examples/Zeolite_attack.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Adversarial attacks on zeolites using the pretrained models\n",
    "\n",
    "In this notebook, we perform an adversarial attack on zeolites using the [SchNet NN potential](https://github.com/learningmatter-mit/NeuralForceField). We will be using the second generation of zeolite models, as shown in our paper.\n",
    "\n",
    "The utilities at `nff` will be used to perform the energy/force predictions. `nglview` will be used to visualize the generated trajectories. A few utility functions from this repo, `robust`, will be used as well. For the sake of generality, all steps for performing the adversarial attack are shown in this notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "716ac9e91dd04b798a282549462ad7f2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": []
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import os\n",
    "import sys\n",
    "sys.path.append('..')\n",
    "\n",
    "import random\n",
    "import robust as rb\n",
    "import torch as ch\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from nff.io import NeuralFF, AtomsBatch, EnsembleNFF\n",
    "from nff.data import Dataset\n",
    "from nff.train import load_model\n",
    "\n",
    "from ase.io import Trajectory, read\n",
    "\n",
    "import nglview as nv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading the dataset, models and initial geometry"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "DEVICE = 1\n",
    "\n",
    "dset = Dataset.from_file('../data/zeolite.pth.tar')\n",
    "\n",
    "PATH = '../models/zeolite'\n",
    "models = []\n",
    "for model_name in sorted(os.listdir(PATH)):\n",
    "    m = NeuralFF.from_file(os.path.join(PATH, model_name), device=DEVICE).model\n",
    "    models.append(m)\n",
    "\n",
    "ensemble = EnsembleNFF(models, device=DEVICE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "CUTOFF = 5\n",
    "\n",
    "def get_atoms(props):\n",
    "    atoms = AtomsBatch(\n",
    "        positions=props['nxyz'][:, 1:],\n",
    "        numbers=props['nxyz'][:, 0],\n",
    "        cell=props['lattice'],\n",
    "        pbc=True,\n",
    "        cutoff=CUTOFF,\n",
    "        props={'energy': 0, 'energy_grad': []},\n",
    "        calculator=ensemble,\n",
    "        nbr_torch=False,\n",
    "        device=DEVICE,\n",
    "    )\n",
    "    _ = atoms.update_nbr_list()\n",
    "    \n",
    "    return atoms\n",
    "\n",
    "initial = get_atoms(random.choice(dset))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Defining the adversarial attack\n",
    "\n",
    "The `Attacker` class allows one to perform an adversarial attack using an ensemble of SchNet models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "../robust/data.py:76: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  return ch.tensor(tensor).reshape(len(tensor), -1)\n",
      "../robust/data.py:77: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  return ch.tensor(tensor)\n"
     ]
    }
   ],
   "source": [
    "energies_per_atom = ch.tensor(dset.props['energy']) / dset.props['num_atoms']\n",
    "\n",
    "energy_dset = rb.PotentialDataset(\n",
    "    ch.zeros_like(energies_per_atom),\n",
    "    energies_per_atom,\n",
    "    energies_per_atom,\n",
    ")\n",
    "\n",
    "loss_fn = rb.loss.AdvLoss(\n",
    "    train=energy_dset,\n",
    "    temperature=20,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "attacker = rb.schnet.Attacker(\n",
    "    initial,\n",
    "    ensemble,\n",
    "    loss_fn,\n",
    "    device=DEVICE,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 100/100 [00:15<00:00,  6.48it/s]\n"
     ]
    }
   ],
   "source": [
    "results = attacker.attack(epochs=100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Post-processing the results: visualizing and plotting\n",
    "\n",
    "After the adversarial attack is performed, we can now visualize it and post-process the results. We start by recalculating the variance in forces and energies."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame(results)\n",
    "df['forces_var'] = [f.var(-1).mean() for f in df['forces']]\n",
    "df['energy_var'] = [e.var() for e in df['energy']]\n",
    "df['energy_avg'] = [e.mean() for e in df['energy']]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then, we reconstruct the trajectory of the adversarial attack using the values of `delta` along the attack."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "newatoms = []\n",
    "for transl in df.delta:\n",
    "    at = initial.copy()\n",
    "    at.translate(transl)\n",
    "    newatoms.append(at)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4bf6dbb6bb5e46faa1b46810c71e4f6a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "NGLWidget(max_frame=99)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "view = nv.show_asetraj(newatoms)\n",
    "view.add_unitcell()\n",
    "\n",
    "view"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we can plot the trajectory of the adversarial attack based on the sampled properties shown before (see Fig. S10 of the paper). Notice how the geometry that maximizes the adversarial loss is not necessarily the one with highest energy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/dskoda/.conda/envs/nff/lib/python3.7/site-packages/ipykernel_launcher.py:13: UserWarning: FixedFormatter should only be used together with FixedLocator\n",
      "  del sys.path[0]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(figsize=(4, 5))\n",
    "ax.spines['right'].set_visible(True)\n",
    "\n",
    "ax.plot(-df.loss, color='k')\n",
    "\n",
    "ax.set_ylabel('Adversarial Loss')\n",
    "\n",
    "COLOR_TAX = '#d62728'\n",
    "tax = ax.twinx()\n",
    "tax.plot(df.energy_avg, color=COLOR_TAX)\n",
    "tax.set_yticklabels(\n",
    "    tax.get_yticks(),\n",
    "    color=COLOR_TAX\n",
    ")\n",
    "tax.set_ylabel('Mean energy (kcal/mol)', color=COLOR_TAX)\n",
    "\n",
    "ax.set_xlabel('Attack step')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "nff",
   "language": "python",
   "name": "nff"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}