#model_name = r'birdnet_dropout_conv'
if args.model_name is None:
#model_name = "lenet"
model_name = 'birdnet'
else:
model_name = args.model_name
print('Model name: ' + model_name)
if args.dataset_name is None:
dataset_name = "Data_2"
else:
dataset_name = args.dataset_name
print('Dataset name: ' + dataset_name)
##################################
# DATASET & MODEL
config = Config(dataset=dataset_name, model=model_name)
#config = Config(dataset='Data_2', model='lenet')
#config = Config(dataset='Data_2', model='birdnet') # failed a couple of times, probably needs augmentation
#config = Config(dataset='Data_2', model='lenet_dropout_conv')
#config = Config(dataset='Data_2', model='birdnet_dropout_conv')
##################################
if args.augment is True:
config.DO_AUGMENT = True
print('Data augmentation is used for conv-based nets.\n')
else:
config.DO_AUGMENT = False
print('No data augmentation is used for conv-based nets.\n')
##################################
if (model_name == 'recurr_lstm') or (model_name == r'conv1d_gru'):
"""
import os, sys
ROOT_DIR = os.path.abspath("/home/kpalmer/AnacondaProjects/upcall-basic-net")
# the path where upcall-basic-net is
sys.path.append(ROOT_DIR)
from upcall.accuracy_measure import map_build_day_file
import glob, os
from upcall.config import Config
from upcall.full_process import full_process_interfaceDSP, full_process_interface
from upcall.RunDetectorDSPdependency import MakeSoundStream
#import matplotlib.pyplot as plt
# parameter
config = Config()
DATA_DIR = r'/home/ys587/__Data'
# truth data:
TruthFolder = r'/home/ys587/__Data/__TruthLabelVAPlusStAndrew'
TruthFolder = r'/cache/kpalmer/quick_ssd/data/dclmmpa2013/LogsWithNewSNR/RavenST/NOPPSet2'
day_file_map = map_build_day_file(TruthFolder)
# testing sound stream of St Andrew DCL
SoundPath = os.path.join(DATA_DIR, r'DCL_St_Andrew/Sound_3_days')
day_list = sorted(glob.glob(os.path.join(TruthFolder + '/', '*')))
SoundFileLoc = '/cache/kpalmer/quick_ssd/data/dclmmpa2013/Testing/Upcalls_NOPPset2'
file_days = os.listdir(SoundFileLoc)
# create a datafram to hold precision-recall cuver of all runs; for the seaborn
pd_prc_list = []
for ii in range(len(pre_rec_list)):
#pre_rec_array_list.append(np.array(pre_rec_list[ii]))
pd_prc = pd.DataFrame(np.array(pre_rec_list[ii]).T,
columns=['mpre', 'mrec'])
pd_prc['run'] = ii
pd_prc_list.append(pd_prc)
pd_prc_all_runs = pd.concat(pd_prc_list)
pd_prc_all_runs.to_csv(os.path.join(seltab_detect_path, 'pre_rec.txt'),
sep='\t')
if __name__ == "__main__":
config = Config()
config.NUM_RUNS = 10
# DCL13
truth_folder = os.path.join(
DATA_DIR, 'DCL_St_Andrew/Sound_3_days_seltab') # DCL13 test dataset
func_folder = '__full_data' # could be __full_data_large as well
# seltab_detect_path = r'/home/ys587/__ExptResult/__V4_Paper'
seltab_detect_path = r'/home/ys587/__ExptResult/__V4_Paper/__result_only_20190114/__negtive_mining'
model_path = glob.glob(seltab_detect_path + '/cv_*')
model_path.sort()
for mm in model_path:
print(os.path.basename(mm))
seltab_detect_path = os.path.join(mm, func_folder)
# the path where upcall-basic-net is
sys.path.append(ROOT_DIR)
from upcall.config import Config
from upcall.train_classifier import lenet, lenet_dropout_input, \
lenet_dropout_conv, lenet_dropout_input_conv, birdnet, birdnet_dropout_input, \
birdnet_dropout_conv, birdnet_dropout_input_conv
from upcall.train_classifier import net_train, model_preprocess
from script.DCLDE2018_train_data import prepare_truth_data
import time
start_time = time.time()
#config = Config(dataset='Data_4', model='birdnet_dropout_input_conv')
config = Config(dataset='Data_1', model='lenet')
#config = Config('Data_1', 'Lenet_dropout_conv') # model of the test case 38-0.8361.hdf5
##################################
# Model of DCLDE 2018 # Build models here
##################################
try:
# use model name in string as function name to generate the model
func_model_generate = globals()[config.MODEL]
model = func_model_generate(config)
except:
print(
"ModelError: Model name is either incorrect or model is not accepted.")
sys.exit()
#if config.MODEL == r'lenet':
from upcall.train_classifier import lenet, lenet_dropout_input, \
lenet_dropout_conv, lenet_dropout_input_conv, birdnet, birdnet_dropout_input, \
birdnet_dropout_conv, birdnet_dropout_input_conv
from upcall.train_classifier import net_train, model_preprocess, net_train_augment
from script.DCLDE2018_train_data import prepare_truth_data
import time
start_time = time.time()
##################################
# select a dataset to train the classifier and a model to run.
##################################
#config = Config(dataset='Data_2', model='birdnet')
#config = Config(dataset='Data_4', model='birdnet_dropout_input_conv')
config = Config(dataset='Data_2', model='birdnet_dropout_conv')
#config = Config(dataset='Data_2', model='lenet')
#config = Config(dataset='Data_2', model='lenet_dropout_conv')
#config = Config('Data_1', 'Lenet_dropout_conv') # model of the test case 38-0.8361.hdf5
##################################
# Classifier training
##################################
# Model of DCLDE 2018 # Build models here
try:
# use model name in string as function name to generate the model
func_model_generate = globals()[config.MODEL]
model = func_model_generate(config)
except:
print(
"ModelError: Model name is either incorrect or model is not accepted.")
Created on 7/11/19
@author: atoultaro
"""
import os
import glob
from upcall.config import Config
from upcall.accuracy_measure import calc_TP_FP_FN, map_build_day_file, \
get_date_str
result_path = os.path.join(
'/home/ys587/__ExptResult/cv_lenet_dropout_conv_train_Data_2_augment',
'__full_data_large')
DATA_DIR = r'/home/ys587/__Data' # Data
truth_folder = os.path.join(DATA_DIR, '__large_test_dataset/__truth_seltab')
config = Config(dataset='Data_2', model='lenet_dropout_conv')
config.NUM_RUNS = 1
for rr in range(config.NUM_RUNS):
day_list = sorted(
glob.glob(
os.path.join(result_path, 'Run' + str(rr), '__TP_FN_FP',
'*.txt'))) # find selection table for all days
print(rr)
for dd in day_list:
basename = os.path.basename(dd)
map_day_file = map_build_day_file(truth_folder)
truth_file = map_day_file[get_date_str(basename)]
accu_result, _ = calc_TP_FP_FN(dd, truth_file, result_path,
config) # TP, FP, TP2, FN
start_time = time.time()
proj_path = u'/mnt/drive_W/projects/2018_ORStateU_CCB_85941/Dep01'
sound_dir = os.path.join(proj_path, r'AIFFs_UTCz')
# sound_dir = r'/mnt/drive_W/projects/2018_ORStateU_CCB_85941/Dep01/detection_results/__sound_test' # for test
model_path = os.path.join(
proj_path,
r'detection_results/__model/lenet_dropout_conv_train_Data_6_augment_Run0_epoch_100_F1_0.8733.hdf5'
)
# seltab_detect_output = os.path.join(proj_path, r'detection_results/__seltab')
seltab_detect_output = os.path.join(proj_path,
r'detection_results/__seltab_test_drive')
best_model_path, best_accu = find_best_model(model_path)
config = Config(dataset=None, model=u"lenet_dropout_conv")
config.SCORE_THRE = 0.5
print('Detecting:')
day_list = os.listdir(sound_dir)
day_list2 = []
for dd in day_list:
day_list2.append(sound_dir + '/' + dd)
day_list = day_list2
del day_list2
day_list.sort()
# day_list = day_list[1:] # avoid the first one, 20190216 has problem
# day_list = day_list[0]
day_list_err = [day_list[ii] for ii in [0]] # debugging 20190215 & 20190612
day_list = day_list_err