def parse_mixin(less, parent=None, **kwargs):
match = MIXIN.match(less)
if not match:
raise ValueError()
code = match.group()
contents = None
if match.group('nested'):
matched_length = len(match.group())
remaining_less = less[matched_length:]
contents = parse_nested(remaining_less)
code += contents + '}'
params = parse_params(match.group('params'))
if contents:
for param in params:
if param['value'] and not param['name']:
param['name'] = param['value']
param['value'] = None
name = match.group('name')
if match.group('nested') and not match.group('param_detect'):
raise ValueError()
return Mixin(parent=parent, code=code, name=name, params=params,
contents=contents)
def __init__(self, initial_param_file, fasta_file):
"""
Initialize TSequence to the dataset you are interested in investigating
Can only be called with the following strings `s`:
'mu', '2mu', or '5mu'
e.g. TSequence('mu')
"""
self.sequences = read_fasta_sequences_to_str(fasta_file)
self.obs = observe_differences(self.sequences[0], self.sequences[1])
self.theta = parse_params(initial_param_file)
self.estimate = None
self.likelihood = None
self.initial_likelihood = None
def parse_mixin(less, parent=None, **kwargs):
match = MIXIN.match(less)
if not match:
raise ValueError()
code = match.group()
contents = None
if match.group('nested'):
matched_length = len(match.group())
remaining_less = less[matched_length:]
contents = parse_nested(remaining_less)
code += contents + '}'
params = parse_params(match.group('params'))
if contents:
for param in params:
if param['value'] and not param['name']:
param['name'] = param['value']
param['value'] = None
name = match.group('name')
if match.group('nested') and not match.group('param_detect'):
raise ValueError()
return Mixin(parent=parent,
code=code,
name=name,
params=params,
contents=contents)
import socket
import sys
import params
from framedSock import file_send
storage_dir = 'clientFiles/' # directory used to send/receive files
switches_var_defaults = (
(('-s', '--server'), 'server', '127.0.0.1:50001'),
(('-d', '--debug'), 'debug', False), # boolean (set if present)
(('-?', '--usage'), 'usage', False), # boolean (set if present)
)
prog_name = 'file_client'
param_map = params.parse_params(switches_var_defaults)
server, usage, debug = param_map['server'], param_map['usage'], param_map[
'debug']
if usage:
params.usage()
try:
server_host, server_port = re.split(':', server)
server_port = int(server_port)
except:
print('Could not parse server:port from ', server)
sys.exit(1)
s = None
for res in socket.getaddrinfo(server_host, server_port, socket.AF_UNSPEC,
def main():
# Parse args
args = parse_cmdline()
params = parse_params(args.template_path, args.param_path)
# TODO: Check if number of available GPUs is same as number specified
# No-reference quality metrics
if params.experiment.train_mode == 'noref':
print('### Experiment Type x (no ref) ###')
# Get metadata for HOG, None for CNN
for param_val in getattr(params.sensor, params.experiment.sat_param):
qual_pipeline = QualityPipeline(params, 'full', None, param_name=params.experiment.sat_param, param_val=param_val)
qual_pipeline.test()
# Normalization
elif params.experiment.train_mode == 'norm':
print('### Experiment Type 2 (norm) ###')
for fold_index in range(params.experiment.num_folds):
# Get metadata for HOG, None for CNN
for param_val in getattr(params.sensor, params.experiment.sat_param):
norm_pipeline = NormPipeline(params, 'full', fold_index, param_name=params.experiment.sat_param, param_val=param_val)
feat_mean, feat_std = norm_pipeline.norm()
# Experiment type 0
elif params.experiment.train_mode == 'basic':
print('### Experiment Type 0 ###')
for fold_index in range(params.experiment.num_folds):
if params.model.method == 'hog':
norm_pipeline = NormPipeline(params, 'train', fold_index)
feat_mean, feat_std = norm_pipeline.norm()
else:
feat_mean, feat_std = None, None
train_pipeline = TrainPipeline(params, 'train', fold_index, feat_mean=feat_mean, feat_std=feat_std)
test_pipeline = TestPipeline(params, 'val', fold_index, feat_mean=feat_mean, feat_std=feat_std)
train_pipeline.save()
test_pipeline.test(0)
for epoch_num in range(1, params.learning.num_epochs+1):
train_pipeline.train(save_flag=True)
test_pipeline.test(epoch_num)
# Experiment type 1
elif params.experiment.train_mode == 'orig':
print('### Experiment Type 1 ###')
mean_list, std_list = [], []
for fold_index in range(params.experiment.num_folds):
# Get metadata for HOG, None for CNN
if params.model.method == 'hog':
norm_pipeline = NormPipeline(params, 'train', fold_index)
feat_mean, feat_std = norm_pipeline.norm()
mean_list.append(feat_mean)
std_list.append(feat_std)
else:
mean_list.append(None), std_list.append(None)
# Create train pipeline
train_pipeline = TrainPipeline(params, 'train', fold_index,
feat_mean=mean_list[fold_index], feat_std=std_list[fold_index])
train_pipeline.save()
for epoch_num in range(params.learning.num_epochs):
train_pipeline.train(save_flag=True)
for fold_index in range(params.experiment.num_folds):
for param_val in getattr(params.sensor, params.experiment.sat_param):
test_pipeline = TestPipeline(params, 'val', fold_index,
param_name=params.experiment.sat_param, param_val=param_val,
feat_mean=mean_list[fold_index], feat_std=std_list[fold_index])
for epoch_num in range(params.learning.num_epochs+1):
test_pipeline.test(epoch_num)
# Experiment type 3
elif args.test_param_path is not None:
feat_mean, feat_std = None, None
test_params = parse_params(args.template_path, args.test_param_path)
print('### Experiment Type 3 ###')
# Get metadata for HOG, None for CNN
for param_val in getattr(params.sensor, params.experiment.sat_param):
for train_fold_index in range(params.experiment.num_folds):
load_pipeline = TestPipeline(params, 'val', train_fold_index,
param_name=params.experiment.sat_param, param_val=param_val,
feat_mean=feat_mean, feat_std=feat_std)
model_path = load_pipeline.get_path(5)
for test_fold_index in range(test_params.experiment.num_folds):
test_pipeline = TestPipeline(test_params, 'val', test_fold_index,
param_name=params.experiment.sat_param, param_val=param_val,
feat_mean=feat_mean, feat_std=feat_std, train_fold_index=train_fold_index)
test_pipeline.test_map(0, model_path=model_path)
# Experiment type 2
elif params.experiment.train_mode == 'degrade':
print('### Experiment Type 2 ###')
for fold_index in range(params.experiment.num_folds):
# Get metadata for HOG, None for CNN
for param_val in getattr(params.sensor, params.experiment.sat_param):
if params.model.method == 'hog':
norm_pipeline = NormPipeline(params, 'train', fold_index, param_name=params.experiment.sat_param, param_val=param_val)
feat_mean, feat_std = norm_pipeline.norm()
else:
feat_mean, feat_std = None, None
train_pipeline = TrainPipeline(params, 'train', fold_index, param_name=params.experiment.sat_param, param_val=param_val,
feat_mean=feat_mean, feat_std=feat_std)
test_pipeline = TestPipeline(params, 'val', fold_index, param_name=params.experiment.sat_param, param_val=param_val,
feat_mean=feat_mean, feat_std=feat_std)
train_pipeline.save()
test_pipeline.test(0)
# Train for num_epochs
for _ in range(1, params.learning.num_epochs+1):
epoch_num = train_pipeline.train(save_flag=True)
test_pipeline.test(epoch_num)
# Invalid experiment type
else:
raise Exception('Invalid experiment train_mode: {}'.format(params.experiment.train_mode))
from sys import argv import tensorflow as tf import numpy as np from data.preprocessing import get_dataset, augment from unet.model import unet from unet.loss import superloss from unet.metrics import Dice, DiceX, DiceG, DiceM from unet.utils import crop_size from unet.preconf import conf os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true' params = parse_params(argv[1:]) # initialization of the data split and classes # the list of id for images used in training imgs_train = params.imgs_train # list of ids of the classes terms = params.terms # list of the id for images used in training to select the model to keep imgs_val = params.imgs_val # the list of id for images used in testing for which fullsize mask will be written imgs_test = params.imgs_test datadir = params.datadir
from ssn.ssn_dataset import SSN_Dataset
# from ssn.ssn_submodule import Contract
from ssn.ssn import Relight_SSN, baseline_2_tbaseline, baseline_2_touchloss
from utils.utils_file import get_cur_time_stamp, create_folder
from utils.net_utils import save_model, get_lr, set_lr
from utils.visdom_utils import setup_visdom, visdom_plot_loss, visdom_log, visdom_show_batch
from utils.tensorboard_utils import *
from torch.utils.tensorboard import SummaryWriter
from params import params as options, parse_params
import matplotlib.pyplot as plt
from evaluation import exp_predict, exp_metric
import pickle
# parse args
params = parse_params()
print("Params: {}".format(params))
exp_name = params.exp_name
# cur_viz = setup_visdom(params.vis_port)
tensorboard_folder = 'tensorboard_log/runs'
os.makedirs(tensorboard_folder, exist_ok=True)
timestamp = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
writer = SummaryWriter(
join(tensorboard_folder, '{}_{}'.format(exp_name, timestamp)))
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if params.cpu:
device = torch.device('cpu')
print("Device: ", device)