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>smiles</th>\n",
       "      <th>canonical</th>\n",
       "      <th>valid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>C=C1C=NOC(OCC=C)=C1C=O</td>\n",
       "      <td>C=CCOC1=C(C=O)C(=C)C=NO1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>C#CCNC1(CN(C)C)SC(=O)O1</td>\n",
       "      <td>C#CCNC1(CN(C)C)OC(=O)S1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>C=C(C(=N)C(=N)N)C1=NOCC1=O</td>\n",
       "      <td>C=C(C(=N)C(=N)N)C1=NOCC1=O</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>C=CC(C)C1(O)COC2(CCS2)C1</td>\n",
       "      <td>C=CC(C)C1(O)COC2(CCS2)C1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CC=C1OC2C3CC4CN4CC2C13</td>\n",
       "      <td>CC=C1OC2C3CC4CN4CC2C13</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9995</th>\n",
       "      <td>COCC(O)CN(C)N=CNC=N</td>\n",
       "      <td>COCC(O)CN(C)N=CNC=N</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9996</th>\n",
       "      <td>C=NN(N)C1CCOC2CCC21C</td>\n",
       "      <td>C=NN(N)C1CCOC2CCC21C</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9997</th>\n",
       "      <td>C=CC1C=CN(C=N)C1C(C)C=C</td>\n",
       "      <td>C=CC(C)C1C(C=C)C=CN1C=N</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9998</th>\n",
       "      <td>C=CC(=N)NC(C=O)C(C)=COC</td>\n",
       "      <td>C=CC(=N)NC(C=O)C(C)=COC</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9999</th>\n",
       "      <td>C=C1C2OCC3(NC4COC4O1)C23</td>\n",
       "      <td></td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>10000 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                          smiles                   canonical  valid\n",
       "0         C=C1C=NOC(OCC=C)=C1C=O    C=CCOC1=C(C=O)C(=C)C=NO1      1\n",
       "1        C#CCNC1(CN(C)C)SC(=O)O1     C#CCNC1(CN(C)C)OC(=O)S1      1\n",
       "2     C=C(C(=N)C(=N)N)C1=NOCC1=O  C=C(C(=N)C(=N)N)C1=NOCC1=O      1\n",
       "3       C=CC(C)C1(O)COC2(CCS2)C1    C=CC(C)C1(O)COC2(CCS2)C1      1\n",
       "4         CC=C1OC2C3CC4CN4CC2C13      CC=C1OC2C3CC4CN4CC2C13      1\n",
       "...                          ...                         ...    ...\n",
       "9995         COCC(O)CN(C)N=CNC=N         COCC(O)CN(C)N=CNC=N      1\n",
       "9996        C=NN(N)C1CCOC2CCC21C        C=NN(N)C1CCOC2CCC21C      1\n",
       "9997     C=CC1C=CN(C=N)C1C(C)C=C     C=CC(C)C1C(C=C)C=CN1C=N      1\n",
       "9998     C=CC(=N)NC(C=O)C(C)=COC     C=CC(=N)NC(C=O)C(C)=COC      1\n",
       "9999    C=C1C2OCC3(NC4COC4O1)C23                                  0\n",
       "\n",
       "[10000 rows x 3 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_gen"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "ad218ce0-652d-4cca-9795-80ceb0fdf016",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "valid\n",
       "1    9346\n",
       "0     654\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_gen['valid'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "d1256812-bdbc-4329-9a0b-3a5508f6d99c",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_gen.to_csv(os.path.join(output_dir, 'generated.csv'), index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7a629e49-e8ca-4b8f-a99b-d56ffabf4b83",
   "metadata": {},
   "source": [
    "# Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "02f13f0a-876c-4193-856d-7824286a33aa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cleaning up reference smiles\n"
     ]
    }
   ],
   "source": [
    "MetricEngine = GetMetrics(\n",
    "    n_jobs=20,\n",
    "    device='cpu',\n",
    "    batch_size=512,\n",
    "    run_fcd=False,\n",
    "    train=dataset.df['smiles'].values.tolist(),\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "54393a62-d450-4ac1-8929-f9569fdea80a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calculating Validity & Uniqueness\n",
      "Computing pre-statistics (fps, ring systems etc.)\n",
      "Calculating Novelty\n",
      "Calculating Diversity\n",
      "Calculating Filters\n"
     ]
    }
   ],
   "source": [
    "metrics = MetricEngine.calculate(\n",
    "    df_gen['smiles'].values.tolist(),\n",
    "    calc_valid=True,\n",
    "    calc_unique=True,\n",
    "    unique_k=10000,\n",
    "    se_k=1000,\n",
    "    sp_k=1000,\n",
    "    verbose=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "4a9dd0bc-10cf-4a80-a328-c0beb508a98d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'#': 10000,\n",
       " 'Validity': 0.9345,\n",
       " '# valid': 9345,\n",
       " 'Uniqueness': 1.0,\n",
       " '# valid & unique': 9345,\n",
       " 'Novelty': 0.9997859818084537,\n",
       " '# novel': 9343,\n",
       " 'IntDiv1': 0.9038065353801495,\n",
       " 'IntDiv2': 0.8970243566455017,\n",
       " 'SEDiv@1k': 0.993,\n",
       " 'SPDiv@1k': 0.9083269846974154,\n",
       " '# scaffolds': 4608,\n",
       " 'ScaffDiv': 0.8961942575643426,\n",
       " 'ScaffUniqueness': 0.4930979133226324,\n",
       " 'FG': 0.23449770942408377,\n",
       " 'RS': 0.45453616673477726,\n",
       " 'Filters': 0.6340288924558587,\n",
       " 'Purchasable_ZINC20': 0.0}"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "317b5ac9-e72e-4cbc-bb49-12fae820f016",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.21"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}