splitter = modsel.KFold(n_splits=3, random_state=42, shuffle=True)
indices = splitter.split(idx)
all_scores = []
for item in indices:
estimator = ARMP_G(iterations=1,
representation='acsf',
representation_params=acsf_params,
batch_size=250,
tensorboard=True)
estimator.set_xyz(xyz)
estimator.set_classes(zs)
estimator.set_properties(ene)
estimator.set_gradients(forces)
estimator.generate_representation(method='fortran')
idx_train = item[0]
idx_test = item[1]
estimator.fit(idx_train)
score = estimator.score(idx_test)
all_scores.append(score)
tf.reset_default_graph()
print(all_scores)
data = h5py.File(
"/Volumes/Transcend/data_sets/CN_isobutane_model/pruned_dft_with_forces/pruned_isopentane_cn_dft.hdf5",
"r")
n_samples = 100
xyz = np.array(data.get("xyz")[-n_samples:])
ene = np.array(data.get("ene")[-n_samples:]) * 2625.50
ene = ene - data.get("ene")[0] * 2625.50
zs = np.array(data["zs"][-n_samples:], dtype=np.int32)
forces = np.array(data.get("forces")[-n_samples:])
idx = list(range(n_samples))
# Creating the estimator
estimator = ARMP_G(iterations=50, batch_size=10)
estimator.set_xyz(xyz)
estimator.set_properties(ene)
estimator.set_classes(zs)
estimator.set_gradients(forces)
estimator.load_nn()
estimator.predict(idx)
estimator.fit(idx)
estimator.score(idx)
estimator.set_xyz(xyz)
estimator.set_classes(zs)
estimator.set_properties(ene)
estimator.set_gradients(forces)
estimator.generate_representation()
idx = np.arange(0, n_samples)
idx_train, idx_test = modsel.train_test_split(idx, test_size=0, random_state=42, shuffle=True)
# estimator.load_nn()
estimator.fit(idx_train)
# print("Done the fitting")
ene_pred, f_pred = estimator.predict(idx_train)
plt.scatter(ene_pred, ene[idx_train])
plt.xlabel("Predicted energies (kJ/mol)")
plt.ylabel("DFT energies (kJ/mol)")
# plt.savefig("mem_aglaia_overfit.png", dpi=200)
plt.show()
score = estimator.score(idx_train)
# print("\n The score is %s" % (str(score)))
# os.remove("predict.tfrecords")
# os.remove("training.tfrecords")
# estimator.save_nn()
n_samples = 10
xyz = np.array(data.get("xyz")[:n_samples])
ene = np.array(data.get("ene")[:n_samples])*2625.50
ene = ene - data.get("ene")[0]*2625.50
zs = np.array(data["zs"][:n_samples], dtype=np.int32)
forces = np.array(data.get("forces")[:n_samples])
# Creating the estimator
estimator = ARMP_G(iterations=10, batch_size=30, hidden_layer_sizes=(40, 20, 10), l2_reg=0.0, tensorboard=False,
store_frequency=5, learning_rate=0.001, representation_params={'nRs2':7, 'nRs3':7, 'nTs':7,
'zeta':220.127, 'eta':30.8065})
estimator.set_xyz(xyz)
estimator.set_properties(ene)
estimator.set_classes(zs)
estimator.set_gradients(forces)
idx = list(range(n_samples))
estimator.load_nn("saved_model")
estimator.fit(idx)
pred_ene, pred_forces = estimator.predict(idx)
score = estimator.score(idx)
# estimator.save_nn()
plt.scatter(ene, pred_ene)
plt.show()