def __init__(self, model, batch_size, hparam_str, session):
"""Create the model.
Args:
model: (BasicModel) The model we are going to train.
batch_size: the size of the batches generated in each iteration.
hparam_str: the hyper-parameters for the input layer.
session: the current tensorflow Session (used for online learning).
"""
self.hparams = ultra.utils.hparams.HParams(
click_model_json='./example/ClickModel/pbm_0.1_1.0_4_1.0.json', # the setting file for the predefined click models.
tau=1, # Scalar for the probability distribution.
oracle_mode=False, # Set True to feed relevance labels instead of simulated clicks.
dynamic_bias_eta_change=0.0, # Set eta change step for dynamic bias severity in training, 0.0 means no change.
dynamic_bias_step_interval=1000, # Set how many steps to change eta for dynamic bias severity in training, 0.0 means no change.
)
print('Create online simluation feed')
print(hparam_str)
self.hparams.parse(hparam_str)
self.click_model = None
with open(self.hparams.click_model_json) as fin:
model_desc = json.load(fin)
self.click_model = cm.loadModelFromJson(model_desc)
self.start_index = 0
self.count = 1
self.rank_list_size = model.rank_list_size
self.max_candidate_num = model.max_candidate_num
self.feature_size = model.feature_size
self.batch_size = batch_size
self.model = model
self.session = session
self.global_batch_count = 0
def loadEstimatorFromFile(self, file_name):
"""Load a propensity estimator from a json file.
Args:
file_name: (string) The path to the json file of the propensity estimator.
"""
with open(file_name) as data_file:
data = json.load(data_file)
self.click_model = CM.loadModelFromJson(data['click_model'])
return
def main():
click_model_json_file = sys.argv[1]
data_dir = sys.argv[2]
output_path = sys.argv[3]
print("Load data from " + data_dir)
train_set = data_utils.read_data(data_dir, 'train')
click_model = None
with open(click_model_json_file) as fin:
model_desc = json.load(fin)
click_model = CM.loadModelFromJson(model_desc)
print("Estimating...")
estimator = RandomizedPropensityEstimator()
estimator.estimateParametersFromModel(click_model, train_set)
print("Output results...")
output_file = output_path + '/randomized_' + click_model_json_file.split(
'/')[-1][:-5] + '.json'
estimator.outputEstimatorToFile(output_file)
def __init__(self, model, batch_size, hparam_str, session=None):
"""Create the model.
Args:
model: (BasicModel) The model we are going to train.
batch_size: the size of the batches generated in each iteration.
hparam_str: the hyper-parameters for the input layer.
"""
self.hparams = ultra.utils.hparams.HParams(
# the setting file for the predefined click models.
# click_model_json='./example/ClickModel/pbm_0.1_1.0_4_1.0.json',
click_model_json='./ClickModel/pbm_0.1_1.0_4_1.0.json',
# Set True to feed relevance labels instead of simulated clicks.
oracle_mode=False,
# Set eta change step for dynamic bias severity in training, 0.0
# means no change.
dynamic_bias_eta_change=0.0,
# Set how many steps to change eta for dynamic bias severity in
# training, 0.0 means no change.
dynamic_bias_step_interval=1000,
)
print('Create simluated clicks feed')
print(hparam_str)
self.hparams.parse(hparam_str)
self.click_model = None
if not self.hparams.oracle_mode:
with open(self.hparams.click_model_json) as fin:
model_desc = json.load(fin)
self.click_model = cm.loadModelFromJson(model_desc)
self.start_index = 0
self.count = 1
self.rank_list_size = model.rank_list_size
self.feature_size = model.feature_size
self.batch_size = batch_size
self.model = model
self.global_batch_count = 0
