2_training/hamiltonian/train.ini

; DeepH-E3 training config — generated by train_ham.py

[basic]
device = cuda
dtype = float
save_dir = /home/apolyukhin/Development/epc_ml/example/diamond/2_training/hamiltonian/results
additional_folder_name = diamond_e3
simplified_output = False
seed = 42
checkpoint_dir =
use_new_hypp = True

[data]
graph_dir =
DFT_data_dir =
processed_data_dir = /home/apolyukhin/Development/epc_ml/example/diamond/2_training/hamiltonian/dataset
save_graph_dir = /home/apolyukhin/Development/epc_ml/example/diamond/2_training/hamiltonian/graph
target_data = hamiltonian
dataset_name = diamond_qe_e3
get_overlap = False

[train]
num_epoch = 500
batch_size = 1
extra_validation = []
extra_val_test_only = True

train_size = 30
val_size   = 10
test_size  = 10

min_lr = 1e-5

[hyperparameters]
learning_rate = 0.002
Adam_betas = (0.9, 0.999)
scheduler_type = 1
scheduler_params = (factor=0.5, cooldown=40, patience=120, threshold=0.05)
revert_decay_patience = 20
revert_decay_rate = 0.8

[target]
target = hamiltonian
target_blocks_type = all
target_blocks =
selected_element_pairs =
convert_net_out = False

[network]
cutoff_radius = 7.4
only_ij = False
spherical_harmonics_lmax = 4
spherical_basis_irreps =
irreps_embed = 64x0e
irreps_mid = 64x0e+32x1o+16x2e+8x3o+8x4e
num_blocks = 3
ignore_parity = False

irreps_embed_node =
irreps_edge_init =
irreps_mid_node =
irreps_post_node =
irreps_out_node =
irreps_mid_edge =
irreps_post_edge =
out_irreps =