3-composition_example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Prediction of experimental band gap\n",
    "\n",
    "This notebooks applies MODNet on the matbench experimental band gap data. It is a good example on how MODNet can be used for a composition only task, i.e. when only the stoichiometry (e.g. SiO2) is known with no structural information."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from matminer.datasets import load_dataset\n",
    "from modnet.models import MODNetModel\n",
    "from modnet.preprocessing import MODData\n",
    "import matplotlib.pyplot as plt \n",
    "from pymatgen.core import Composition"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1. Import dataset:\n",
    "The matbench_expt_gap dataset contains measured band gaps for 4604 compositions of inorganic semiconductors from Zhuo et al., JPCL."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "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>composition</th>\n",
       "      <th>gap expt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>(Ag, Au, S)</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>(Ag, W, Br)</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>(Ag, Ge, Pb, S)</td>\n",
       "      <td>1.83</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>(Ag, Ge, Pb, Se)</td>\n",
       "      <td>1.51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>(Ag, B, Br)</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        composition  gap expt\n",
       "0       (Ag, Au, S)      0.00\n",
       "1       (Ag, W, Br)      0.00\n",
       "2   (Ag, Ge, Pb, S)      1.83\n",
       "3  (Ag, Ge, Pb, Se)      1.51\n",
       "4       (Ag, B, Br)      0.00"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from matminer.datasets import load_dataset\n",
    "\n",
    "df = load_dataset(\"matbench_expt_gap\")\n",
    "df[\"composition\"] = df[\"composition\"].map(Composition) # maps composition to a pymatgen composition object\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us have a look at the band gap distribution:"
   ]
  },
  {
   "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>gap expt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>4604.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.975951</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>1.445034</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.812500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>11.700000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          gap expt\n",
       "count  4604.000000\n",
       "mean      0.975951\n",
       "std       1.445034\n",
       "min       0.000000\n",
       "25%       0.000000\n",
       "50%       0.000000\n",
       "75%       1.812500\n",
       "max      11.700000"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f9c7b4f52b0>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots(facecolor=\"w\")\n",
    "ax.hist(df.where(df[\"gap expt\"] == 0)[\"gap expt\"], bins=1, density=False, label=\"Zero band gap\")\n",
    "ax.hist(df.where(df[\"gap expt\"] > 0)[\"gap expt\"], bins=11, density=False, label=\"Non-zero band gap\")\n",
    "ax.set_ylabel(\"Frequency\")\n",
    "ax.set_xlabel(\"Band gap (eV)\")\n",
    "ax.legend()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Training \n",
    "## a) create MODData\n",
    "We create a MODData based on the above dataframe."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:44:22,918 - modnet - INFO - Loaded CompositionOnlyMatminer2023Featurizer featurizer.\n"
     ]
    }
   ],
   "source": [
    "# This instantiates the MODData\n",
    "data = MODData(\n",
    "    materials=df[\"composition\"], # you can provide composition objects to MODData\n",
    "    targets=df[\"gap expt\"], \n",
    "    target_names=[\"gap_expt_eV\"]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:44:23,032 - modnet - INFO - Loaded <modnet.preprocessing.MODData object at 0x7f9c7b613430> object, created with modnet version 0.2.0~develop\n"
     ]
    }
   ],
   "source": [
    "# Featurization of the moddata\n",
    "# It will automatically apply composition only featurizers\n",
    "# data.featurize() #this takes around 10 mins to run\n",
    "# data.save(\"out/gap_comp.mdt\")\n",
    "data = MODData.load(\"out/gap_comp.mdt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## b) Train-test split\n",
    "In order to test the model on unseen data, we split the dataset in two parts: a training and testing set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "split = train_test_split(range(100), test_size=0.1, random_state=1234)\n",
    "train, test = data.split(split)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Important features are selected. This takes around 1 minute."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:44:23,207 - modnet - INFO - Multiprocessing on 1 workers.\n",
      "2023-01-24 14:44:23,212 - modnet - INFO - Computing \"self\" MI (i.e. information entropy) of features\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 270/270 [00:08<00:00, 30.56it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:44:32,082 - modnet - INFO - Computing cross NMI between all features...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "100%|██████████| 13695/13695 [00:43<00:00, 317.79it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:45:15,852 - modnet - INFO - Starting target 1/1: gap_expt_eV ...\n",
      "2023-01-24 14:45:15,852 - modnet - INFO - Computing mutual information between features and target...\n",
      "2023-01-24 14:45:17,016 - modnet - INFO - Computing optimal features...\n",
      "2023-01-24 14:45:20,506 - modnet - INFO - Selected 50/166 features...\n",
      "2023-01-24 14:45:22,876 - modnet - INFO - Selected 100/166 features...\n",
      "2023-01-24 14:45:24,018 - modnet - INFO - Selected 150/166 features...\n",
      "2023-01-24 14:45:24,114 - modnet - INFO - Done with target 1/1: gap_expt_eV.\n",
      "2023-01-24 14:45:24,114 - modnet - INFO - Merging all features...\n",
      "2023-01-24 14:45:24,115 - modnet - INFO - Done.\n"
     ]
    }
   ],
   "source": [
    "train.feature_selection(n=-1)\n",
    "# if you want to use precomputed cross_nmi of the MP. This saves time :\n",
    "# data.feature_selection(n=-1, use_precomputed_cross_nmi)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## c) Create MODNet model\n",
    "\n",
    "A `MODNetModel` is created with custom hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2023-01-24 14:45:24.217663: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
     ]
    }
   ],
   "source": [
    "model = MODNetModel([[['gap_expt_eV']]],\n",
    "                    weights={'gap_expt_eV':1},\n",
    "                    num_neurons = [[256], [128], [16], [16]],\n",
    "                    n_feat = 150,\n",
    "                    act =  \"elu\"\n",
    "                   )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## d) Model fitting\n",
    "The model is fitted with custom hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/ppdebreuck/anaconda3/envs/modnet-develop/lib/python3.8/site-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n",
      "  super(Adam, self).__init__(name, **kwargs)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 0: loss: 1.111, val_loss:1.162 val_mae:1.162\n",
      "epoch 1: loss: 0.995, val_loss:0.979 val_mae:0.979\n",
      "epoch 2: loss: 0.905, val_loss:0.843 val_mae:0.843\n",
      "epoch 3: loss: 0.821, val_loss:0.747 val_mae:0.747\n",
      "epoch 4: loss: 0.754, val_loss:0.652 val_mae:0.652\n",
      "epoch 5: loss: 0.694, val_loss:0.554 val_mae:0.554\n",
      "epoch 6: loss: 0.650, val_loss:0.455 val_mae:0.455\n",
      "epoch 7: loss: 0.619, val_loss:0.359 val_mae:0.359\n",
      "epoch 8: loss: 0.593, val_loss:0.312 val_mae:0.312\n",
      "epoch 9: loss: 0.582, val_loss:0.323 val_mae:0.323\n",
      "epoch 10: loss: 0.569, val_loss:0.325 val_mae:0.325\n",
      "epoch 11: loss: 0.558, val_loss:0.330 val_mae:0.330\n",
      "epoch 12: loss: 0.556, val_loss:0.336 val_mae:0.336\n",
      "epoch 13: loss: 0.556, val_loss:0.334 val_mae:0.334\n",
      "epoch 14: loss: 0.550, val_loss:0.328 val_mae:0.328\n",
      "epoch 15: loss: 0.543, val_loss:0.327 val_mae:0.327\n",
      "epoch 16: loss: 0.537, val_loss:0.324 val_mae:0.324\n",
      "epoch 17: loss: 0.530, val_loss:0.322 val_mae:0.322\n",
      "epoch 18: loss: 0.520, val_loss:0.321 val_mae:0.321\n",
      "epoch 19: loss: 0.512, val_loss:0.322 val_mae:0.322\n",
      "epoch 20: loss: 0.508, val_loss:0.325 val_mae:0.325\n",
      "epoch 21: loss: 0.504, val_loss:0.330 val_mae:0.330\n",
      "epoch 22: loss: 0.501, val_loss:0.331 val_mae:0.331\n",
      "epoch 23: loss: 0.498, val_loss:0.330 val_mae:0.330\n",
      "epoch 24: loss: 0.491, val_loss:0.352 val_mae:0.352\n",
      "epoch 25: loss: 0.485, val_loss:0.371 val_mae:0.371\n",
      "epoch 26: loss: 0.483, val_loss:0.361 val_mae:0.361\n",
      "epoch 27: loss: 0.476, val_loss:0.358 val_mae:0.358\n",
      "epoch 28: loss: 0.471, val_loss:0.367 val_mae:0.367\n",
      "epoch 29: loss: 0.469, val_loss:0.363 val_mae:0.363\n",
      "epoch 30: loss: 0.461, val_loss:0.339 val_mae:0.339\n",
      "epoch 31: loss: 0.455, val_loss:0.346 val_mae:0.346\n",
      "epoch 32: loss: 0.449, val_loss:0.358 val_mae:0.358\n",
      "epoch 33: loss: 0.446, val_loss:0.360 val_mae:0.360\n",
      "epoch 34: loss: 0.441, val_loss:0.360 val_mae:0.360\n",
      "epoch 35: loss: 0.436, val_loss:0.356 val_mae:0.356\n",
      "epoch 36: loss: 0.434, val_loss:0.362 val_mae:0.362\n",
      "epoch 37: loss: 0.428, val_loss:0.388 val_mae:0.388\n",
      "epoch 38: loss: 0.426, val_loss:0.391 val_mae:0.391\n",
      "epoch 39: loss: 0.421, val_loss:0.358 val_mae:0.358\n",
      "epoch 40: loss: 0.421, val_loss:0.358 val_mae:0.358\n",
      "epoch 41: loss: 0.421, val_loss:0.376 val_mae:0.376\n",
      "epoch 42: loss: 0.408, val_loss:0.407 val_mae:0.407\n",
      "epoch 43: loss: 0.412, val_loss:0.400 val_mae:0.400\n",
      "epoch 44: loss: 0.407, val_loss:0.377 val_mae:0.377\n",
      "epoch 45: loss: 0.402, val_loss:0.362 val_mae:0.362\n",
      "epoch 46: loss: 0.398, val_loss:0.360 val_mae:0.360\n",
      "epoch 47: loss: 0.393, val_loss:0.375 val_mae:0.375\n",
      "epoch 48: loss: 0.393, val_loss:0.377 val_mae:0.377\n",
      "epoch 49: loss: 0.386, val_loss:0.403 val_mae:0.403\n",
      "epoch 50: loss: 0.385, val_loss:0.422 val_mae:0.422\n",
      "epoch 51: loss: 0.386, val_loss:0.409 val_mae:0.409\n",
      "epoch 52: loss: 0.380, val_loss:0.371 val_mae:0.371\n",
      "epoch 53: loss: 0.378, val_loss:0.358 val_mae:0.358\n",
      "epoch 54: loss: 0.373, val_loss:0.388 val_mae:0.388\n",
      "epoch 55: loss: 0.379, val_loss:0.392 val_mae:0.392\n",
      "epoch 56: loss: 0.373, val_loss:0.372 val_mae:0.372\n",
      "epoch 57: loss: 0.360, val_loss:0.376 val_mae:0.376\n",
      "epoch 58: loss: 0.358, val_loss:0.398 val_mae:0.398\n",
      "epoch 59: loss: 0.356, val_loss:0.399 val_mae:0.399\n",
      "epoch 60: loss: 0.356, val_loss:0.376 val_mae:0.376\n",
      "epoch 61: loss: 0.347, val_loss:0.367 val_mae:0.367\n",
      "epoch 62: loss: 0.341, val_loss:0.355 val_mae:0.355\n",
      "epoch 63: loss: 0.341, val_loss:0.359 val_mae:0.359\n",
      "epoch 64: loss: 0.339, val_loss:0.382 val_mae:0.382\n",
      "epoch 65: loss: 0.335, val_loss:0.361 val_mae:0.361\n",
      "epoch 66: loss: 0.325, val_loss:0.366 val_mae:0.366\n",
      "epoch 67: loss: 0.325, val_loss:0.380 val_mae:0.380\n",
      "epoch 68: loss: 0.327, val_loss:0.380 val_mae:0.380\n",
      "epoch 69: loss: 0.321, val_loss:0.392 val_mae:0.392\n",
      "epoch 70: loss: 0.314, val_loss:0.366 val_mae:0.366\n",
      "epoch 71: loss: 0.307, val_loss:0.346 val_mae:0.346\n",
      "epoch 72: loss: 0.309, val_loss:0.374 val_mae:0.374\n",
      "epoch 73: loss: 0.312, val_loss:0.368 val_mae:0.368\n",
      "epoch 74: loss: 0.299, val_loss:0.332 val_mae:0.332\n",
      "epoch 75: loss: 0.318, val_loss:0.345 val_mae:0.345\n",
      "epoch 76: loss: 0.301, val_loss:0.373 val_mae:0.373\n",
      "epoch 77: loss: 0.293, val_loss:0.319 val_mae:0.319\n",
      "epoch 78: loss: 0.304, val_loss:0.311 val_mae:0.311\n",
      "epoch 79: loss: 0.294, val_loss:0.384 val_mae:0.384\n",
      "epoch 80: loss: 0.303, val_loss:0.370 val_mae:0.370\n",
      "epoch 81: loss: 0.299, val_loss:0.307 val_mae:0.307\n",
      "epoch 82: loss: 0.316, val_loss:0.312 val_mae:0.312\n",
      "epoch 83: loss: 0.281, val_loss:0.356 val_mae:0.356\n",
      "epoch 84: loss: 0.270, val_loss:0.325 val_mae:0.325\n",
      "epoch 85: loss: 0.273, val_loss:0.304 val_mae:0.304\n",
      "epoch 86: loss: 0.271, val_loss:0.360 val_mae:0.360\n",
      "epoch 87: loss: 0.263, val_loss:0.399 val_mae:0.399\n",
      "epoch 88: loss: 0.273, val_loss:0.342 val_mae:0.342\n",
      "epoch 89: loss: 0.244, val_loss:0.308 val_mae:0.308\n",
      "epoch 90: loss: 0.252, val_loss:0.328 val_mae:0.328\n",
      "epoch 91: loss: 0.246, val_loss:0.339 val_mae:0.339\n",
      "epoch 92: loss: 0.240, val_loss:0.310 val_mae:0.310\n",
      "epoch 93: loss: 0.239, val_loss:0.316 val_mae:0.316\n",
      "epoch 94: loss: 0.240, val_loss:0.351 val_mae:0.351\n",
      "epoch 95: loss: 0.242, val_loss:0.347 val_mae:0.347\n",
      "epoch 96: loss: 0.234, val_loss:0.331 val_mae:0.331\n",
      "epoch 97: loss: 0.228, val_loss:0.337 val_mae:0.337\n",
      "epoch 98: loss: 0.228, val_loss:0.330 val_mae:0.330\n",
      "epoch 99: loss: 0.219, val_loss:0.347 val_mae:0.347\n"
     ]
    }
   ],
   "source": [
    "model.fit(train,\n",
    "          val_fraction = 0.1,\n",
    "          lr = 0.0002,\n",
    "          batch_size = 64,\n",
    "          loss = 'mae',\n",
    "          epochs = 100,\n",
    "          verbose = 1,\n",
    "         )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 3. Predicting on test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "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>gap_expt_eV</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>id40</th>\n",
       "      <td>0.344833</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id35</th>\n",
       "      <td>3.012862</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id81</th>\n",
       "      <td>-0.301482</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id61</th>\n",
       "      <td>2.006864</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id98</th>\n",
       "      <td>1.696285</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      gap_expt_eV\n",
       "id40     0.344833\n",
       "id35     3.012862\n",
       "id81    -0.301482\n",
       "id61     2.006864\n",
       "id98     1.696285"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred = model.predict(test)\n",
    "pred.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TEST MAE: 0.32 eV\n"
     ]
    }
   ],
   "source": [
    "mae_test = np.absolute(pred.values-test.df_targets.values).mean()\n",
    "print(\"TEST MAE: {:.2f} eV\".format(mae_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This concludes the notebook."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8.12 ('modnet-develop')",
   "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": "3.8.12"
  },
  "vscode": {
   "interpreter": {
    "hash": "65cdb1bf34a883aa160cb191d1776d32605e2c21ff3abc3b3101a9562d1e4c9d"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}