def getSVMEpsRegressorParams(kernel):
param_grid = {}
baseParams = {
"C": genPowerTen(-1, 1, 9),
"epsilon": genLinear(0.1, 0.9, step=0.2),
"kernel": [kernel]
}
if kernel == "poly":
params = {
"coef0": genLinear(-1, 1, 3),
"degree": genLinear(1, 5, step=1),
"gamma": ['auto']
}
if kernel == "linear":
params = {}
if kernel == "sigmoid":
params = {"coef0": genLinear(-1, 1, 3), "gamma": ['auto']}
if kernel == "rbf":
params = {"gamma": ['auto']}
params = dict(baseParams.items() + params.items())
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getKernelRidgeRegressorParams():
param_grid = {}
params = {
"alpha": genPowerTen(-1, 1, 9),
"coef0": genLinear(-1, 1, 3),
"degree": genLinear(1, 5, step=1),
"kernel": ['linear', 'poly', 'rbf', 'sigmoid']
} #, 'precomputed']}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getXGBRegressorParams():
params = {
"gamma": genLinear(0, 1, step=0.2),
"max_depth": genLinear(2, 8, step=2),
"learning_rate": genPowerTen(-2, -0.5, 4),
"n_estimators": [50, 100, 200, 350, 500],
"reg_alpha": genPowerTen(-2, 1, 4),
"reg_lambda": genPowerTen(-2, 1, 4)
}
param_grid = {}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getSGDRegressorParams():
param_grid = {}
params = {
"alpha":
genPowerTen(-1, 1, 4),
"epsilon":
genLinear(0.05, 0.25, step=0.05),
"eta0":
genPowerTen(-3, -1, 5),
"l1_ratio":
genPowerTen(-2, 0, 5),
"learning_rate": ["constant", "optimal", "invscaling"],
"loss": [
"squared_loss", "huber", "epsilon_insensitive",
"squared_epsilon_insensitive"
],
"max_iter": [1000],
"penalty": ["l1", "l2"],
"power_t":
genPowerTwo(-3, -1, 3),
"tol": [0.001]
}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getRadiusNeighborsRegressorParams():
params = {
"algorithm": ['auto', 'ball_tree', 'kd_tree', 'brute'],
"leaf_size":
genLinear(10, 50, step=10),
"metric": [
'minkowski', 'cityblock', 'cosine', 'euclidean', 'l1', 'l2',
'manhattan'
],
"radius":
genLinear(0.5, 1.5, step=0.5),
"weights": ['uniform', 'distance']
}
param_grid = {}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getHuberRegressorParams():
param_grid = {}
params = {
"alpha": genPowerTen(-5, -3, 9),
"epsilon": genLinear(1.05, 1.65, step=0.05),
"max_iter": [1000],
"tol": [0.00001]
}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getMLPRegressorParams():
param_grid = {}
params = {
"activation": ["identity", "logistic", "tanh", "relu"],
"alpha": genPowerTen(-1, 1, 9),
"beta_1": genLinear(0.81, 0.99, step=0.04),
"hidden_layer_sizes": [(10, ), (25, ), (50, )], #, (100,), (250,)],
#"learning_rate": ["constant", "invscaling", "adaptive"],
#"momentum": genLinear(0.75, 0.95, step=0.05),
"max_iter": [500]
}
#"power_t": genLinear(0.25, 0.75, step=0.25)}
#"solver": ["lbfgs", "sgd", "adam"]}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getSGDClassifierParams():
param_grid = {}
params = {
"alpha": genPowerTen(-4, 4, 100),
"epsilon": genLinear(0.05, 0.25, step=0.05),
"eta0": genPowerTen(-3, -1, 5),
"l1_ratio": genPowerTen(-2, 0, 5),
"learning_rate": ["constant", "optimal", "invscaling"],
"loss": ["modified_huber", "log"],
"max_iter": [1000],
"penalty": ["l1", "l2"],
"power_t": genPowerTwo(-3, -1, 3),
"tol": [0.001]
}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval