Add cgcnn (#183)

* Add cgcnn framework * Add crystals to properties * Apply style on cgcnn framework * Add descriptions for crystal properties * Add training pipeline for cgcnn * Apply style * Update documentation * Add cgcnn training unit test and fix style * Enable cgcnn models * Increase samples_for_evaluation in the parameters * Rename CGCNN classes * Apply renaming in the cgccn unit test * Added the cgcnn properties to the factory for properties * Add examples for cgccn training * Proper handling of the classification output * Added cgcnn notebook * Apply style * Fix import in properties Co-authored-by: Jannis Born <[email protected]>
GT4SD · Jan 19, 2023 · 608f9a7 · 608f9a7
1 parent 2e6463d
commit 608f9a7
Show file tree

Hide file tree

Showing 27 changed files with 3,741 additions and 1 deletion.
diff --git a/examples/crystals/README.md b/examples/crystals/README.md
@@ -0,0 +1,43 @@
+## Crystals
+
+A simple example to train CGCNN based models to predict material properties.
+
+Make sure to activate the conda environment:
+
+```console
+conda activate gt4sd
+```
+
+To launch a training execute the following command from the GT4SD root. Examples of a dataset, including he sample-classification that we use in the given example, can be found in the CGCNN's official implementation repo in the following [link](https://github.com/txie-93/cgcnn/tree/master/data).
+
+```console
+
+gt4sd-trainer  --training_pipeline_name cgcnn \
+    --task classification \
+    --datapath sample-classification \
+    --atom_fea_len 64 \
+    --h_fea_len 128 \
+    --n_conv 3 \
+    --n_h 1 \
+    --epochs 30 \
+    --batch_size 256 \
+    --lr 0.01 \
+    --momentum 0.9 \
+    --weight_decay 0.0 \
+    --optim SGD 
+```
+
+The code is adapted from: <https://github.com/txie-93/cgcnn>.  
+
+```bibtex
+@article{xie2018crystal,
+  title={Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties},
+  author={Xie, Tian and Grossman, Jeffrey C},
+  journal={Physical review letters},
+  volume={120},
+  number={14},
+  pages={145301},
+  year={2018},
+  publisher={APS}
+}
+```
diff --git a/notebooks/cgcnn-demo.ipynb b/notebooks/cgcnn-demo.ipynb
@@ -0,0 +1,228 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Inference using CGCNN Model"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "In this notebook we show how to perform inference using GT4SD and CGCNN-based models. The current existing models (algorithm_version=v0) and the sample dataset have been obtained from the [official CGCNN repository](https://github.com/txie-93/cgcnn)."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Formation Energy\n",
+    "\n",
+    "This method predicts the formation energy per atom using the CGCNN framework (unit eV/atom)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import FormationEnergyParameters, FormationEnergy\n",
+    "\n",
+    "model_parameters = FormationEnergyParameters(algorithm_version=\"v0\")\n",
+    "model = FormationEnergy(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Absolute Energy\n",
+    "\n",
+    "This method predicts the absolute energy of crystals using the CGCNN framework (unit eV/atom)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import AbsoluteEnergy, AbsoluteEnergyParameters\n",
+    "\n",
+    "model_parameters = AbsoluteEnergyParameters(algorithm_version=\"v0\")\n",
+    "model = AbsoluteEnergy(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Band Gap\n",
+    "\n",
+    "This method predicts the band gap of crystals using the CGCNN framework (unit eV)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import BandGapParameters, BandGap\n",
+    "\n",
+    "model_parameters = BandGapParameters(algorithm_version=\"v0\")\n",
+    "model = BandGap(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Fermi Energy\n",
+    "\n",
+    "This method predicts the Fermi energy of crystals using the CGCNN framework (unit eV/atom)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import FermiEnergyParameters, FermiEnergy\n",
+    "\n",
+    "model_parameters = FermiEnergyParameters(algorithm_version=\"v0\")\n",
+    "model = FermiEnergy(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Bulk Moduli\n",
+    "\n",
+    "This method predicts the bulk moduli of crystals using the CGCNN framework (unit log(GPa))."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import BulkModuliParameters, BulkModuli\n",
+    "\n",
+    "model_parameters = BulkModuliParameters(algorithm_version=\"v0\")\n",
+    "model = BulkModuli(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Shear Moduli\n",
+    "\n",
+    "This method predicts the shear moduli of crystals using the CGCNN framework (unit log(GPa))."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import ShearModuliParameters, ShearModuli\n",
+    "\n",
+    "model_parameters = ShearModuliParameters(algorithm_version=\"v0\")\n",
+    "model = ShearModuli(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Poisson Ratio\n",
+    "\n",
+    "This method predicts the poisson ratio of crystals using the CGCNN framework."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import PoissonRatioParameters, PoissonRatio\n",
+    "\n",
+    "model_parameters = PoissonRatioParameters(algorithm_version=\"v0\")\n",
+    "model = PoissonRatio(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Metal-Semiconductor classifier\n",
+    "\n",
+    "This method predicts if the provided crystals are metal (1) or semiconductors (0) using the CGCNN framework."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from gt4sd.properties.crystals.core import MetalSemiconductorClassifierParameters, MetalSemiconductorClassifier\n",
+    "\n",
+    "model_parameters = MetalSemiconductorClassifierParameters(algorithm_version=\"v0\")\n",
+    "model = MetalSemiconductorClassifier(model_parameters)\n",
+    "\n",
+    "model(input=\"cgcnn-sample\")"
+   ]
+  }
+ ],
+ "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": "3.8.15"
+  },
+  "vscode": {
+   "interpreter": {
+    "hash": "7f6df040e92be56c42e1ba5ffdd58832f97e24eff2af0498cbc88845f2e95a48"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}