Merge pull request #5 from mhangaard/atoml_api

adsorbate fingerprinting tutorial

docs/9_adsorbate_fingerprints.py

@@ -0,0 +1,86 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Tue Jan 30 21:01:59 2018
+
+@author: mhangaard
+"""
+import numpy as np
+from ase.build import bulk
+from catkit.surface import SlabGenerator
+from ase.build import add_adsorbate
+from ase.data import covalent_radii
+from atoml.fingerprint.db2thermo import get_radius, attach_adsorbate_info
+from atoml.fingerprint.setup import return_fpv, get_combined_descriptors
+from atoml.fingerprint import AdsorbateFingerprintGenerator
+from atoml.utilities.clean_data import clean_infinite
+
+# Define an adsorbate.
+ads = 'C'
+
+# Create a bulk.
+atoms = bulk('Pd', 'fcc', a=4, cubic=True)
+atoms[3].symbol = 'Cu'
+
+# Create some surfaces.
+Gen = SlabGenerator(
+    atoms,
+    miller_index=[1, 1, 1],
+    layers=3,
+    fixed=1,
+    vacuum=8,
+)
+terminations = Gen.get_unique_terminations()
+# Loop over unique surface terminations.
+images = []
+for i, t in enumerate(terminations):
+    slab = Gen.get_slab(iterm=i)
+    slab.center(axis=2, vacuum=5)
+    # Add an adsorbate to each slab.
+    radii = np.average([get_radius(a.symbol) for a in slab])
+    height = radii + covalent_radii[6]
+    add_adsorbate(slab, ads, height, position=slab.get_positions()[-1, :2])
+    # Some help is needed for AtoML to positively identify the adsorbate.
+    # A string in Hill notation is enough, if the adsorbate consists
+    #   of different elements from the slab.
+    slab.info['species'] = ads
+    # The number of layers also makes some of the fingerprinting easier.
+    slab.info['layers'] = Gen.layers
+    # Append them to a list.
+    images.append(slab)
+
+# Some information is expected to be attached to atoms objects.
+#   There are various ways of doing this, but the easiest is to call
+AtoML_atoms = attach_adsorbate_info(images)
+# This is where checks should be made
+
+# Instantiate the fingerprint generator for adsorbate structures.
+fingerprinter = AdsorbateFingerprintGenerator()
+
+# All user methods under the fingerprinter accepts an atoms object and
+#   returns a vector.
+functions = [
+    fingerprinter.ads_nbonds,
+    fingerprinter.primary_addatom,
+    fingerprinter.primary_adds_nn,
+    fingerprinter.Z_add,
+    fingerprinter.ads_av,
+    fingerprinter.primary_surf_nn,
+    fingerprinter.primary_surfatom,
+    ]
+# This list is passed on to the following setup functions,
+#    along with a list of atoms.
+
+# Get and print the names of features.
+features_labels = get_combined_descriptors(functions)
+for l in range(len(features_labels)):
+    print(l, features_labels[l])
+
+# Get a matrix containing the fingerprints.
+unlabeled_data_matrix = return_fpv(AtoML_atoms, functions)
+print(np.shape(unlabeled_data_matrix), 'data matrix created.')
+
+# Cleanup in case some of the functions are returning NaNs or Infs
+clean_data_matrix = clean_infinite(unlabeled_data_matrix)['train']
+
+# Ready for Machine learning.
+print(np.shape(clean_data_matrix), 'data matrix returned.')