def __init__(self,
rec_class=None,
api_path=None,
extra_payload=dict(),
user='******',
rec_score_file='rec_state.obj',
verbose=True,
warm_start=False,
n_recs=1,
datasets=False,
use_knowledgebase=False,
term_condition='n_recs',
max_time=5):
"""Initializes AI managing agent."""
# default supervised learning recommender settings
self.DEFAULT_REC_CLASS = RandomRecommender
self.DEFAULT_REC_ARGS = {'metric': 'accuracy'}
# recommendation engines for different problem types
# will be expanded as more types of probles are supported
# (classification, regression, unsupervised, etc.)
self.rec_engines = {"classification": None}
# Request manager settings
self.n_recs = n_recs if n_recs > 0 else 1
self.continuous = n_recs < 1
# api parameters, will be removed from self once the recommenders no longer
# call the api directly.
# See #98 <https://github.com/EpistasisLab/pennai/issues/98>
if api_path == None:
api_path = ('http://' + os.environ['LAB_HOST'] + ':' +
os.environ['LAB_PORT'])
self.user = user
self.api_path = api_path
self.api_key = os.environ['APIKEY']
self.verbose = verbose #False: no printouts, True: printouts on updates
# file name of stored scores for the recommender
self.rec_score_file = rec_score_file
# timestamp of the last time new experiments were processed
self.last_update = 0
# api
self.labApi = api_utils.LabApi(api_path=self.api_path,
user=self.user,
api_key=self.api_key,
extra_payload=extra_payload,
verbose=self.verbose)
self.load_options() #loads algorithm parameters to self.ui_options
self.initilize_recommenders(rec_class) # set self.rec_engines
# build dictionary of ml ids to names conversion
self.ml_id_to_name = self.labApi.get_ml_id_dict()
# print('ml_id_to_name:',self.ml_id_to_name)
# dictionary of dataset threads, initilized and used by q_utils.
# Keys are datasetIds, values are q_utils.DatasetThread instances.
#WGL: this should get moved to the request manager
self.dataset_threads = {}
# local dataframe of datasets and their metafeatures
self.dataset_mf_cache = pd.DataFrame()
# store dataset_id to hash dictionary
self.dataset_mf_cache_id_hash_lookup = {}
if use_knowledgebase:
self.load_knowledgebase()
# set termination condition
self.term_condition = term_condition
if self.term_condition == 'n_recs':
self.term_value = self.n_recs
elif self.term_condition == 'time':
self.term_value = max_time
else:
self.term_value = None
# start the request manager
self.requestManager = RequestManager(
ai=self,
defaultTermConditionStr=self.term_condition,
defaultTermParam=self.term_value)
# if there is a pickle file, load it as the recommender scores
assert not (warm_start), "The `warm_start` option is not yet supported"
# for comma-separated list of datasets in datasets, turn AI request on
assert not (
datasets
), "The `datasets` option is not yet supported: " + str(datasets)
def __init__(self,
rec=None,
api_path=None,
extra_payload=dict(),
user='******',
rec_score_file='rec_state.obj',
verbose=True,
warm_start=False,
n_recs=1,
datasets=False,
use_knowledgebase=False,
term_condition='n_recs',
max_time=5):
"""initializes AI managing agent."""
# recommender settings
if api_path == None:
api_path = ('http://' + os.environ['LAB_HOST'] + ':' +
os.environ['LAB_PORT'])
self.rec = rec
self.n_recs = n_recs if n_recs > 0 else 1
self.continous = n_recs < 1
# api parameters, will be removed from self once the recommenders no longer
# call the api directly.
# See #98 <https://github.com/EpistasisLab/pennai/issues/98>
self.user = user
self.api_path = api_path
self.api_key = os.environ['APIKEY']
self.verbose = verbose #False: no printouts, True: printouts on updates
# file name of stored scores for the recommender
self.rec_score_file = rec_score_file
# timestamp of the last time new experiments were processed
self.last_update = 0
# api
self.labApi = api_utils.LabApi(api_path=self.api_path,
user=self.user,
api_key=self.api_key,
extra_payload=extra_payload,
verbose=self.verbose)
self.load_options() #loads algorithm parameters to self.ui_options
# create a default recommender if not set
if (rec):
self.rec = rec
else:
self.rec = RandomRecommender(ml_p=self.labApi.get_all_ml_p())
# set the registered ml parameters in the recommender
ml_p = self.labApi.get_all_ml_p()
assert ml_p is not None
assert len(ml_p) > 0
self.rec.ml_p = ml_p
# if hasattr(self.rec,'mlp_combos'):
# self.rec.mlp_combos = self.rec.ml_p['algorithm']+'|'+self.rec.ml_p['parameters']
logger.debug('self.rec.ml_p:\n' + str(self.rec.ml_p.head()))
# tmp = self.labApi.get_all_ml_p()
# tmp.to_csv('ml_p_options.csv')
# build dictionary of ml ids to names conversion
self.ml_id_to_name = self.labApi.get_ml_id_dict()
# print('ml_id_to_name:',self.ml_id_to_name)
# dictionary of dataset threads, initilized and used by q_utils.
# Keys are datasetIds, values are q_utils.DatasetThread instances.
#WGL: this should get moved to the request manager
self.dataset_threads = {}
# local dataframe of datasets and their metafeatures
self.dataset_mf = pd.DataFrame()
if use_knowledgebase:
self.load_knowledgebase()
# set termination condition
self.term_condition = term_condition
if self.term_condition == 'n_recs':
self.term_value = self.n_recs
elif self.term_condition == 'time':
self.term_value = max_time
else:
self.term_value = None
# start the request manager
self.requestManager = RequestManager(
ai=self,
defaultTermConditionStr=self.term_condition,
defaultTermParam=self.term_value)
# if there is a pickle file, load it as the recommender scores
assert not (warm_start), "The `warm_start` option is not yet supported"
# for comma-separated list of datasets in datasets, turn AI request on
assert not (
datasets
), "The `datasets` option is not yet supported: " + str(datasets)