def __init__(self, model_name, Instance=None, resolution=50):
self.geoms = []
self.figure = plt.figure()
self.ax = Axes3D(self.figure)
#self.ax.set_zlim(bottom=-1,top=5)
plt.title(model_name)
self.resolution = resolution
self.theta = np.linspace(60 * np.pi / 180, 300 * np.pi / 180,
self.resolution)
self.rhs = np.linspace(2, 8, self.resolution)
self.meshX, self.meshY = np.meshgrid(self.rhs,
self.theta * 180 / np.pi)
meshZ = np.zeros((len(self.theta), len(self.rhs)))
self.surf = self.ax.plot_surface(self.meshX,
self.meshY,
meshZ,
cmap=cm.magma,
animated=True)
self.model = model_name
self.create_geoms()
self.Evaluation = NN.AtomicNeuralNetInstance()
self.Evaluation.CalcDatasetStatistics = False
self.Evaluation.TextOutput = False
if Instance != None:
self.Evaluation._IsFromCheck = True
self.Evaluation.Rs = Instance.Rs
self.Evaluation.R_Etas = Instance.R_Etas
self.Evaluation.NumberOfRadialFunctions = Instance.NumberOfRadialFunctions
self.Evaluation.Etas = Instance.Etas
self.Evaluation.Lambs = Instance.Lambs
self.Evaluation.Zetas = Instance.Zetas
self.Data = None
self.ZData = []
def get_multi_data(model, data_files):
predictions = []
targets = []
i = 0
while i < len(data_files):
try:
data_file = data_files[i]
print(data_file)
#Load trainings instance
Training = _NN.AtomicNeuralNetInstance()
Training.TextOutput = False
Training.CalcDatasetStatistics = False
Reader = _readers.QE_MD_Reader()
Reader.E_conv_factor = 13.605698066
Reader.read_folder(data_file)
offset = (
-774.90203736 * Reader.nr_atoms_per_type[1] +
Reader.nr_atoms_per_type[0] * -185.29265964) * 13.605698066
Training.prepare_evaluation(model, Reader.nr_atoms_per_type)
Training.create_eval_data(Reader.geometries)
temp = [
pred[0] / sum(Reader.nr_atoms_per_type)
for pred in Training.eval_dataset_energy(Training.EvalData)
]
predictions += temp
targets += list(
np.divide(np.subtract(Reader.e_pot, offset),
sum(Reader.nr_atoms_per_type)))
i += 1
gc.collect()
except:
print("Failed for file " + str(data_file) + " retrying...")
return predictions, targets
from NeuralNetworks import NeuralNetworkUtilities
from NeuralNetworks.data_generation import data_readers
from os import listdir
from os.path import isfile, join
import numpy as np
input_reader = data_readers.SimpleInputReader()
mypath = "/home/afuchs/Documents/Validation_big/"
files = [f for f in listdir(mypath) if isfile(join(mypath, f))]
for file in files:
input_reader.read(join(mypath, file), skip=3)
energies = np.asarray(input_reader.energies)
Training = NeuralNetworkUtilities.AtomicNeuralNetInstance()
Training.TextOutput = False
Training.CalcDatasetStatistics = False
Training.prepare_evaluation(
"/home/afuchs/Documents/NiAu_Training/multi_morse_smallds",
nr_atoms_per_type=[1, 146])
#Training.create_eval_data(input_reader.geometries)
#out=Training.eval_dataset_energy(Training.EvalData)
out = []
for geometry in input_reader.geometries:
out.append(Training.energy_for_geometry(geometry))
offset = (-774.90203736 * 146 + 1 * -185.29265964) * 13.605698066
ref = energies - offset
res = out
min_ref = energies - min(energies)
min_res = res - min(res)
rel_error = []
for this in data_files:
folder_files = [os.path.join(this, f) for f in os.listdir(this)]
for myfile in folder_files:
temp = open(myfile, "r").read()
if "error" in temp:
print("Error in :" + str(myfile))
print(data_files)
Multi = _NN.MultipleInstanceTraining()
Multi.GlobalLearningRate = learning_rate
for i in range(len(data_files)):
data_file = data_files[i]
print(data_file)
#Load trainings instance
Training = _NN.AtomicNeuralNetInstance()
Training.IsPartitioned = False
Training.IncludeMorse = True
Training.CalcDatasetStatistics = True
Training.UseForce = force
#Default symmetry function set
#Training.NumberOfRadialFunctions=15
bohr2ang = 0.529177249
#Training.Lambs=[1.0,-1.0]
#Training.Zetas=[0.2,0.5,1,3,10]#[0.025,0.045,0.075,0.1,0.15,0.2,0.3,0.5,0.7,1,1.5,2,3,5,10,18,36,100]
#Training.Etas=[0.01]
Training.Rs = [
0, 0, 0, 0, 0, 0, 0, 0
] #,1.16674542, 1.81456625,2.3, 2.89256287, 4.53134823, 6.56226301, 6.92845869]
Training.R_Etas = [
0.4 / bohr2ang**2, 0.2 / bohr2ang**2, 0.1 / bohr2ang**2,