class AIController:
#TODO: relay routings if AIController is on a different machine
def __init__(self, config, app, verbose_start=False):
self.config = config
self.app = app
if verbose_start:
print('AIController'.ljust(18), end='')
try:
self.middleware = AIMiddleware(config)
self.login_check = None
self._initBottle()
except Exception as e:
if verbose_start:
LogDecorator.print_status('fail')
raise Exception(
f'Could not launch AIController (message: "{str(e)}").')
if verbose_start:
LogDecorator.print_status('ok')
def loginCheck(self,
project=None,
admin=False,
superuser=False,
canCreateProjects=False,
extend_session=False):
return self.login_check(project, admin, superuser, canCreateProjects,
extend_session)
def addLoginCheckFun(self, loginCheckFun):
self.login_check = loginCheckFun
def _initBottle(self):
@self.app.get('/<project>/listModelStates')
def list_model_states(project):
'''
Returns a list of saved AI model states'
metadata for a given project.
Also checks whether model states have been
shared through the model marketplace.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
return {'modelStates': self.middleware.listModelStates(project)}
#TODO: deprecated; replace with workflow:
@self.app.post('/<project>/startTraining')
def start_training(project):
'''
Manually requests the AIController to train the model.
This still only works if there is no training process ongoing.
Otherwise the request is aborted.
'''
if self.loginCheck(project=project, admin=True):
try:
params = request.json
if 'minNumAnnoPerImage' in params:
minNumAnnoPerImage = int(params['minNumAnnoPerImage'])
else:
minNumAnnoPerImage = 0 #TODO
if 'maxNum_train' in params:
maxNumImages_train = int(params['maxNum_train'])
else:
maxNumImages_train = -1 #TODO
status = self.middleware.start_training(
project=project,
minTimestamp='lastState',
minNumAnnoPerImage=minNumAnnoPerImage,
maxNumImages=maxNumImages_train,
maxNumWorkers=1)
except Exception as e:
status = str(e)
return {'status': status}
else:
abort(401, 'unauthorized')
#TODO: deprecated; replace with workflow:
@self.app.post('/<project>/startInference')
def start_inference(project):
'''
Manually requests the AIController to issue an inference job.
'''
if self.loginCheck(project=project, admin=True):
try:
params = request.json
if 'maxNum_inference' in params:
maxNumImages_inference = int(
params['maxNum_inference'])
else:
maxNumImages_inference = -1 #TODO
status = self.middleware.start_inference(
project=project,
forceUnlabeled=False, #TODO
maxNumImages=maxNumImages_inference,
maxNumWorkers=-1) #TODO
except Exception as e:
status = str(e)
return {'status': status}
else:
abort(401, 'unauthorized')
#TODO: deprecated; replace with workflow:
@self.app.post('/<project>/start')
def start_model(project):
'''
Manually launches one of the model processes (train, inference, both, etc.),
depending on the provided flags.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
params = request.json
doTrain = 'train' in params and params['train'] is True
doInference = 'inference' in params and params[
'inference'] is True
if 'minNumAnnoPerImage' in params:
minNumAnnoPerImage = int(params['minNumAnnoPerImage'])
else:
minNumAnnoPerImage = 0 #TODO
if 'maxNum_train' in params:
maxNumImages_train = int(params['maxNum_train'])
else:
maxNumImages_train = -1 #TODO
if 'maxNum_inference' in params:
maxNumImages_inference = int(params['maxNum_inference'])
else:
maxNumImages_inference = -1 #TODO
if doTrain:
if doInference:
status = self.middleware.start_train_and_inference(
project=project,
minTimestamp='lastState',
minNumAnnoPerImage=minNumAnnoPerImage,
maxNumWorkers_train=1, #TODO
forceUnlabeled_inference=False,
maxNumImages_inference=maxNumImages_inference,
maxNumWorkers_inference=-1) #TODO
else:
#TODO: expand to other tasks and requests
if self.middleware.task_ongoing(
project,
('AIController.start_training',
'AIWorker.call_train',
'AIWorker.call_average_model_states')):
raise Exception(
'A training process is already ongoing for project "{}".'
.format(project))
status = self.middleware.start_training(
project=project,
numEpochs=1,
minTimestamp='lastState',
minNumAnnoPerImage=minNumAnnoPerImage,
maxNumImages=maxNumImages_train,
maxNumWorkers=1) #TODO
else:
status = self.middleware.start_inference(
project=project,
forceUnlabeled=False,
maxNumImages=maxNumImages_inference,
maxNumWorkers=-1) #TODO
return {'status': status}
except Exception as e:
abort(400, 'bad request')
@self.app.post('/<project>/launchWorkflow')
def launch_workflow(project):
'''
New way of submitting jobs. This starts entire workflows, which
can be a chain of multiple training and inference jobs in a row.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
params = request.json
result = self.middleware.launch_task(project,
params['workflow'],
username)
return result
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.post('/<project>/abortWorkflow')
def abort_workflow(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
params = request.json
taskID = params['taskID']
self.middleware.revoke_task(project, taskID, username)
return {'status': 0}
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.get('/<project>/status')
def check_status(project):
'''
Queries the middleware for any ongoing training worker processes
and returns the status of each in a dict.
'''
if self.loginCheck(project=project):
try:
queryProject = 'project' in request.query
queryTasks = 'tasks' in request.query
queryWorkers = 'workers' in request.query
status = self.middleware.check_status(
project, queryProject, queryTasks, queryWorkers)
except Exception as e:
status = str(e)
return {'status': status}
else:
abort(401, 'unauthorized')
#TODO: REPLACED WITH GENERIC FN OF ProjectAdministration
# @self.app.get('/<project>/getAImodelSettings')
# def get_ai_model_info(project):
# '''
# Returns the model class and settings for the AI model
# and the AL criterion.
# '''
# if not self.login_check(project=project, admin=True):
# abort(401, 'unauthorized')
# return { 'settings': self.middleware.getProjectModelSettings(project) }
@self.app.get('/<project>/getSavedWorkflows')
def get_saved_workflows(project):
'''
Returns all the model workflows saved for this project,
also made by other users.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
workflows = self.middleware.getSavedWorkflows(project)
return {'workflows': workflows}
except Exception as e:
return {'status': str(e)}
@self.app.post('/<project>/saveWorkflow')
def save_workflow(project):
'''
Receives a workflow definition through JSON, verifies it
by parsing, and stores it in the database if valid. If
the flag "set_default" is given and set to True, the pro-
vided workflow will be set as the default, to be executed
automatically.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
username = html.escape(request.get_cookie('username'))
workflow = request.json['workflow']
workflowName = request.json['workflow_name']
try:
# for updating existing workflows
workflowID = request.json['workflow_id']
except:
workflowID = None
try:
setDefault = request.json['set_default']
except:
setDefault = False
status = self.middleware.saveWorkflow(project, username,
workflow, workflowID,
workflowName, setDefault)
return {'response': status}
except Exception as e:
return {'response': {'status': 1, 'message': str(e)}}
@self.app.post('/<project>/setDefaultWorkflow')
def set_default_workflow(project):
'''
Receives a string (ID) of a workflow and sets it as default
for a given project.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
workflowID = request.json['workflow_id']
status = self.middleware.setDefaultWorkflow(
project, workflowID)
return status
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.post('/<project>/deleteWorkflow')
def delete_workflow(project):
'''
Receives a string (ID) or list of strings (IDs) for work-
flow(s) to be deleted. They can only be deleted by the
authors or else super users.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
username = html.escape(request.get_cookie('username'))
workflowID = request.json['workflow_id']
status = self.middleware.deleteWorkflow(
project, username, workflowID)
return status
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.get('/<project>/getAvailableAImodels')
def get_available_ai_models(project):
'''
Returns all available AI models (class, name) that are
installed in this instance of AIDE and compatible with
the project's annotation and prediction types.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
return self.middleware.getAvailableAImodels(project)
@self.app.post('/<project>/verifyAImodelOptions')
def verify_model_options(project):
'''
Receives JSON-encoded options and verifies their
correctness with the AI model (either specified through
the JSON arguments, or taken from the default project
option). If the AI model does not support verification
(as is the case in legacy models), a warning is returned.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
modelOptions = request.json['options']
try:
modelLibrary = request.json['ai_model_library']
except:
modelLibrary = None
status = self.middleware.verifyAImodelOptions(
project, modelOptions, modelLibrary)
return {'status': status}
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.post('/<project>/saveAImodelSettings')
def save_model_settings(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
settings = request.json['settings']
response = self.middleware.updateAImodelSettings(
project, settings)
return {'status': 0, 'message': response}
except Exception as e:
return {'status': 1, 'message': str(e)}
class AIController:
#TODO: relay routings if AIController is on a different machine
def __init__(self, config, app, dbConnector, taskCoordinator, verbose_start=False, passive_mode=False):
self.config = config
self.app = app
if verbose_start:
print('AIController'.ljust(LogDecorator.get_ljust_offset()), end='')
try:
self.middleware = AIMiddleware(config, dbConnector, taskCoordinator, passive_mode)
self.login_check = None
self._initBottle()
except Exception as e:
if verbose_start:
LogDecorator.print_status('fail')
raise Exception(f'Could not launch AIController (message: "{str(e)}").')
if verbose_start:
LogDecorator.print_status('ok')
def loginCheck(self, project=None, admin=False, superuser=False, canCreateProjects=False, extend_session=False):
return self.login_check(project, admin, superuser, canCreateProjects, extend_session)
def addLoginCheckFun(self, loginCheckFun):
self.login_check = loginCheckFun
def _initBottle(self):
@self.app.get('/<project>/listModelStates')
def list_model_states(project):
'''
Returns a list of saved AI model states'
metadata for a given project.
Also checks whether model states have been
shared through the model marketplace.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
latestOnly = bool(request.params.get('latest_only'))
except:
latestOnly = False
return {'modelStates': self.middleware.listModelStates(project, latestOnly) }
@self.app.post('/<project>/deleteModelStates')
def delete_model_states(project):
'''
Receives a list of model IDs and launches a background
task to delete them.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
modelStateIDs = request.json['model_ids']
taskID = self.middleware.deleteModelStates(project, username, modelStateIDs)
return {
'status': 0,
'task_id': taskID
}
except Exception as e:
return {
'status': 1,
'message': str(e)
}
@self.app.post('/<project>/duplicateModelState')
def duplicate_model_state(project):
'''
Receives a model state ID and creates a copy of it in this project.
This copy receives the current date, which makes it the most recent
model state.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
modelStateID = request.json['model_id']
taskID = self.middleware.duplicateModelState(project, username, modelStateID, skipIfLatest=True)
return {
'status': 0,
'task_id': taskID
}
except Exception as e:
return {
'status': 1,
'message': str(e)
}
@self.app.get('/<project>/getModelTrainingStatistics')
def get_model_training_statistics(project):
'''
Launches a background task to assemble training
stats as (optionally) returned by the model across
all saved model states.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
#TODO: permit filtering for model state IDs (change to POST?)
taskID = self.middleware.getModelTrainingStatistics(project, username, modelStateIDs=None)
return {
'status': 0,
'task_id': taskID
}
except Exception as e:
return {
'status': 1,
'message': str(e)
}
@self.app.post('/<project>/launchWorkflow')
def launch_workflow(project):
'''
New way of submitting jobs. This starts entire workflows, which
can be a chain of multiple training and inference jobs in a row.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
params = request.json
result = self.middleware.launch_task(project, params['workflow'], username)
return result
except Exception as e:
return { 'status': 1,
'message': str(e) }
@self.app.post('/<project>/abortWorkflow')
def abort_workflow(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
params = request.json
taskID = params['taskID']
self.middleware.revoke_task(project, taskID, username)
return { 'status': 0 }
except Exception as e:
return { 'status': 1,
'message': str(e) }
@self.app.post('/<project>/abortAllWorkflows')
def abort_all_workflows(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'forbidden')
try:
username = html.escape(request.get_cookie('username'))
self.middleware.revoke_all_tasks(project, username)
return { 'status': 0 }
except Exception as e:
return { 'status': 1,
'message': str(e) }
@self.app.get('/<project>/status')
def check_status(project):
'''
Queries the middleware for any ongoing training worker processes
and returns the status of each in a dict.
'''
if self.loginCheck(project=project):
try:
queryProject = 'project' in request.query
queryTasks = 'tasks' in request.query
queryWorkers = 'workers' in request.query
nudgeWatchdog = 'nudge_watchdog' in request.query
recheckAutotrainSettings = 'recheck_autotrain_settings' in request.query
status = self.middleware.check_status(
project,
queryProject, queryTasks, queryWorkers, nudgeWatchdog, recheckAutotrainSettings)
except Exception as e:
status = str(e)
return { 'status' : status }
else:
abort(401, 'unauthorized')
@self.app.get('/<project>/getSavedWorkflows')
def get_saved_workflows(project):
'''
Returns all the model workflows saved for this project,
also made by other users.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
workflows = self.middleware.getSavedWorkflows(project)
return { 'workflows': workflows }
except Exception as e:
return { 'status': str(e) }
@self.app.post('/<project>/saveWorkflow')
def save_workflow(project):
'''
Receives a workflow definition through JSON, verifies it
by parsing, and stores it in the database if valid. If
the flag "set_default" is given and set to True, the pro-
vided workflow will be set as the default, to be executed
automatically.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
username = html.escape(request.get_cookie('username'))
workflow = request.json['workflow']
workflowName = request.json['workflow_name']
try:
# for updating existing workflows
workflowID = request.json['workflow_id']
except:
workflowID = None
try:
setDefault = request.json['set_default']
except:
setDefault = False
status = self.middleware.saveWorkflow(project, username, workflow, workflowID, workflowName, setDefault)
return { 'response': status }
except Exception as e:
return { 'response': {'status':1, 'message':str(e)} }
@self.app.post('/<project>/setDefaultWorkflow')
def set_default_workflow(project):
'''
Receives a string (ID) of a workflow and sets it as default
for a given project.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
workflowID = request.json['workflow_id']
status = self.middleware.setDefaultWorkflow(project, workflowID)
return status
except Exception as e:
return {'status':1, 'message':str(e)}
@self.app.post('/<project>/deleteWorkflow')
def delete_workflow(project):
'''
Receives a string (ID) or list of strings (IDs) for work-
flow(s) to be deleted. They can only be deleted by the
authors or else super users.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
username = html.escape(request.get_cookie('username'))
workflowID = request.json['workflow_id']
status = self.middleware.deleteWorkflow(project, username, workflowID)
return status
except Exception as e:
return {'status':1, 'message':str(e)}
@self.app.post('/<project>/deleteWorkflowHistory')
def delete_workflow_history(project):
'''
Receives a string (ID) or list of strings (IDs) for work-
flow(s) to be deleted. They can only be deleted by the
authors or else super users.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
workflowID = request.json['workflow_id']
revokeRunning = (request.json['revoke_running'] if 'revoke_running' in request.json else False)
status = self.middleware.deleteWorkflow_history(project, workflowID, revokeRunning)
return status
except Exception as e:
return {'status':1, 'message':str(e)}
@self.app.get('/<project>/getAImodelTrainingInfo')
def get_ai_model_training_info(project):
'''
Returns information required to determine whether AI models can be trained
for a given project.
This includes:
- Whether an AI model library is configured for the project
- Whether at least consumer for each AIController and AIWorker is
connected and available
Returns a dict of this information accordingly.
'''
if not self.loginCheck(project, admin=True):
abort(401, 'unauthorized')
try:
status = self.middleware.get_ai_model_training_info(project)
return {'response': status}
except Exception as e:
return {'status':1, 'message':str(e)}
@self.app.get('/<project>/getAvailableAImodels')
def get_available_ai_models(project):
'''
Returns all available AI models (class, name) that are
installed in this instance of AIDE and compatible with
the project's annotation and prediction types.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
return self.middleware.getAvailableAImodels(project)
@self.app.get('/getAvailableAImodels')
def get_available_ai_models():
'''
Returns all available AI models (class, name) that are
installed in this instance of AIDE.
'''
if not self.loginCheck(canCreateProjects=True):
abort(401, 'unauthorized')
return self.middleware.getAvailableAImodels(None)
@self.app.post('/<project>/verifyAImodelOptions')
def verify_model_options(project):
'''
Receives JSON-encoded options and verifies their
correctness with the AI model (either specified through
the JSON arguments, or taken from the default project
option). If the AI model does not support verification
(as is the case in legacy models), a warning is returned.
'''
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
modelOptions = request.json['options']
try:
modelLibrary = request.json['ai_model_library']
except:
modelLibrary = None
status = self.middleware.verifyAImodelOptions(project, modelOptions, modelLibrary)
return {'status': status}
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.post('/<project>/saveAImodelSettings')
def save_model_settings(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
settings = request.json['settings']
response = self.middleware.updateAImodelSettings(project, settings)
return {'status': 0, 'message': response}
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.get('/<project>/getLabelclassAutoadaptInfo')
def get_labelclass_autoadapt_info(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
try:
modelID = request.params.get('model_id')
except:
modelID = None
response = self.middleware.getLabelclassAutoadaptInfo(project, modelID)
return {'status': 0, 'message': response}
except Exception as e:
return {'status': 1, 'message': str(e)}
@self.app.post('/<project>/saveLabelclassAutoadaptInfo')
def save_labelclass_autoadapt_info(project):
if not self.loginCheck(project=project, admin=True):
abort(401, 'unauthorized')
try:
enabled = parse_boolean(request.params.get('enabled'))
response = self.middleware.setLabelclassAutoadaptEnabled(project, enabled)
return {'status': 0, 'message': response}
except Exception as e:
return {'status': 1, 'message': str(e)}