generative-flow-qm9.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: selfies in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (2.1.1)\n",
      "Requirement already satisfied: tensorflow_probability in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (0.19.0)\n",
      "Requirement already satisfied: numpy>=1.13.3 in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (1.22.1)\n",
      "Requirement already satisfied: cloudpickle>=1.3 in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (2.2.0)\n",
      "Requirement already satisfied: decorator in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (5.1.1)\n",
      "Requirement already satisfied: absl-py in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (1.3.0)\n",
      "Requirement already satisfied: dm-tree in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (0.1.8)\n",
      "Requirement already satisfied: six>=1.10.0 in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (1.16.0)\n",
      "Requirement already satisfied: gast>=0.3.2 in c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages (from tensorflow_probability) (0.4.0)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n",
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n",
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n",
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n",
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n",
      "WARNING: Ignoring invalid distribution -rotobuf (c:\\users\\user\\onedrive\\desktop\\github\\small-molecules\\env\\lib\\site-packages)\n"
     ]
    }
   ],
   "source": [
    "! pip install selfies tensorflow_probability"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\user\\OneDrive\\Desktop\\GitHub\\small-molecules\\env\\lib\\site-packages\\tqdm\\auto.py:22: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "c:\\Users\\user\\OneDrive\\Desktop\\GitHub\\small-molecules\\env\\lib\\site-packages\\torchvision\\models\\detection\\anchor_utils.py:63: UserWarning: Failed to initialize NumPy: module compiled against API version 0x10 but this version of numpy is 0xf (Triggered internally at ..\\torch\\csrc\\utils\\tensor_numpy.cpp:77.)\n",
      "  device: torch.device = torch.device(\"cpu\"),\n",
      "Skipped loading some Jax models, missing a dependency. jax requires jaxlib to be installed. See https://github.com/google/jax#installation for installation instructions.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import os\n",
    "\n",
    "import deepchem as dc\n",
    "from deepchem.models.normalizing_flows import NormalizingFlow, NormalizingFlowModel\n",
    "from deepchem.models.optimizers import Adam\n",
    "from deepchem.data import NumpyDataset\n",
    "from deepchem.splits import RandomSplitter\n",
    "from deepchem.molnet import load_qm9\n",
    "\n",
    "import selfies as sf\n",
    "\n",
    "import rdkit\n",
    "from rdkit import Chem\n",
    "from rdkit.Chem import Draw\n",
    "\n",
    "from IPython.display import Image, display\n",
    "\n",
    "import tensorflow as tf\n",
    "import tensorflow_probability as tfp\n",
    "\n",
    "tfd = tfp.distributions\n",
    "tfb = tfp.bijectors\n",
    "tfk = tf.keras\n",
    "\n",
    "tfk.backend.set_floatx('float64')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[H]O[C@@]1(C([H])([H])[H])C(=O)N(C([H])([H])[H...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[H]C([H])([H])C([H])([H])C([H])([H])[C@@]1([H]...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[H]C1=C([H])[C@]2([H])[N@H+]3C([H])([H])[C@]2(...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[H]C(=O)C([H])([H])[C@]1([H])[N@H+](C2([H])C([...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[H]OC([H])([H])C1([C@@]2([H])C([H])([H])[C@@]2...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                              smiles\n",
       "0  [H]O[C@@]1(C([H])([H])[H])C(=O)N(C([H])([H])[H...\n",
       "1  [H]C([H])([H])C([H])([H])C([H])([H])[C@@]1([H]...\n",
       "2  [H]C1=C([H])[C@]2([H])[N@H+]3C([H])([H])[C@]2(...\n",
       "3  [H]C(=O)C([H])([H])[C@]1([H])[N@H+](C2([H])C([...\n",
       "4  [H]OC([H])([H])C1([C@@]2([H])C([H])([H])[C@@]2..."
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tasks, datasets, transformers = load_qm9(featurizer=\"ECFP\")\n",
    "df = pd.DataFrame(data={\"smiles\" : datasets[0].ids})\n",
    "\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = df[[\"smiles\"]].sample(2500, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'H': 1,\n",
       " 'F': 1,\n",
       " 'Cl': 1,\n",
       " 'Br': 1,\n",
       " 'I': 1,\n",
       " 'B': 3,\n",
       " 'B+1': 2,\n",
       " 'B-1': 4,\n",
       " 'O': 2,\n",
       " 'O+1': 3,\n",
       " 'O-1': 1,\n",
       " 'N': 3,\n",
       " 'N+1': 4,\n",
       " 'N-1': 2,\n",
       " 'C': 4,\n",
       " 'C+1': 5,\n",
       " 'C-1': 3,\n",
       " 'P': 5,\n",
       " 'P+1': 6,\n",
       " 'P-1': 4,\n",
       " 'S': 6,\n",
       " 'S+1': 7,\n",
       " 'S-1': 5,\n",
       " '?': 3}"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sf.set_semantic_constraints() # reset constraints\n",
    "constraints = sf.get_semantic_constraints()\n",
    "constraints[\"?\"] = 3\n",
    "\n",
    "sf.set_semantic_constraints(constraints)\n",
    "constraints"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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>selfies</th>\n",
       "      <th>len</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>101707</th>\n",
       "      <td>[H]c1nc(=O)onc1F</td>\n",
       "      <td>[H][C][=N][C][=Branch1][C][=O][O][N][=C][Ring1...</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>105311</th>\n",
       "      <td>[H]OC([H])([H])C#N</td>\n",
       "      <td>[H][O][C][Branch1][C][H][Branch1][C][H][C][#N]</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>61184</th>\n",
       "      <td>[H]c1nc(F)nc(F)c1F</td>\n",
       "      <td>[H][C][=N][C][Branch1][C][F][=N][C][Branch1][C...</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>66050</th>\n",
       "      <td>[H]C#CC#Cc1nnoc1[H]</td>\n",
       "      <td>[H][C][#C][C][#C][C][N][=N][O][C][=Ring1][Bran...</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48158</th>\n",
       "      <td>[H]c1nc([H])n([H])n1</td>\n",
       "      <td>[H][C][N][=C][Branch1][C][H][N][Branch1][C][H]...</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      smiles  \\\n",
       "101707      [H]c1nc(=O)onc1F   \n",
       "105311    [H]OC([H])([H])C#N   \n",
       "61184     [H]c1nc(F)nc(F)c1F   \n",
       "66050    [H]C#CC#Cc1nnoc1[H]   \n",
       "48158   [H]c1nc([H])n([H])n1   \n",
       "\n",
       "                                                  selfies  len  \n",
       "101707  [H][C][=N][C][=Branch1][C][=O][O][N][=C][Ring1...   16  \n",
       "105311     [H][O][C][Branch1][C][H][Branch1][C][H][C][#N]   18  \n",
       "61184   [H][C][=N][C][Branch1][C][F][=N][C][Branch1][C...   18  \n",
       "66050   [H][C][#C][C][#C][C][N][=N][O][C][=Ring1][Bran...   19  \n",
       "48158   [H][C][N][=C][Branch1][C][H][N][Branch1][C][H]...   20  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def process_smiles(smiles):\n",
    "    return sf.encoder(smiles)\n",
    "\n",
    "def keys_int(symbol_to_int):\n",
    "    d = {}\n",
    "    for i, key in enumerate(symbol_to_int.keys()):\n",
    "        d[i] = key\n",
    "    return d\n",
    "\n",
    "data[\"selfies\"] = data[\"smiles\"].apply(process_smiles)\n",
    "\n",
    "data[\"len\"] = data[\"smiles\"].apply(lambda x: len(x))\n",
    "data.sort_values(by=\"len\").head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "selfies_list = np.asanyarray(data[\"selfies\"])\n",
    "selfies_alphabet = sf.get_alphabet_from_selfies(selfies_list)\n",
    "selfies_alphabet.add(\"[nop]\") # Ass the \"no operation\" symbol as a padding character\n",
    "selfies_alphabet.add(\".\")\n",
    "selfies_alphabet = list(sorted(selfies_alphabet))\n",
    "\n",
    "largest_selfie_len = max(sf.len_selfies(s) for s in selfies_list)\n",
    "symbol_to_int = dict((c, i) for i, c in enumerate(selfies_alphabet))\n",
    "int_mol = keys_int(symbol_to_int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "onehots = sf.batch_selfies_to_flat_hot(selfies_list, symbol_to_int, largest_selfie_len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_tensor = tf.convert_to_tensor(onehots, dtype=\"float64\")\n",
    "noise_tensor = tf.random.uniform(shape=input_tensor.shape, minval=0, maxval=1, dtype=\"float64\")\n",
    "dequantized_data = tf.add(input_tensor, noise_tensor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2000, 3050)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ds = NumpyDataset(dequantized_data)\n",
    "splitter = RandomSplitter()\n",
    "train, val, test = splitter.train_valid_test_split(dataset=ds, seed=42)\n",
    "train_idx, val_idx, test_idx = splitter.split(dataset=ds, seed=42)\n",
    "\n",
    "dim = len(train.X[0])\n",
    "train.X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_smiles = data['smiles'].iloc[train_idx].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "base_dist = tfd.MultivariateNormalDiag(loc=np.zeros(dim), scale_diag=np.ones(dim))\n",
    "\n",
    "if dim % 2 == 0:\n",
    "    permutation = tf.cast(np.concatenate((np.arange(dim / 2, dim), np.arange(0, dim / 2))), tf.int32)\n",
    "else:\n",
    "    permutation = tf.cast(np.concatenate((np.arange(dim / 2 + 1, dim), np.arange(0, dim / 2))), tf.int32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_layers = 8\n",
    "flow_layers = []\n",
    "\n",
    "Made = tfb.AutoregressiveNetwork(params=2, hidden_units=[521, 521], activation=\"relu\")\n",
    "\n",
    "for i in range(num_layers):\n",
    "    flow_layers.append(\n",
    "        (tfb.MaskedAutoregressiveFlow(shift_and_log_scale_fn=Made))\n",
    "    )\n",
    "    \n",
    "    permutation = tf.cast(np.random.permutation(np.arange(0, dim)), tf.int32)\n",
    "    flow_layers.append(tfb.Permute(permutation=permutation))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "nf = NormalizingFlow(base_distribution=base_dist, flow_layers=flow_layers)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "nfm = NormalizingFlowModel(nf, learning_rate = 1e-4, batch_size = 128, model_dir=\"model/generative/flow/generative-normalizing-flow\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "losses = []\n",
    "val_losses = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From c:\\Users\\user\\OneDrive\\Desktop\\GitHub\\small-molecules\\env\\lib\\site-packages\\tensorflow\\python\\autograph\\pyct\\static_analysis\\liveness.py:83: Analyzer.lamba_check (from tensorflow.python.autograph.pyct.static_analysis.liveness) is deprecated and will be removed after 2023-09-23.\n",
      "Instructions for updating:\n",
      "Lambda fuctions will be no more assumed to be used in the statement where they are used, or at least in the same block. https://github.com/tensorflow/tensorflow/issues/56089\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 3050) for input KerasTensor(type_spec=TensorSpec(shape=(None, 3050), dtype=tf.float64, name='input_1'), name='input_1', description=\"created by layer 'input_1'\"), but it was called on an input with incompatible shape (1, 128, 3050).\n",
      "CPU times: total: 29min 14s\n",
      "Wall time: 9min 54s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "max_epochs = 10\n",
    "\n",
    "for epoch in range(max_epochs):\n",
    "    loss = nfm.fit(train, nb_epoch=1, all_losses=losses)\n",
    "    val_loss = nfm.create_nll(val.X)\n",
    "    val_losses.append(val_loss.numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x240a2908610>"
      ]
     },
     "execution_count": 18,
     "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": [
    "fig, ax = plt.subplots()\n",
    "ax.scatter(range(len(losses)), losses, label = \"Training loss\")\n",
    "ax.scatter(range(len(val_losses)), val_losses, label = \"Validation loss\")\n",
    "plt.legend(loc=\"upper right\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "generated_samples = nfm.flow.sample(10) # generaive modeling\n",
    "log_probs = nfm.flow.log_prob(generated_samples) # probability density estimation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "mols = tf.math.floor(generated_samples) # quantize data\n",
    "mols = tf.clip_by_value(mols, 0, 1) # Set nefative values to 0 and values > 1 to 1\n",
    "\n",
    "int_to_symbol = dict((i, c) for i, c in enumerate(selfies_alphabet))\n",
    "\n",
    "# for idx, mol in enumerate(mols):\n",
    "#     mol = mol.numpy().tolist()\n",
    "#     print(sf.encoding_to_selfies([mol]))\n",
    "    \n",
    "mols = mols.numpy().tolist()\n",
    "selfies_molecule = sf.encoding_to_selfies(mols, vocab_itos=int_to_symbol, enc_type=\"one_hot\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "smile_molecule = sf.decoder(selfies_molecule)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SELFIES:  [-\\Ring2][C-1][C@@H1][/N+1].[=N][=N+1][H].[=Branch1]\n",
      "SMILES:  [C-1][C@@H1]/[N+1].N=[N+1][H]\n"
     ]
    }
   ],
   "source": [
    "print(\"SELFIES: \", selfies_molecule)\n",
    "print(\"SMILES: \", smile_molecule)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<rdkit.Chem.rdchem.Mol at 0x240cced83c0>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Chem.MolFromSmiles(sf.decoder(selfies_molecule), sanitize=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# let's generate 5 molecules\n",
    "\n",
    "# import random\n",
    "# int_to_symbol = dict((i, c) for i, c in enumerate(selfies_alphabet))\n",
    "\n",
    "# selfies_molecules = []\n",
    "\n",
    "# for i in range(5):\n",
    "#     n_steps = random.randint(10, 30)\n",
    "#     generated_samples = nfm.flow.sample(n_steps) # generaive modeling\n",
    "#     log_probs = nfm.flow.log_prob(generated_samples) # probability density estimation\n",
    "    \n",
    "#     mols = tf.math.floor(generated_samples) # quantize data\n",
    "#     mols = tf.clip_by_value(mols, 0, 1) # Set nefative values to 0 and values > 1 to 1\n",
    "#     mols = mols.numpy().tolist()\n",
    "#     selfies_molecule.append(sf.encoding_to_selfies(mols, vocab_itos=int_to_symbol, enc_type=\"one_hot\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['.',\n",
       " '[#Branch1]',\n",
       " '[#Branch2]',\n",
       " '[#C]',\n",
       " '[#N]',\n",
       " '[-\\\\Ring2]',\n",
       " '[/C]',\n",
       " '[/H]',\n",
       " '[/N+1]',\n",
       " '[/N]',\n",
       " '[/O]',\n",
       " '[=Branch1]',\n",
       " '[=Branch2]',\n",
       " '[=C]',\n",
       " '[=N+1]',\n",
       " '[=N]',\n",
       " '[=O]',\n",
       " '[=Ring1]',\n",
       " '[=Ring2]',\n",
       " '[Branch1]',\n",
       " '[Branch2]',\n",
       " '[C+1]',\n",
       " '[C-1]',\n",
       " '[C@@H1]',\n",
       " '[C@@]',\n",
       " '[C@H1]',\n",
       " '[C@]',\n",
       " '[CH1+1]',\n",
       " '[C]',\n",
       " '[F]',\n",
       " '[H]',\n",
       " '[N+1]',\n",
       " '[N-1]',\n",
       " '[N@@H1+1]',\n",
       " '[N@H1+1]',\n",
       " '[NH1+1]',\n",
       " '[NH1]',\n",
       " '[NH2+1]',\n",
       " '[N]',\n",
       " '[O-1]',\n",
       " '[O]',\n",
       " '[P]',\n",
       " '[Ring1]',\n",
       " '[Ring2]',\n",
       " '[S]',\n",
       " '[\\\\C]',\n",
       " '[\\\\H]',\n",
       " '[\\\\N]',\n",
       " '[\\\\O]',\n",
       " '[nop]']"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "selfies_alphabet\n",
    "# tf.saved_model.save(nfm, \"model/generative/flow/generative-flow-model-qm9\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# nfm.from_dir(\"model/generative/generative-flow-model-qm9\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "nfm_model = NormalizingFlowModel(nf, learning_rate = 1e-4, batch_size = 128, model_dir=\"model/generative/flow/generative-normalizing-flow\")\n",
    "nfm_model.restore()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "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": "3.10.9"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "4df9ae2f3669d2915414309207e4c3fb96f9f71dd598ede9d89ce16f72233d5f"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}