def train(user_conf):
"""
Parameters
----------
user_conf : dict
Json dict (created with json.dumps) with the user's configuration parameters that will replace the defaults.
Must be loaded with json.loads()
For example:
user_conf={'num_classes': 'null', 'lr_step_decay': '0.1', 'lr_step_schedule': '[0.7, 0.9]', 'use_early_stopping': 'false'}
"""
CONF = config.CONF
# Update the conf with the user input
for group, val in sorted(CONF.items()):
for g_key, g_val in sorted(val.items()):
g_val['value'] = json.loads(user_conf[g_key])
# Check the configuration
try:
config.check_conf(conf=CONF)
except Exception as e:
raise BadRequest(e)
CONF = config.conf_dict(conf=CONF)
timestamp = datetime.now().strftime('%Y-%m-%d_%H%M%S')
config.print_conf_table(CONF)
K.clear_session() # remove the model loaded for prediction
train_fn(TIMESTAMP=timestamp, CONF=CONF)
# Sync with NextCloud folders (if NextCloud is available)
try:
mount_nextcloud(paths.get_models_dir(), 'ncplants:/models')
except Exception as e:
print(e)
def train(**args):
"""
Train an image classifier
"""
update_with_query_conf(user_args=args)
CONF = config.conf_dict
timestamp = datetime.now().strftime('%Y-%m-%d_%H%M%S')
config.print_conf_table(CONF)
K.clear_session() # remove the model loaded for prediction
train_fn(TIMESTAMP=timestamp, CONF=CONF)
# Sync with NextCloud folders (if NextCloud is available)
try:
mount_nextcloud(paths.get_models_dir(), 'rshare:/models')
except Exception as e:
print(e)
from imgclas import paths, utils, config, test_utils
from imgclas.data_utils import load_class_names, load_class_info, mount_nextcloud
from imgclas.train_runfile import train_fn
# TODO: Move to proper marshalling for arguments
# The point is that some fields need additional information than the one that is contained in the config.yaml
# --> define an example for each arg in the config.yaml --> create the schema for that arg
# type_map = {'int': fields.Int, 'str': fields.Str, 'float': fields.Float, 'dict': fields.Dict,
# 'bool': fields.Bool, 'list': fields.List}
# field_type = type_map.get(g_val['type'], fields.Field)
# parser[g_key] = field_type(**opt_args)
# --> another option is to add a marshmallow schema to each config args
# Mount NextCloud folders (if NextCloud is available)
try:
mount_nextcloud('rshare:/data/dataset_files', paths.get_splits_dir())
mount_nextcloud('rshare:/data/images', paths.get_images_dir())
#mount_nextcloud('rshare:/models', paths.get_models_dir())
except Exception as e:
print(e)
# Empty model variables for inference (will be loaded the first time we perform inference)
loaded_ts, loaded_ckpt = None, None
graph, model, conf, class_names, class_info = None, None, None, None, None
# Additional parameters
allowed_extensions = set(['png', 'jpg', 'jpeg', 'PNG', 'JPG',
'JPEG']) # allow only certain file extensions
top_K = 5 # number of top classes predictions to return
import numpy as np
import requests
from werkzeug.exceptions import BadRequest
import tensorflow as tf
from tensorflow.keras.models import load_model
from tensorflow.keras import backend as K
from imgclas import paths, utils, config
from imgclas.data_utils import load_class_names, load_class_info, mount_nextcloud
from imgclas.test_utils import predict
from imgclas.train_runfile import train_fn
# Mount NextCloud folders (if NextCloud is available)
try:
mount_nextcloud('ncplants:/data/dataset_files', paths.get_splits_dir())
mount_nextcloud('ncplants:/data/images', paths.get_images_dir())
#mount_nextcloud('ncplants:/models', paths.get_models_dir())
except Exception as e:
print(e)
# Empty model variables for inference (will be loaded the first time we perform inference)
loaded = False
graph, model, conf, class_names, class_info = None, None, None, None, None
# Additional parameters
allowed_extensions = set(['png', 'jpg', 'jpeg', 'PNG', 'JPG',
'JPEG']) # allow only certain file extensions
top_K = 5 # number of top classes predictions to return