LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=1, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=2, variance=1.0, lengthscale=1.0)),
GPyMT(GPy.kern.Matern52(input_dim=2, variance=1.0, lengthscale=1.0)),
CHOR(Chorus_config),
]
fmap = ModelMapMap.ModelMapMapDifference0_1(auth_model, map_models)
# -----------------------------------------------------------------------------------------------------------------
#
# TPC-C Scenario 1 - Predicting Performance with Increased CPU Resources
#
# -----------------------------------------------------------------------------------------------------------------
experiment_tpcc_1 = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename = authoritative_dataset_filename,
domain_columns = domain_columns,
codomain_columns = codomain_columns,
num_runs = num_runs,
sampling_budgets = sampling_budgets,
authoritative_models = authoritative_models,
authoritative_model_dataset_range = authoritative_model_dataset_range,
map_models = [
LINM(),
]
direct_models = []
Application_Experiment_2a = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
authoritative_model_dataset_range=authoritative_model_dataset_range,
unknown_function_dataset_range=unknown_function_dataset_range,
map=ModelMapMap.ModelMapMapComposition0_1(auth_model, map_models),
direct_modeling_methods=direct_models,
active_learning=False,
model_selection=False,
logger=logger,
title="APP-Scenario2a-")
map_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=3, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
map_models = [
LINM(),
]
direct_models = []
Application_Experiment_3a = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
authoritative_model_dataset_range=authoritative_model_dataset_range,
unknown_function_dataset_range=unknown_function_dataset_range,
map=ModelMapMap.ModelMapMapComposition0_1(auth_model, map_models),
direct_modeling_methods=direct_models,
active_learning=False,
model_selection=False,
logger=logger,
title="APP-Scenario3a-")
map_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=3, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=1, variance=1.0, lengthscale=1.0)),
]
experiment_tpcc_4a = ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
unknown_dataset_filename=unknown_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=authoritative_models,
authoritative_model_dataset_range=authoritative_model_dataset_range,
unknown_function_dataset_range=unknown_function_dataset_range,
map=ModelMapMap.ModelMapMapComposition0_1(auth_model, map_models_a),
direct_modeling_methods=direct_models,
active_learning=False,
model_selection=False,
log2_columns=log2_columns,
normalization='N',
logger=logger,
title="TPC-C-Scenario4a-")
map_models_b = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=2, variance=1.0, lengthscale=1.0)),
]
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=3, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=5, variance=1.0, lengthscale=1.0)),
GPyMT(GPy.kern.Matern52(input_dim=5, variance=1.0, lengthscale=1.0)),
CHOR(Chorus_config),
]
fmap = ModelMapMap.ModelMapMapDifferenceL_1(auth_model, map_models,
["inode_size", "bg_count"])
# -----------------------------------------------------------------------------------------------------------------
#
# Filesystem Scenario 3 - Modeling the effects of disk type (SATA -> 500SAS) and I/O scheduler (CFQ -> Deadline)
#
# -----------------------------------------------------------------------------------------------------------------
Filesystem_Experiment_3 = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
authoritative_model_dataset_range=authoritative_model_dataset_range,
map_models = [
LINM(),
]
direct_models = []
Application_Experiment_1a = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
authoritative_model_dataset_range=authoritative_model_dataset_range,
unknown_function_dataset_range=unknown_function_dataset_range,
map=ModelMapMap.ModelMapMapComposition0_1(auth_model, map_models),
direct_modeling_methods=direct_models,
active_learning=False,
model_selection=False,
logger=logger,
title="APP-Scenario1a-"
)
map_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=2, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=5, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=7, variance=1.0, lengthscale=1.0)),
GPyMT(GPy.kern.Matern52(input_dim=7, variance=1.0, lengthscale=1.0)),
CHOR(Chorus_config),
]
fmap = ModelMapMap.ModelMapMapCompositionL_1(auth_model, map_models, ["inode_size", "bg_count", "disk_type", "io_scheduler"])
# -----------------------------------------------------------------------------------------------------------------
#
# Filesystem Scenario 5 - Modeling the effects changing filesystem (ext3 -> ext4)
#
# -----------------------------------------------------------------------------------------------------------------
Filesystem_Experiment_5 = ModelMapExperiment.ModelMapExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
unknown_dataset_filename=unknown_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=num_runs,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
def experiment_func(iteration):
# create logger with 'modelmap'
logger = logging.getLogger("ModelMap")
logger.setLevel(logging.DEBUG)
# create file handler which logs even debug messages
file_logger = logging.FileHandler('modelmap.log')
# create console handler with a higher log level
console_logger = logging.StreamHandler()
# ----------------------------------------------------------------------------
file_logger.setLevel(logging.DEBUG)
console_logger.setLevel(logging.DEBUG)
# ----------------------------------------------------------------------------
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
file_logger.setFormatter(formatter)
console_logger.setFormatter(formatter)
# add the handlers to the logger
logger.addHandler(file_logger)
logger.addHandler(console_logger)
######################################################################################################
# Chorus regressor
fit_dimension = 0
# Chorus G_RGN model
G_RGN_config = {"fit_column": fit_dimension + 1,
"type": "poly",
"order": "4"
}
# Chorus B_SVM global model
B_SVM_config = {"USE_AVG": "true"}
# Chorus B_LIN global region split model
B_LIN_config = {"max_regions": 2,
"specdb": {0: (0.0, 1.0, 0),
1: (0.0, 1.0, 0),
2: (0.0, 1.0, 0)}
}
# Full Chorus3 ensemble model
Chorus_config = {
"models": {
"Chorus_G_RGN": G_RGN_config,
"Chorus_B_CNST": dict(),
"Chorus_B_LIN": B_LIN_config,
"Chorus_B_SVM": B_SVM_config,
},
"max_regions": 2,
"n_way_cross": 5,
"specdb": {0: (0.0, 1.0, 0),
1: (0.0, 1.0, 0),
2: (0.0, 1.0, 0)}
}
######################################################################################################
seed = 1
LIN = type('LIN', LinearRegression.__bases__, dict(LinearRegression.__dict__))
LIN.label = "Linear regression"
POL = type('POL', PolynomialRegressor.__bases__, dict(PolynomialRegressor.__dict__))
POL.label = "Polynomial regression"
LINSVR = type('POL', LinearSVR.__bases__, dict(LinearSVR.__dict__))
LINSVR.label = "Linear SVR"
GPyR = type('GPyR', GPyRegressor.__bases__, dict(GPyRegressor.__dict__))
GPyR.label = "Gaussian Process"
GPyMT = type('GPyMT', GPyRegressorMultiTask.__bases__, dict(GPyRegressorMultiTask.__dict__))
GPyMT.label = "Multi-Task Gaussian Process*"
CHOR = ChorusIncrementalRegressor
CHOR.label = "Chorus*"
######################################################################################################
# Authoritative model
auth_model = DatabaseRegressor()
# Dataset file
authoritative_dataset_filename = "Datasets/tpcc_blade_3min_10WH_20170725.csv"
unknown_dataset_filename = "Datasets/tpcc_blade_3min_20WH.csv"
# Columns containing the function domain
domain_columns = ["CPU", "disk_bandwidth_quanta", "buffer_pool"]
# Columns containing the function codomain
codomain_columns = "TPMC"
# list of Number of samples
sampling_budgets = [5, 10]
# Authoritative model
authoritative_model_dataset_range = {
"CPU": [0, 1],
"disk_bandwidth_quanta": [1, 48],
"buffer_pool": [1, 1024]
}
# Unknown model
unknown_function_dataset_range = {
"CPU": [0, 1],
"disk_bandwidth_quanta": [1, 48],
"buffer_pool": [1, 1024]
}
log2_columns = ["CPU", "disk_bandwidth_quanta", "buffer_pool"]
map_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=2, variance=1.0, lengthscale=1.0)),
]
direct_models = [
LIN(),
POL(degree=4),
LINSVR(random_state=seed),
GPyR(GPy.kern.Matern52(input_dim=3, variance=1.0, lengthscale=1.0)),
GPyMT(GPy.kern.Matern52(input_dim=3, variance=1.0, lengthscale=1.0)),
CHOR(Chorus_config),
]
fmap = ModelMapMap.ModelMapMapDifferenceL_1(auth_model, map_models, ["disk_bandwidth_quanta"])
# -----------------------------------------------------------------------------------------------------------------
#
# TPC-C Scenario 5 - Modeling VM packing optimization (10WH -> 20WH map class (L>0:1))
#
# -----------------------------------------------------------------------------------------------------------------
experiment = ModelMapVMPackingExperiment.ModelMapVMPackingExperiment(
authoritative_dataset_filename=authoritative_dataset_filename,
unknown_dataset_filename=unknown_dataset_filename,
domain_columns=domain_columns,
codomain_columns=codomain_columns,
num_runs=iteration,
sampling_budgets=sampling_budgets,
authoritative_models=auth_model,
authoritative_model_dataset_range=authoritative_model_dataset_range,
unknown_function_dataset_range=unknown_function_dataset_range,
map = fmap,
direct_modeling_methods=direct_models,
log2_columns=log2_columns,
logger=logger,
title="TPC-C-Scenario5-"
)
# Run the experiment
experiment.run()