def predict_url(args):
"""
Function to predict an url
"""
# Check user configuration
update_with_query_conf(args)
conf = config.conf_dict
merge = True
catch_url_error(args['urls'])
# Load model if needed
if loaded_ts != conf['testing']['timestamp'] or loaded_ckpt != conf[
'testing']['ckpt_name']:
load_inference_model(timestamp=conf['testing']['timestamp'],
ckpt_name=conf['testing']['ckpt_name'])
conf = config.conf_dict
# Make the predictions
with graph.as_default():
pred_lab, pred_prob = test_utils.predict(
model=model,
X=args['urls'],
conf=conf,
top_K=top_K,
filemode='url',
merge=merge,
use_multiprocessing=False
) # safer to avoid memory fragmentation in failed queries
if merge:
pred_lab, pred_prob = np.squeeze(pred_lab), np.squeeze(pred_prob)
return format_prediction(pred_lab, pred_prob)
def predict_data(args, merge=True):
"""
Function to predict an image in binary format
"""
#####FIXME: Remove after DEEPaaS upgrade
if type(args['files']) is not list:
args['files'] = [args['files']]
########################################
catch_localfile_error(args['files'])
if not loaded:
load_inference_model()
filenames = []
images = [f.read() for f in args['files']]
for image in images:
f = tempfile.NamedTemporaryFile(delete=False)
f.write(image)
f.close()
filenames.append(f.name)
try:
with graph.as_default():
pred_lab, pred_prob = predict(
model=model,
X=filenames,
conf=conf,
top_K=top_K,
filemode='local',
merge=merge,
use_multiprocessing=False
) # safer to avoid memory fragmentation in failed queries
except Exception as e:
raise e
finally:
for f in filenames:
os.remove(f)
if merge:
pred_lab, pred_prob = np.squeeze(pred_lab), np.squeeze(pred_prob)
return format_prediction(pred_lab, pred_prob)
def predict_data(args):
"""
Function to predict an image in binary format
"""
# Check user configuration
update_with_query_conf(args)
conf = config.conf_dict
merge = True
catch_localfile_error(args['files'])
# Load model if needed
if loaded_ts != conf['testing']['timestamp'] or loaded_ckpt != conf[
'testing']['ckpt_name']:
load_inference_model(timestamp=conf['testing']['timestamp'],
ckpt_name=conf['testing']['ckpt_name'])
conf = config.conf_dict
# Create a list with the path to the images
filenames = [f.filename for f in args['files']]
# Make the predictions
try:
with graph.as_default():
pred_lab, pred_prob = test_utils.predict(
model=model,
X=filenames,
conf=conf,
top_K=top_K,
filemode='local',
merge=merge,
use_multiprocessing=False
) # safer to avoid memory fragmentation in failed queries
finally:
for f in filenames:
os.remove(f)
if merge:
pred_lab, pred_prob = np.squeeze(pred_lab), np.squeeze(pred_prob)
return format_prediction(pred_lab, pred_prob)
def predict_url(args, merge=True):
"""
Function to predict an url
"""
catch_url_error(args['urls'])
if not loaded:
load_inference_model()
with graph.as_default():
pred_lab, pred_prob = predict(
model=model,
X=args['urls'],
conf=conf,
top_K=top_K,
filemode='url',
merge=merge,
use_multiprocessing=False
) # safer to avoid memory fragmentation in failed queries
if merge:
pred_lab, pred_prob = np.squeeze(pred_lab), np.squeeze(pred_prob)
return format_prediction(pred_lab, pred_prob)
print("++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++después")
# INCISO: Ahora la parte que continúa está basada en el predicting a datasplit txt file que incluye Ignacio en el notebook
# 3.0 . Esta preparación previa es necesaria para computar la matriz de confusión.
#
# OJO: ahora lo que le vas a dar para testear el modelo dado por el timestamp SÍ se encuentra en data/dataset_files
# Y ES CON LO QUE TÚ QUIERES TESTEAR EL MODELO.
SPLIT_NAME = input("Indica el nombre del split con el que evaluas. Es de data/dataset_files. Ejemplos: val train ...: ")
# Load the data
X, y = load_data_splits(splits_dir=paths.get_ts_splits_dir(),
im_dir=conf['general']['images_directory'],
split_name=SPLIT_NAME)
# Predict
# Añade esto si quieres no usar aumentacion en la validacion:
#
print(conf['augmentation']['val_mode'])
pred_lab, pred_prob = predict(model, X, conf, top_K=TOP_K, filemode='local')
#Ahora guardamos las predicciones
# Save the predictions
pred_dict = {'filenames': list(X),
'pred_lab': pred_lab.tolist(),
'pred_prob': pred_prob.tolist()}
if y is not None:
pred_dict['true_lab'] = y.tolist()
# No incluimos la parte de guardarlo en json porque lo vamos a utilizar ahora mismo.
# Importamos el warning este porque Ignacio lo sugiere.
import warnings
warnings.filterwarnings("ignore") # To ignore UndefinedMetricWarning: [Recall/Precision/F-Score] is ill-defined and being set to 0.0 in labels with no [true/predicted] samples.
# INCISO: Sacamos por pantalla distintas métricas relevantes SOBRE EL SPLIT SELECCIONADO