def GraficarLearningCurve(self, id_experiment, id_weigths_1_de_4,
id_weigths_2_de_4, id_weigths_3_de_4,
id_weigths_4_de_4):
wr = WeigthsRepo(database_name=self.data_base,
id_experiment=id_experiment)
data_weigths_1_de_4 = wr.GetWeithsInfoById(id_weigths_1_de_4)
test_error_1_de_4 = data_weigths_1_de_4[13]
training_error_1_de_4 = data_weigths_1_de_4[8]
data_weigths_2_de_4 = wr.GetWeithsInfoById(id_weigths_2_de_4)
test_error_2_de_4 = data_weigths_2_de_4[13]
training_error_2_de_4 = data_weigths_2_de_4[8]
data_weigths_3_de_4 = wr.GetWeithsInfoById(id_weigths_3_de_4)
test_error_3_de_4 = data_weigths_3_de_4[13]
training_error_3_de_4 = data_weigths_3_de_4[8]
data_weigths_4_de_4 = wr.GetWeithsInfoById(id_weigths_4_de_4)
test_error_4_de_4 = data_weigths_4_de_4[13]
training_error_4_de_4 = data_weigths_4_de_4[8]
test_x_size = np.asarray([175500, 351000, 526500, 684000], dtype=int)
test_y_error = np.asarray([
test_error_1_de_4, test_error_2_de_4, test_error_3_de_4,
test_error_4_de_4
],
dtype=np.float64)
train_x_size = np.asarray([175500, 351000, 526500, 684000], dtype=int)
train_y_error = np.asarray([
training_error_1_de_4, training_error_2_de_4,
training_error_3_de_4, training_error_4_de_4
],
dtype=np.float64)
my_xticks = [175500, 351000, 526500, 684000]
plt.xticks(test_x_size, my_xticks)
plt.xticks(train_x_size, my_xticks)
plt.plot(
test_x_size,
test_y_error,
'r--',
)
plt.plot(train_x_size, train_y_error, 'b--')
plt.ylabel('Error')
plt.xlabel('Tamaño training set')
plt.title('Learning curve(' + str(id_experiment) + ')')
plt.show()
return
def BuildWeigthsErrorAndCost_TestSet(self, noRows_Testset, skipuntilidw,
justOne):
# Calculo total del costo y error por todos los datos, pero por cada conjunto de pesos generados
# experiment_repo = Experiments.ExperimentsRepo.ExperimentsRepo(bd, id_experiment)
# weigths_repo = WeigthsRepo.WeigthsRepo(bd, weigths_path)
database_relative_path = self.data_base
id_experiment = self.id_experiment
wr = WeigthsRepo(database_name=database_relative_path,
id_experiment=id_experiment)
wm = WeightsManager(database_name=database_relative_path,
weights_repo=wr)
weigthsOfExperiment = wm.GetListOfWeightsByIdExperiment(id_experiment)
print(
'--------------------------- Test SET -------------------------------------------------'
)
print("Calculando Errores en TestSet y costos en Test set")
for w in weigthsOfExperiment:
idW = w[0]
if idW < skipuntilidw:
continue
# Ahora cargamos los weights
iws = wm.LoadWeightsXId(idW)
random_droput = np.random.RandomState(12345)
rng_droput = T.shared_randomstreams.RandomStreams(
random_droput.randint(999899))
# terminar de instanciar esto
cnn = CNNGCresc(
layers_metaData=layers_metaData,
initWeights=iws,
srng=rng_droput,
no_channels_imageInput=1,
isTraining=0,
pDropOut=0.7 # antes 0.60
)
# exp 9 pDropout=0.65
logger = LoggerRepo(id_experiment=id_experiment,
database_name=database_relative_path)
experimentsManager = ExperimentsManager(database_relative_path)
self.experiment = experimentsManager.LoadExpermentById(
self.id_experiment)
self.data_set_repo = DataSetRepo(
list_PKL_files=self.experiment.pkl_test_referenceList,
batch_size=self.experiment.batch_size,
superbatch_Size=self.experiment.super_batch_size,
no_rows=noRows_Testset)
self.data_set_manager = DataSetManager(
self.experiment.batch_size, self.experiment.super_batch_size,
self.data_set_repo)
validator = Validator_ValTest(
data_set_manager=self.data_set_manager,
logger=logger,
cnn=cnn,
weightsRepo=wr)
averageError = validator.CalculateError()
wm.UpdateTestErrorWeigth(idW, averageError)
print("--------[Test Set] El error promedio es: " +
str(averageError))
averageCost = validator.CalculateCost()
wm.UpdateTestCostWeigth(idW, averageCost)
print("--------[Test Set] El costo promedio es: " +
str(averageCost))
if justOne == True:
break
print("End Test :)")
def BuildResults_ValSet(self, noRows_valset, skipuntilidw, justOne):
# Calculo total del costo y error por todos los datos, pero por cada conjunto de pesos generados
#experiment_repo = Experiments.ExperimentsRepo.ExperimentsRepo(bd, id_experiment)
#weigths_repo = WeigthsRepo.WeigthsRepo(bd, weigths_path)
database_relative_path = self.data_base
id_experiment = self.id_experiment
wr = WeigthsRepo(database_name=database_relative_path,
id_experiment=id_experiment)
wm = WeightsManager(database_name=database_relative_path,
weights_repo=wr)
weigthsOfExperiment = wm.GetListOfWeightsByIdExperiment(id_experiment)
print(
'--------------------------- Validation SET -------------------------------------------------'
)
print("Calculando Errores en validationSet y costos en validation set")
csvfile = open(
r'D:\Gyo\Dev\Thesis\dist2\analisys\DataSet_analisys_resultts.csv',
'w',
newline='')
csvwriter = csv.writer(csvfile, delimiter=',', quoting=csv.QUOTE_NONE)
for w in weigthsOfExperiment:
idW = w[0]
if idW < skipuntilidw:
continue
# Ahora cargamos los weights
iws = wm.LoadWeightsXId(idW)
random_droput = np.random.RandomState(12345)
rng_droput = T.shared_randomstreams.RandomStreams(
random_droput.randint(999899))
# terminar de instanciar esto
cnn = CNNGCresc(
layers_metaData=layers_metaData,
initWeights=iws,
srng=rng_droput,
no_channels_imageInput=1,
isTraining=0,
pDropOut=0.7 # antes 0.60
)
# exp 9 pDropout=0.65
logger = LoggerRepo(id_experiment=id_experiment,
database_name=database_relative_path)
experimentsManager = ExperimentsManager(database_relative_path)
self.experiment = experimentsManager.LoadExpermentById(
self.id_experiment)
self.data_set_repo = DataSetRepo(
list_PKL_files=self.experiment.pkl_validation_referenceList,
batch_size=self.experiment.batch_size,
superbatch_Size=self.experiment.super_batch_size,
no_rows=noRows_valset)
self.data_set_manager = DataSetManager(
self.experiment.batch_size, self.experiment.super_batch_size,
self.data_set_repo)
validator = Validator_ValTest(
data_set_manager=self.data_set_manager,
logger=logger,
cnn=cnn,
weightsRepo=wr,
removeRandom=True)
result = validator.CalculateResults()
rrr = zip(result[0], result[1], result[2])
for rr in rrr:
csvwriter.writerow([rr[0], rr[1], rr[2]])
csvfile.close()
if justOne == True:
break
print("End Validation :)")
from Core.ExperimentsManager import ExperimentsManager
from Core.Predictor import Predictor
from Core.Trainer import Trainer
from Core.WeightsManager import WeightsManager
from IGenericServices.ExperimentsRepo import ExperimentsRepo
from IGenericServices.LoggerRepo import LoggerRepo
from IGenericServices.WeightsRepo import WeigthsRepo
from PIL import Image
database_relative_path = "../BD/FR2.0.db"
idExperiment = 10
idW = 265
#Ahora cargamos los weights
wr = WeigthsRepo(database_name=database_relative_path,
id_experiment=idExperiment)
wm = WeightsManager(database_name=database_relative_path, weights_repo=wr)
iws = wm.LoadWeightsXId(idW)
random_droput = np.random.RandomState(12345)
rng_droput = T.shared_randomstreams.RandomStreams(
random_droput.randint(999899))
#terminar de instanciar esto
cnn = CNNGCresc(layers_metaData=layers_metaData,
initWeights=iws,
srng=rng_droput,
no_channels_imageInput=1,
isTraining=1,
pDropOut=0.7)