def __init__(self,
model_options,
input_options,
stock_code=None,
load=False,
saved_model_dir=None,
saved_model_path=None):
"""Initializes the model. Creates a new model or loads a saved model."""
Model.__init__(self,
model_options,
input_options,
stock_code=stock_code)
# Please check scipy SVR documentation for details
if not load or saved_model_dir is None:
self.model = [
SVR(kernel=self.model_options["kernel"],
degree=self.model_options["degree"],
gamma=self.model_options["gamma"],
coef0=self.model_options["coef0"],
tol=self.model_options["tol"],
C=self.model_options["C"],
epsilon=self.model_options["epsilon"],
shrinking=self.model_options["shrinking"],
max_iter=self.model_options["max_iter"])
for _ in range(model_options["predict_n"])
]
else:
model_path = saved_model_path if saved_model_path is not None else self.get_saved_model_path(
saved_model_dir)
if model_path is not None:
self.load_model(path.join(saved_model_dir, model_path),
self.SKLEARN_MODEL)
def __init__(self,
dataset_name: str,
hyperparameters: Dict,
infra_s3: Dict,
features: list,
target: str,
h2o_ip: str,
data_dir: str,
training_job_dir: str = None,
clean: bool = False,
model_id: str = None):
Model.__init__(self, dataset_name, hyperparameters, infra_s3, features,
target, data_dir, training_job_dir, clean)
self.ip = h2o_ip.split(':')[0]
self.port = h2o_ip.split(':')[1]
if model_id:
self.model_id = model_id
timestamp = self.model_id[-23:]
self.model_filename = os.path.join(
*(training_job_dir.split('/')[:-1] +
[timestamp, self.MODEL_FILENAME]))
else:
self.model_id = '-'.join(training_job_dir.split('/')[1:])
self.model_filename = os.path.join(training_job_dir,
self.MODEL_FILENAME)
logging.info('======== Model ID ========\n{}'.format(self.model_id))
def __init__(self, gamma_0=None, gamma_f=None, n=None):
"""
Initialise when we create a the model.
"""
# We initialise gamma_0, gamma_f and n if the user entered something.
Model.__init__(self, gamma_0, gamma_f, n)
'''
We define the fixed attribute (that won't change for all the equations of this model).
display : how we show it to the user to ask him to fill the value
var_name : how the variable is written in the .conf file
value : The value of the variable. Set to "" (empty string) by default.
'''
# FILL HERE : 3/4 put the good attributes of your model (specifics variables that the user need to give)
self.fixed_attributes = [{
"display": "a",
"var_name": "a_spec",
"value": ""
}, {
"display": "b",
"var_name": "b_spec",
"value": ""
}, {
"display": "d",
"var_name": "d_spec",
"value": ""
}]
# We run the setup function to ask every attribute to the user.
self.setup()
def __init__(self, model_options, input_options, stock_code):
"""Initializes the model."""
Model.__init__(self,
model_options,
input_options,
stock_code=stock_code)
def __init__(self,
session,
initial_embeddings: Optional[np.ndarray],
static: str='non-static',
input_length=20,
embeddings_dim=50,
learning_rate=1,
num_filters=100,
regularization_rate=0.01,
ckpt_file: Optional[str] = None):
Model.__init__(self)
self.sess = session
self.learning_rate = learning_rate
self.input_length = input_length
assert static in ['non-static', 'static', 'rand', 'both']
self.static = static
self.embeddings_dim = embeddings_dim
if initial_embeddings is None:
assert static == 'rand'
self.initial_embeddings = np.random.rand(VOCABULARY_SIZE, self.embeddings_dim).astype(np.float32)
else:
self.initial_embeddings = initial_embeddings
self.num_filters = num_filters
self.regularization_rate = regularization_rate
if ckpt_file:
self.ckpt_file = ckpt_file
else:
ckpt_dir = os.path.join('tmp', 'models', str(self))
os.makedirs(ckpt_dir, exist_ok=True)
self.ckpt_file = os.path.join(ckpt_dir, 'yoon_kim.ckpt')
self._build_model()
self._add_training_objectives()
self._load_or_init()
def __init__(self, init_avg=1):
Model.__init__(self)
self.VERSION = 1
self.name = "TimeItem-average"
self.model_time = True
self._init_avg = init_avg
def __init__(self,
model_options,
input_options,
stock_code=None,
load=False,
saved_model_dir=None,
saved_model_path=None,
build_model=True):
"""Initializes the model. Creates a new model or loads a saved model."""
Model.__init__(self,
model_options,
input_options,
stock_code=stock_code)
self.input_shape = get_input_shape(input_options)
if not load or saved_model_dir is None and build_model:
self.build_model()
else:
model_path = saved_model_path if saved_model_path is not None else self.get_saved_model_path(
saved_model_dir)
if model_path is not None:
self.load_model(path.join(saved_model_dir, model_path),
Model.KERAS_MODEL)
def __init__(self, feature_set, predict_table_name):
Model.__init__(self, feature_set)
self.predict_table_name = predict_table_name
self.x = None
self.y = None
self.clf = None
self.predict_x = None
self.predict_id = None
self.vectorizer = None
def __init__(self, ph):
Model.__init__(self, ph)
self.input_shape = INPUT_SHAPES[ph['dataset']]
self.output_shape = OUTPUT_SHAPES[self.ph['dataset']]
self.pretrained_model_fp = self.ph.setdefault('pretrained_model_fp',
None)
if self.pretrained_model_fp:
print('training on a pretrained model')
def __init__(self, model_options, load=False, saved_model_dir=None, saved_model_path=None):
Model.__init__(self, model_options)
if not load or saved_model_dir is None:
self.model = linear_model.LinearRegression()
else:
model_path = saved_model_path if saved_model_path is not None else self.get_saved_model_path(saved_model_dir)
if model_path is not None:
with open(saved_model_dir + "/" + model_path, "rb") as model_file:
self.model = pickle.load(model_file)
def __init__(self, alpha=1.0, beta=0.1, KC=3.5, KI=2.5):
Model.__init__(self)
self.VERSION = 1
self.name = "Hierarchical"
self._alpha = alpha
self._beta = beta
self._KC = KC
self._KI = KI
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self, alpha=1.0, beta=0.1, KC=1, KI=1):
Model.__init__(self)
self.VERSION = 2
self.name = "Prior-current"
self._alpha = alpha
self._beta = beta
self._KC = KC
self._KI = KI
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self,
tensor,
keep_prob=1.0,
num_classes=1000,
retrain_layer=[],
weights_path='./weights/vgg16.npy'):
# Call the parent class, which will create the graph
Model.__init__(self, tensor, keep_prob, num_classes, retrain_layer,
weights_path)
# Call the create function to build the computational graph
self.final, self.endpoints = self.create()
def __init__(self, alpha=0.8, beta=0.08, KC=0.075, KI=0.1):
Model.__init__(self)
self.VERSION = 4
self.name = "TimeHierarchical"
self.model_time = True
self._alpha = alpha
self._beta = beta
self._KC = KC
self._KI = KI
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self, alpha=1.0, beta=0.1, K=1, init_avg=0, floating_start=True):
Model.__init__(self)
self.VERSION = 3
self.name = "Basic-Time"
self.model_time = True
self._alpha = alpha
self._beta = beta
self._K = K
self._init_avg = init_avg
self._floating_start = floating_start
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self, alpha=1.0, beta=0.1, K=1, concepts=None):
Model.__init__(self)
self.VERSION = 2
self.name = "TimeConcepts"
self.model_time = True
self._alpha = alpha
self._beta = beta
self._K = K
self._concepts = sorted(concepts.keys()) if concepts is not None else "All"
self._init_concept_map(concepts)
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self, alpha=1.0, beta=0.1, KC=1, KI=1, init_avg=0, first_level=3):
Model.__init__(self)
self.VERSION = 4
self.name = "Prior-Current-Time"
self.model_time = True
self._alpha = alpha
self._beta = beta
self._KC = KC
self._KI = KI
self._init_avg = init_avg
self._first_level = first_level
self.decay_function = lambda x: alpha / (1 + beta * x)
def __init__(self,
from_ckpt=False,
n_dim=None,
r=2,
opt_params=default_opt,
log_prefix='./run'):
# perform the usual initialization
self.r = r
Model.__init__(self,
from_ckpt=from_ckpt,
n_dim=n_dim,
r=r,
opt_params=opt_params,
log_prefix=log_prefix)
def __init__(self,
tensor,
keep_prob=1.0,
num_classes=1001,
retrain_layer=[],
weights_path='./weights/resnet_v2_101.ckpt'):
# Call the parent class
Model.__init__(self, tensor, keep_prob, num_classes, retrain_layer,
weights_path)
# TODO This implementation has a problem while validation (is still set to training)
is_training = True if retrain_layer else False
with slim.arg_scope(resnet_arg_scope()):
self.final, self.endpoints = resnet_v2_101(self.tensor,
num_classes=num_classes,
is_training=is_training)
def __init__(self, ph):
Model.__init__(self, ph)
self.input_size = INPUT_SIZE[self.ph['dataset']]
self.output_size = OUTPUT_SIZE[self.ph['dataset']]
self.pretrained_model_fp = self.ph.setdefault('pretrained_model_fp', None)
self.fine_tune = self.ph['fine_tune']
if self.pretrained_model_fp:
print('training on pretrained model')
else:
print('training from scratch without pretrained model')
if self.fine_tune:
print('fine tuning pretrained model')
else:
print('pretrained model weights are frozen')
def __init__(self,
from_ckpt=False,
n_dim=None,
r=2,
pool_size=4,
strides=4,
opt_params=default_opt,
log_prefix='./run'):
# perform the usual initialization
self.r = r
self.pool_size = pool_size
self.strides = strides
Model.__init__(self,
from_ckpt=from_ckpt,
n_dim=n_dim,
r=r,
opt_params=opt_params,
log_prefix=log_prefix)
def __init__(self,
tensor,
keep_prob=1.0,
num_classes=1001,
retrain_layer=[],
weights_path='./weights/resnet_v2_101.ckpt'):
# Call the parent class
Model.__init__(self, tensor, keep_prob, num_classes, retrain_layer,
weights_path)
# Create the Graph
is_training = True if retrain_layer else False
with slim.arg_scope(resnet_arg_scope()):
self.final, self.endpoints = resnet_v2_101(
self.tensor,
num_classes=num_classes,
is_training=is_training,
global_pool=True # True: both height_out and width_out equal one
)
def __init__(self, session, corpus, sampler, k, factor_reg, bias_reg):
self.sampler = sampler # 选择sample方式
self.lfactor_reg = factor_reg # 设置正则率
self.bias_reg = bias_reg # 设置bias正则率
self.K = k # k是latent factor model的维度
#
self.u, self.i, self.j, self.mf_auc, self.bprloss, self.train_op = BPR.bpr_mf(
corpus.user_count,
corpus.item_count,
k,
regulation_rate=factor_reg,
bias_reg=bias_reg)
Model.__init__(self, corpus, session)
print("bpr's restore...")
self.restore()
print("bpr - k=%d, reg_lf: %.2f, reg_bias=%.2f" %
(k, factor_reg, bias_reg))
def __init__(self,
model_options,
input_options,
stock_code=None,
load=False,
saved_model_dir=None,
saved_model_path=None):
"""Initializes the model. Creates a new model or loads a saved model."""
Model.__init__(self,
model_options,
input_options,
stock_code=stock_code)
if not load or saved_model_dir is None:
self.model = linear_model.LinearRegression()
else:
model_path = saved_model_path if saved_model_path is not None else self.get_saved_model_path(
saved_model_dir)
if model_path is not None:
self.load_model(path.join(saved_model_dir, model_path),
self.SKLEARN_MODEL)
def __init__(self, session, corpus, sampler, k, k2, factor_reg, bias_reg):
self.K = k
self.K2 = k2
self.lam = factor_reg
self.bias_reg = bias_reg
self.sampler = sampler
self.u, self.i, self.j, self.iv, self.jv, self.loss, self.auc, self.train_op = VBPR.vbpr(
corpus.user_count,
corpus.item_count,
len(corpus.image_features[1]),
hidden_dim=k,
hidden_img_dim=k2,
l2_regulization=factor_reg,
bias_regulization=bias_reg)
Model.__init__(self, corpus, session)
print("vbpr's restore...")
self.restore()
print("VBPR - K=%d, K2=%d, reg_lf=%.2f, reg_bias=%.2f" %
(k, k2, factor_reg, bias_reg))
def __init__(self, model_options, load=False, saved_model_dir=None):
Model.__init__(self, model_options)
print(saved_model_dir)
# Please check scipy SVR documentation for details
if not load or saved_model_dir is None:
self.model = SVR(kernel=self.model_options["kernel"],
degree=self.model_options["degree"],
gamma=self.model_options["gamma"],
coef0=self.model_options["coef0"],
tol=self.model_options["tol"],
C=self.model_options["C"],
epsilon=self.model_options["epsilon"],
shrinking=self.model_options["shrinking"],
cache_size=self.model_options["cache_size"],
verbose=self.model_options["verbose"],
max_iter=self.model_options["max_iter"])
else:
model_path = self.get_saved_model_path(saved_model_dir)
if model_path is not None:
with open(saved_model_dir + "/" + model_path,
"rb") as model_file:
self.model = pickle.load(model_file)
def __init__(self,
session,
learning_rate=0.01,
input_length=1014,
regularization_rate=0.1,
alphabet_size=70,
ckpt_file: Optional[str] = None):
Model.__init__(self)
self.sess = session
self.learning_rate = learning_rate
self.input_length = input_length
self.alphabet_size = alphabet_size
self.reg_rate = regularization_rate
if ckpt_file:
self.ckpt_file = ckpt_file
else:
ckpt_dir = os.path.join('tmp', 'models', str(self))
os.makedirs(ckpt_dir, exist_ok=True)
self.ckpt_file = os.path.join(ckpt_dir, 'char_cnn.ckpt')
self._build_model()
self._add_training_objectives()
self._load_or_init()
def __init__(self,
dataset_name: str,
hyperparameters: Dict,
infra_s3: Dict,
infra_sm: Dict,
features: List,
target: str,
data_dir: str,
training_job_common_dir: str,
training_job_dir: str,
aws_model_id: str = None,
clean: bool = False):
Model.__init__(self, dataset_name, hyperparameters, infra_s3, features,
target, data_dir, training_job_dir, clean)
self.infra_sm = infra_sm
self.aws_model_id = aws_model_id
if not aws_model_id:
# data-config-timestamp
self.aws_model_id = '-'.join(
training_job_dir.replace('_', '').split('/')[1:])
s3_timestamp = training_job_dir.split('/')[-1]
else:
s3_timestamp = self.aws_model_id[-23:]
logging.info('======== Model ID ========\n{}'.format(
self.aws_model_id))
self.training_job_common_dir = training_job_common_dir
# data/ml/<dataset>/<config>
self.s3_ml_data_common_dir = '/'.join(
[self.infra_s3['s3_folder_ml']] +
training_job_dir.split('/')[1:-1])
# A new local folder is created for the logs and plots to avoid erasing previous data
# However, we need to use the S3 folder previously generated by the prediction step (e.g. if we run only eval)
# data/ml/<dataset>/<config>/<timestamp>
self.s3_ml_data_job_dir = '/'.join(
[self.s3_ml_data_common_dir, s3_timestamp])
self.csv_training_filename = 'training.csv'
self.csv_validation_filename = 'validation.csv'
self.csv_testing_filename = 'testing.csv'
self.s3_training_csv_path = self.s3_filepath(
filetype='ml_data',
filename=self.csv_training_filename,
common=True,
uri=True)
self.s3_validation_csv_path = self.s3_filepath(
filetype='ml_data',
filename=self.csv_validation_filename,
common=True,
uri=True)
self.s3_testing_csv_path = self.s3_filepath(
filetype='ml_data',
filename=self.csv_testing_filename,
common=True,
uri=True)
self.local_training_preds_file = self.local_filepath(
'training_preds.csv')
self.local_validation_preds_file = self.local_filepath(
'validation_preds.csv')
self.local_testing_preds_file = self.local_filepath(
'testing_preds.csv')
self.s3_preds_folder = self.s3_filepath(filetype='ml_data', uri=True)
self.s3_training_preds_file = self.s3_filepath(
filetype='ml_data', filename='training.csv.out')
self.s3_validation_preds_file = self.s3_filepath(
filetype='ml_data', filename='validation.csv.out')
self.s3_testing_preds_file = self.s3_filepath(
filetype='ml_data', filename='testing.csv.out')
boto3_sess = boto3.Session()
self.boto3_sm = boto3_sess.client('sagemaker')
self.container = self._get_container(boto3_sess)
def __init__(self, ph):
Model.__init__(self, ph)
self.input_size = INPUT_SIZE[self.ph['dataset']]
def __init__(self, prediction_model, time_model):
Model.__init__(self)
self._prediction_model = prediction_model
self._time_model = time_model
def __init__(self, ph):
Model.__init__(self, ph)
def __init__(self, ph):
Model.__init__(self, ph)
self.pretrained_model_fp = None
if 'pretrained_model_fp' in self.ph:
self.pretrained_model_fp = self.ph['pretrained_model_fp']
def __init__(self, pre_word, post_word):
Model.__init__(self, pre_word, post_word)
# root = {"word_A": [sub-tree, count], "word_B": [sub-tree, count],...}
self.root = {}
self.sizemo = None
def __init__(self, env):
Model.__init__(self, env)
size, num_rocks = self.model_spec.split('x')
self.size = int(size)
self.num_rocks = int(num_rocks)
def __init__(self, pre_words, post_words):
Model.__init__(self, pre_words, post_words)
self.sizemo = None
def __init__(self, ph):
Model.__init__(self, ph)
self.input_shape = INPUT_SHAPES[self.ph['dataset']]
def __init__(self, pre_words, post_words): Model.__init__(self, pre_words, post_words)