def fill_params(expt_name, chkpt_num, batch_sz, gpus,
sampler_fname, model_fname, augmentor_fname, **args):
params = {}
#Model params
params["in_dim"] = 1
params["output_spec"] = collections.OrderedDict(synapse_label=1)
params["depth"] = 4
params["batch_norm"] = True
#Training procedure params
params["max_iter"] = 1000000
params["lr"] = 0.00001
params["test_intv"] = 100
params["test_iter"] = 10
params["avgs_intv"] = 50
params["chkpt_intv"] = 10000
params["warm_up"] = 50
params["chkpt_num"] = chkpt_num
params["batch_size"] = batch_sz
#Sampling params
params["data_dir"] = os.path.expanduser("~/seungmount/research/Nick/datasets/SNEMI3D/")
assert os.path.isdir(params["data_dir"]),"nonexistent data directory"
params["train_sets"] = ["K_val"]
params["val_sets"] = ["K_val"]
params["patchsz"] = (18,160,160)
params["sampler_spec"] = dict(input=params["patchsz"],
synapse_label=params["patchsz"])
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = "experiments/{}".format(expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["tb_train"] = os.path.join(params["expt_dir"], "tb/train")
params["tb_val"] = os.path.join(params["expt_dir"], "tb/val")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
params["sampler_class"] = utils.load_source(sampler_fname).Sampler
params["augmentor_constr"] = utils.load_source(augmentor_fname).get_augmentation
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [params["in_dim"], params["output_spec"],
params["depth"]]
params["model_kwargs"] = {"bn" : params["batch_norm"]}
#modules used for record-keeping
params["modules_used"] = [__file__, model_fname, sampler_fname,
augmentor_fname, "loss.py"]
return params
def create_features(features):
import timeit
source = load_source()
start = timeit.default_timer()
compute_features(source, features)
end = timeit.default_timer()
print("save all features takes ", (end-start))
def create_features(features):
import timeit
source = load_source()
start = timeit.default_timer()
compute_features(source, features)
end = timeit.default_timer()
print("save all features takes ", (end - start))
def fill_params(expt_name, chkpt_num, gpus, nobn, model_name, dset_names, tag):
params = {}
#Model params
params["in_dim"] = 1
params["output_spec"] = collections.OrderedDict(soma_label=1)
params["depth"] = 4
params["batch_norm"] = not (nobn)
params["activation"] = F.sigmoid
params["chkpt_num"] = chkpt_num
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params[
"expt_dir"] = "/jukebox/wang/zahra/conv_net/training/experiment_dirs/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["log_tag"] = "fwd_" + tag if len(tag) > 0 else "fwd"
params["output_tag"] = tag
#Dataset params
params[
"data_dir"] = "/jukebox/wang/pisano/conv_net/annotations/all_better_res/h129/otsu/inputRawImages"
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["dsets"] = dset_names
params["input_spec"] = collections.OrderedDict(
input=(20, 192, 192)) #dp dataset spec
params["scan_spec"] = collections.OrderedDict(soma_label=(1, 20, 192, 192))
params["scan_params"] = dict(stride=(0.75, 0.75, 0.75), blend="bump")
#Use-specific Module imports
params["model_class"] = utils.load_source('models/RSUNet.py').Model
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_dim"], params["output_spec"], params["depth"]
]
params["model_kwargs"] = {"bn": params["batch_norm"]}
#Modules used for record-keeping
params["modules_used"] = [__file__, 'models/RSUNet.py', "layers.py"]
return params
def fill_params(expt_name, chkpt_num, gpus, nobn, model_fname, dset_names,
tag):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 192, 192))
params["output_spec"] = collections.OrderedDict(soma=(1, 20, 192, 192))
params["width"] = [32, 40, 80]
params["activation"] = sigmoid
params["chkpt_num"] = chkpt_num
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params[
"expt_dir"] = "/jukebox/wang/zahra/conv_net/training/prv/experiment_dirs/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["log_tag"] = "fwd_" + tag if len(tag) > 0 else "fwd"
params["output_tag"] = tag
#Dataset params
params[
"data_dir"] = "/home/wanglab/mounts/wang/zahra/conv_net/annotations/prv/hypothalamus/inputs"
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["dsets"] = dset_names
params["input_spec"] = collections.OrderedDict(
input=(20, 192, 192)) #dp dataset spec
params["scan_spec"] = collections.OrderedDict(soma=(1, 20, 192, 192))
params["scan_params"] = dict(stride=(0.5, 0.5, 0.5), blend="bump")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#Modules used for record-keeping
params["modules_used"] = [__file__, model_fname, "layers.py"]
return params
def fill_params(expt_name, chkpt_num, gpus, nobn, model_fname, dset_name, tag,
jobid):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 32, 32))
params["output_spec"] = collections.OrderedDict(soma=(1, 20, 32, 32))
params["width"] = [32, 40, 80]
params["activation"] = sigmoid
params["chkpt_num"] = chkpt_num
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = "/tigress/zmd/3dunet_data/cfos/experiments/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["log_tag"] = "fwd_" + tag if len(tag) > 0 else "fwd"
params["output_tag"] = tag
params["jobid"] = jobid
#Dataset params
params["data_dir"] = "/scratch/gpfs/zmd/{}".format(dset_name)
# assert os.path.isdir(params["data_dir"]),"nonexistent data directory"
params["dsets"] = dset_name
params["input_spec"] = collections.OrderedDict(
input=(20, 32, 32)) #dp dataset spec
params["scan_spec"] = collections.OrderedDict(soma=(1, 20, 32, 32))
params["scan_params"] = dict(stride=(0.1, 0.1, 0.1), blend="bump")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#Modules used for record-keeping
params["modules_used"] = [__file__, model_fname, "layers.py"]
return params
def fill_params(expt_name, chkpt_num, gpus,
nobn, model_name, tag, jobid):
params = {}
#Model params
params["in_dim"] = 1
params["output_spec"] = collections.OrderedDict(soma_label=1)
params["depth"] = 4
params["batch_norm"] = not(nobn)
params["activation"] = F.sigmoid
params["chkpt_num"] = chkpt_num
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = "/tigress/zmd/3dunet_data/experiments/{}".format(expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["log_tag"] = "fwd_" + tag if len(tag) > 0 else "fwd"
params["output_tag"] = tag
#Dataset params
params["data_dir"] = "/scratch/gpfs/zmd/20180327_jg40_bl6_sim_03"
assert os.path.isdir(params["data_dir"]),"nonexistent data directory"
params["jobid"] = jobid
params["input_spec"] = collections.OrderedDict(input=(20,192,192)) #dp dataset spec
params["scan_spec"] = collections.OrderedDict(soma_label=(1,20,192,192))
params["scan_params"] = dict(stride=(0.75,0.75,0.75), blend="bump")
#Use-specific Module imports
params["model_class"] = utils.load_source('models/RSUNet.py').Model
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [params["in_dim"], params["output_spec"],
params["depth"] ]
params["model_kwargs"] = { "bn" : params["batch_norm"] }
#Modules used for record-keeping
params["modules_used"] = [__file__, 'models/RSUNet.py', "layers.py"]
return params
def fill_params(expt_name, chkpt_num, batch_sz, gpus, sampler_fname,
model_fname, augmentor_fname, **args):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 192, 192))
params["output_spec"] = collections.OrderedDict(cleft=(1, 20, 192, 192))
params["width"] = [32, 40, 80]
#Training procedure params
params["max_iter"] = 1000000
params["lr"] = 0.00001
params["test_intv"] = 100
params["test_iter"] = 10
params["avgs_intv"] = 50
params["chkpt_intv"] = 1000
params["warm_up"] = 50
params["chkpt_num"] = chkpt_num
params["batch_size"] = batch_sz
#Sampling params
params["data_dir"] = "/tigress/zmd/3dunet_data/ctb/training_inputs"
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["train_sets"] = [
"z269stackstart150", "z269stackstart475", "z266stackstart350",
"z266stackstart250", "z268stackstart300",
"z265_zpln165-191_x6325_y4458", "z265_zpln315-340_x4785_y3793"
]
params["val_sets"] = ["z269stackstart100"]
params["patchsz"] = (20, 192, 192)
params["sampler_spec"] = dict(input=params["patchsz"],
soma_label=params["patchsz"])
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = "/tigress/zmd/3dunet_data/ctb/network/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["tb_train"] = os.path.join(params["expt_dir"], "tb/train")
params["tb_val"] = os.path.join(params["expt_dir"], "tb/val")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
params["sampler_class"] = utils.load_source(sampler_fname).Sampler
params["augmentor_constr"] = utils.load_source(
augmentor_fname).get_augmentation
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#modules used for record-keeping
params["modules_used"] = [
__file__, model_fname, sampler_fname, augmentor_fname, "loss.py"
]
return params
def fill_params_train(expt_name, batch_sz, gpus, sampler_fname, model_fname,
augmentor_fname, **args):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 192, 192))
params["output_spec"] = collections.OrderedDict(soma=(1, 20, 192, 192))
params["width"] = [32, 40, 80]
#Training procedure params
params["max_iter"] = 51
params["lr"] = 0.00001
params["test_intv"] = 25
params["test_iter"] = 10
params["avgs_intv"] = 10
params["chkpt_intv"] = 10
params["warm_up"] = 5
params["chkpt_num"] = 0
params["batch_size"] = batch_sz
#Sampling params
print("the working directory is: {}\n".format(os.getcwd()))
params["data_dir"] = os.path.join(os.path.dirname(os.getcwd()), 'demo')
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["train_sets"] = ["train"]
params["val_sets"] = ["val"]
params["patchsz"] = (20, 192, 192)
params["sampler_spec"] = dict(input=params["patchsz"],
soma_label=params["patchsz"])
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = os.path.join(params["data_dir"],
"experiments/{}".format(expt_name))
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["tb_train"] = os.path.join(params["expt_dir"], "tb/train")
params["tb_val"] = os.path.join(params["expt_dir"], "tb/val")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
params["sampler_class"] = utils.load_source(sampler_fname).Sampler
params["augmentor_constr"] = utils.load_source(
augmentor_fname).get_augmentation
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#modules used for record-keeping
params["modules_used"] = [
__file__, model_fname, sampler_fname, augmentor_fname, "loss.py"
]
return params
def fill_params(expt_name, chkpt_num, batch_sz, gpus, sampler_fname,
model_fname, augmentor_fname, **args):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 192, 192))
params["output_spec"] = collections.OrderedDict(cleft=(1, 20, 192, 192))
params["width"] = [32, 40, 80]
#Training procedure params
params["max_iter"] = 1000000
params["lr"] = 0.00001
params["test_intv"] = 100
params["test_iter"] = 10
params["avgs_intv"] = 50
params["chkpt_intv"] = 10
params["warm_up"] = 50
params["chkpt_num"] = chkpt_num
params["batch_size"] = batch_sz
#Sampling params
params["data_dir"] = "/tigress/zmd/3dunet_data/prv/all_inputs/"
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["train_sets"] = [
'zd_ann_prv_jg05_neocortex_z310-449_01',
'zd_ann_prv_jg24_neocortex_z300-400_01',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0650-0700_01',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0450-0500_01',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0450-0500_00',
'cj_ann_prv_jg05_neocortex_z250-449_02',
'zd_ann_prv_jg32_hypothal_z710-810_02',
'cj_ann_prv_jg05_hypothal_z661-760_02',
'JGANNOTATION_20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_01',
'cj_ann_prv_jg29_neocortex_z700-800_02',
'cj_ann_prv_jg24_neocortex_z300-400_01',
'cj_ann_prv_jg32_neocortex_z650-810_01',
'20180305_jg_bl6f_prv_11_647_010na_7d5um_250msec_10povlp_ch00_C00_300-345_00',
'JGANNOTATION_20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0450-0500_01',
'cj_ann_prv_jg24_neocortex_z300-400_02',
'cj_ann_prv_jg24_hypothal_z550-650_01',
'cj_ann_prv_jg29_hypothal_z580-700_01',
'zd_ann_prv_jg29_neocortex_z300-500_01',
'20180305_jg_bl6f_prv_11_647_010na_7d5um_250msec_10povlp_ch00_C00_300-345_01',
'zd_ann_prv_jg32_neocortex_z650-810_01',
'zd_ann_prv_jg24_hypothal_z550-650_01',
'20180215_jg_bl6f_prv_10_647_010na_z7d5um_250msec_10povlap_ch00_z200-400_y4500-4850_x2050-2400',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0650-0700_06',
'cj_ann_prv_jg32_hypothal_z650-810_01',
'20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_01',
'JGANNOTATION_20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0450-0500_02',
'20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_03',
'cj_ann_prv_jg05_neocortex_z310-449_01'
]
params["val_sets"] = [
'cj_ann_prv_jg29_hypothal_z700-800_02',
'JGANNOTATION_20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_02',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0650-0700_00',
'20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_02',
'JGANNOTATION_20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_03',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0450-0500_06',
'20180306_jg_bl6f_prv_16_647_010na_7d5um_250msec_10povlp_ch00_C00_Z0650-0700_05',
'20180305_jg_bl6f_prv_12_647_010na_7d5um_250msec_10povlp_ch00_C00_400-440_00'
]
params["patchsz"] = (20, 192, 192)
params["sampler_spec"] = dict(input=params["patchsz"],
soma_label=params["patchsz"])
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params["expt_dir"] = "/tigress/zmd/3dunet_data/prv/experiments/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["tb_train"] = os.path.join(params["expt_dir"], "tb/train")
params["tb_val"] = os.path.join(params["expt_dir"], "tb/val")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
params["sampler_class"] = utils.load_source(sampler_fname).Sampler
params["augmentor_constr"] = utils.load_source(
augmentor_fname).get_augmentation
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#modules used for record-keeping
params["modules_used"] = [
__file__, model_fname, sampler_fname, augmentor_fname, "loss.py"
]
return params
def fill_params(expt_name, chkpt_num, batch_sz, gpus, sampler_fname,
model_fname, augmentor_fname, **args):
params = {}
#Model params
params["in_spec"] = dict(input=(1, 20, 32, 32))
params["output_spec"] = collections.OrderedDict(cleft=(1, 20, 32, 32))
params["width"] = [32, 40, 80]
#Training procedure params
params["max_iter"] = 1000000
params["lr"] = 0.00001
params["test_intv"] = 100
params["test_iter"] = 10
params["avgs_intv"] = 50
params["chkpt_intv"] = 1000
params["warm_up"] = 50
params["chkpt_num"] = chkpt_num
params["batch_size"] = batch_sz
#Sampling params
params[
"data_dir"] = "/home/wanglab/Documents/cfos_inputs/otsu_and_guassian_screened"
assert os.path.isdir(params["data_dir"]), "nonexistent data directory"
params["train_sets"] = [
'dp_ann_201812_pcdev_lob6_4_forebrain_cortex_z200-219',
'tp_ann_201812_pcdev_lob6_9_forebrain_hypothal_z520-539',
'tp_ann_201812_pcdev_crus1_23_forebrain_cortex_z290-309',
'jd_ann_201904_an19_ymazefos_020719_thal_z350-369',
'jd_ann_201904_an21_ymazefos_020719_hypothal_z450-469',
'dp_ann_201904_an19_ymazefos_020719_pfc_z380-399',
'dp_ann_201904_an21_ymazefos_020719_hypothal_z450-469',
'tp_ann_201904_an10_ymzefos_020719_cortex_z280-279',
'jd_ann_201904_an22_ymazefos_020719_pfc_z150-169',
'jd_ann_201904_an22_ymazefos_020719_cb_z160-179',
'dp_ann_201904_an22_ymazefos_020719_cb_z160-179',
'tp_ann_201904_an19_ymazefos_020719_pfc_z380-399',
'dp_ann_201904_an12_ymazefos_020719_hypothal_z420-449',
'tp_ann_201812_pcdev_crus1_23_forebrain_midbrain_z260-279',
'tp_ann_201904_an4_ymazefos_020119_cortex_z200-219',
'tp_ann_201904_an4_ymazefos_020119_pfc_z200-219',
'tp_ann_201812_pcdev_lob6_4_forebrain_cortex_z200-219',
'dp_ann_201904_an19_ymazefos_020719_cortex_z380-399_02',
'tp_ann_201904_an22_ymazefos_020719_pfc_z150-169',
'jd_ann_201904_an30_ymazefos_020719_pfc_z410-429',
'jd_ann_201904_an10_ymazefos_020719_hypothal_z460-479',
'jd_ann_201904_an10_ymazefos_020719_pb_z260-279',
'jd_ann_201904_an30_ymazefos_020719_cortex_z400-419',
'dp_ann_201904_an19_ymazefos_020719_cortex_z350-369',
'dp_ann_an16_ymazecfos_z260-299_retrosplenial_cropped'
]
params["val_sets"] = [
'dp_ann_201904_an12_ymazefos_020719_cortex_z371-390',
'dp_ann_201904_an19_ymazefos_020719_cb_z380-399',
'dp_ann_201812_pcdev_lob6_9_forebrain_hypothal_z520-539',
'jd_ann_201904_an30_ymazefos_020719_striatum_z416-435',
'tp_ann_201904_an30_ymazefos_020719_striatum_z416-435',
'dp_ann_an22_ymazecfos_z230-249_sm_cortex_cropped',
'dp_ann_201904_an19_ymazefos_020719_thal_z350-369'
]
params["patchsz"] = (20, 32, 32)
params["sampler_spec"] = dict(input=params["patchsz"],
soma_label=params["patchsz"])
#GPUS
params["gpus"] = gpus
#IO/Record params
params["expt_name"] = expt_name
params[
"expt_dir"] = "/home/wanglab/Documents/cfos_net/experiment_dirs/{}".format(
expt_name)
params["model_dir"] = os.path.join(params["expt_dir"], "models")
params["log_dir"] = os.path.join(params["expt_dir"], "logs")
params["fwd_dir"] = os.path.join(params["expt_dir"], "forward")
params["tb_train"] = os.path.join(params["expt_dir"], "tb/train")
params["tb_val"] = os.path.join(params["expt_dir"], "tb/val")
#Use-specific Module imports
params["model_class"] = utils.load_source(model_fname).Model
params["sampler_class"] = utils.load_source(sampler_fname).Sampler
params["augmentor_constr"] = utils.load_source(
augmentor_fname).get_augmentation
#"Schema" for turning the parameters above into arguments
# for the model class
params["model_args"] = [
params["in_spec"], params["output_spec"], params["width"]
]
params["model_kwargs"] = {}
#modules used for record-keeping
params["modules_used"] = [
__file__, model_fname, sampler_fname, augmentor_fname, "loss.py"
]
return params
