config = join_dicts(
PRESETS_TRAIN['numu_v3'],
{
# Config:
'batch_size' : 1024,
#'vars_input_slice',
#'vars_input_png2d',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset' :
'numu/prod4/fd_fhc/dataset_lstm_ee_fd_fhc_nonswap_std_cut.csv.xz',
'early_stop' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'min_delta' : 0,
'patience' : 40,
},
},
'epochs' : 200,
'loss' : 'mean_absolute_percentage_error',
'max_prongs' : None,
'model' : {
'name' : 'lstm_v3_stack',
'kwargs' : {
'batchnorm' : True,
'layers_pre' : [128, 128, 128],
'layers_post' : [128, 128, 128],
'lstm3d_spec' : [ (32, 'forward') ],
'lstm2d_spec' : [ (32, 'forward') ],
'n_resblocks' : 0,
},
},
'noise' : None,
'optimizer' : {
'name' : 'RMSprop',
'kwargs' : { 'lr' : 0.001 },
},
'prong_sorters' : None,
'regularizer' : {
'name' : 'l1',
'kwargs' : { 'l' : 0.001 },
},
'schedule' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'factor' : 0.5,
'patience' : 5,
'cooldown' : 0
},
},
'seed' : 1337,
'steps_per_epoch' : 250,
'test_size' : 200000,
'weights' : {
'name' : 'flat',
'kwargs' : { 'bins' : 50, 'range' : (0, 5) },
},
# Args:
'vars_mod_png2d' : None,
'vars_mod_png3d' : None,
'vars_mod_slice' : None,
'outdir' : \
'numu/prod4/final_model_perturbations/stack_of_3dlstms/',
}
)
config = join_dicts(
PRESETS_TRAIN['nue_v3'],
{
# Config:
'batch_size' : 1024,
#'vars_input_slice',
#'vars_input_png2d',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset' :
'nue/prod4/fd_fhc/dataset_lstm_ee_fd_fhc_fluxswap_std_cut.csv.xz',
'early_stop' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'min_delta' : 0,
'patience' : 40,
},
},
'epochs' : 200,
'loss' : 'mean_absolute_percentage_error',
'max_prongs' : None,
'model' : {
'name' : 'lstm_v3',
'kwargs' : {
'batchnorm' : True,
'layers_pre' : [ 128, 128, 128 ],
'layers_post' : [ 128, 128, 128 ],
'lstm_units2d' : 32,
'lstm_units3d' : 32,
'n_resblocks' : 0,
},
},
'noise' : {
'affected_vars_slice' : [ 'calE', 'orphCalE', 'remPngCalE' ],
'affected_vars_png2d' : [ 'png2d.calE', 'png2d.weightedCalE' ],
'affected_vars_png3d' : [
'png.calE',
'png.weightedCalE',
'png.bpf[0].overlapE',
'png.bpf[1].overlapE',
'png.bpf[2].overlapE',
],
},
'optimizer' : {
'name' : 'RMSprop',
'kwargs' : { 'lr' : 0.001 },
},
'prong_sorters' : None,
'regularizer' : {
'name' : 'l1',
'kwargs' : { 'l' : 0.001 },
},
'schedule' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'factor' : 0.5,
'patience' : 5,
'cooldown' : 0
},
},
'seed' : 0,
'steps_per_epoch' : 250,
'test_size' : 200000,
'weights' : {
'name' : 'flat',
'kwargs' : { 'bins' : 25, 'range' : (0, 5), 'clip' : 50 },
},
# Args:
'vars_mod_png2d' : None,
'vars_mod_png3d' : None,
'vars_mod_slice' : None,
'outdir' : 'nue/prod4/01_initial_studies/03_lstm_v3_final_wclip_noise/',
}
)
config = join_dicts(
PRESETS_TRAIN['dune_numu_v1'],
{
# Config:
'batch_size': 1024,
#'vars_input_slice',
#'vars_input_png2d',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset': 'dune/numu/dataset_rnne_dune_numu.csv.xz',
'early_stop': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'min_delta': 0,
'patience': 40,
},
},
'epochs': 200,
'loss': 'mean_absolute_percentage_error',
'max_prongs': None,
'model': {
'name': 'lstm_v2',
'kwargs': {
'batchnorm': True,
'layers_pre': [128, 128, 128],
'layers_post': [128, 128, 128],
'lstm_units': 32,
'n_resblocks': 0,
},
},
'noise': None,
'optimizer': {
'name': 'RMSprop',
'kwargs': {
'lr': 0.001
},
},
'prong_sorters': None,
'regularizer': {
'name': 'l1',
'kwargs': {
'l': 0.001
},
},
'schedule': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'factor': 0.5,
'patience': 5,
'cooldown': 0
},
},
'seed': 0,
'steps_per_epoch': 250,
'test_size': 0.2,
'weights': None,
# Args:
'vars_mod_png2d': None,
'vars_mod_png3d': None,
'vars_mod_slice': None,
'outdir': 'dune/numu/01_rnne_v1/',
})
config = join_dicts(
PRESETS_TRAIN['numu_v2'],
{
# Config:
'batch_size':
1024,
#'vars_input_slice',
#'vars_input_png2d',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset':
'numu/prod4/fd_fhc/dataset_lstm_ee_fd_fhc_nonswap_std_cut.csv.xz',
'early_stop': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'min_delta': 0,
'patience': 40,
},
},
'epochs':
200,
'loss':
'mean_absolute_percentage_error',
'max_prongs':
None,
'model': {
'name': 'lstm_v2',
'kwargs': {
'batchnorm': True,
'layers_pre': [128, 128, 128],
'layers_post': [128, 128, 128],
'lstm_units': 32,
'n_resblocks': 0,
},
},
'noise':
None,
'optimizer': {
'name': 'RMSprop',
'kwargs': {
'lr': 0.001
},
},
'prong_sorters':
None,
'regularizer': {
'name': 'l1',
'kwargs': {
'l': 0.001
},
},
'schedule': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'factor': 0.5,
'patience': 5,
'cooldown': 0
},
},
'seed':
1337,
'steps_per_epoch':
250,
'test_size':
200000,
'weights':
None,
# Args:
'vars_mod_png2d':
None,
'vars_mod_png3d': [
'-png.cvnpart.neutronid',
'-png.cvnpart.pizeroid',
'-png.bpf[2].pid',
],
'vars_mod_slice':
None,
'outdir':
'numu/prod4/initial_studies/lstm_v2_final/',
})
config = join_dicts(
PRESETS_TRAIN['numu_v3'],
{
# Config:
'batch_size' : 1024,
#'vars_input_slice',
#'vars_input_png2d',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset' : (
'numu/prod4/nd_fhc'
'/dataset_lstm_ee_nd_fhc_nonswap_loose_cut.csv.xz'
),
'early_stop' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'min_delta' : 0,
'patience' : 40,
},
},
'epochs' : 200,
'loss' : 'mean_absolute_percentage_error',
'max_prongs' : None,
'model' : {
'name' : 'lstm_v3',
'kwargs' : {
'batchnorm' : True,
'layers_pre' : [ 128, 128, 128 ],
'layers_post' : [ 128, 128, 128 ],
'lstm_units2d' : 32,
'lstm_units3d' : 32,
'n_resblocks' : 0,
},
},
'noise' : None,
'optimizer' : {
'name' : 'RMSprop',
'kwargs' : { 'lr' : 0.001 },
},
'prong_sorters' : None,
'regularizer' : {
'name' : 'l1',
'kwargs' : { 'l' : 0.001 },
},
'schedule' : {
'name' : 'standard',
'kwargs' : {
'monitor' : 'val_loss',
'factor' : 0.5,
'patience' : 5,
'cooldown' : 0
},
},
'seed' : 1337,
'steps_per_epoch' : 250,
'test_size' : 200000,
'weights' : None,
# Args:
'vars_mod_png2d' : None,
'vars_mod_png3d' : None,
'vars_mod_slice' : None,
'outdir' : (
'numu/prod4/04_nd_weights_fine_tune/02_weights_fine_tune_nd/'
),
}
)
config = join_dicts(
PRESETS_TRAIN['nue_v3'],
{
# Config:
'batch_size': 1024,
#'vars_input_slice',
#'vars_input_png3d',
#'vars_target_total',
#'vars_target_primary',
'dataset':
'nue/prod4/fd_fhc/dataset_lstm_ee_fd_fhc_fluxswap_std_cut.csv.xz',
'early_stop': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'min_delta': 0,
'patience': 40,
},
},
'epochs': 200,
'loss': 'mean_absolute_percentage_error',
'max_prongs': 5,
'model': {
'name': 'lstm_v1',
'kwargs': {
'batchnorm': False,
'lstm_units': 32,
},
},
'noise': None,
'optimizer': {
'name': 'RMSprop',
'kwargs': {
'lr': 0.001
},
},
'prong_sorters': None,
'regularizer': None,
'schedule': {
'name': 'standard',
'kwargs': {
'monitor': 'val_loss',
'factor': 0.5,
'patience': 5,
'cooldown': 0
},
},
'seed': 0,
'steps_per_epoch': 250,
'test_size': 200000,
#'test_size' : 0.1,
'vars_input_png2d': None,
'weights': None,
# Args:
'vars_mod_png2d': None,
'vars_mod_png3d': None,
'vars_mod_slice': None,
'outdir': 'nue/prod4/01_initial_studies/01_original/',
})
