def convertLimits(limits, form = None):
if not isinstance(limits, list):
raise ValueError("No way to convert limits because {0} is not a list.".format(limits))
if not all(limits):
return limits ## return list because there is a None in the list
if form is None:
if len(limits) < 2:
raise ValueError("List {0} does not have two or more elements".format(limits))
if isinstance(limits[0], int):
dist = rint(limits[0], limits[-1])
elif isinstance(limits[0], float):
dist = rfloat(limits[0], limits[-1])
elif isinstance(limits[0], str):
dist = limits
else:
raise ValueError("Not sure how to process list {0}.".format(limits))
else:
if form == "uniform":
if len(limits) < 2:
raise ValueError("List {0} does not have two or more elements".format(limits))
if isinstance(limits[0], int):
dist = rint(limits[0], limits[-1])
elif isinstance(limits[0], float):
dist = rfloat(limits[0], limits[-1])
else:
raise ValueError("Form {0} is not supported.".format(form))
return dist
def opti_forest(data, n_jobs=1,n_iter=10):
model = RandomForestRegressor()
param_dist = {'n_estimators': rint(15, 30),
# 'criterion': ['mse','mae'], # not in 0.18 but in 0.19
'min_samples_split': rint(2, 10),
'min_samples_leaf': rint(1, 5),
'min_weight_fraction_leaf': [0.02],# opti said this is good
'max_features': [None], # None is best
'min_impurity_split': [0.03, 0.02, 0.01, 0.04], # min_impurity_decrease
"bootstrap": [True], # false conflicts with oob score thing
"oob_score": [False]}
X,y = getXY(data,data.keys())
blu = rsearch(model, param_distributions=param_dist, n_iter=n_iter,n_jobs=n_jobs)
blu.fit(X, y)
print blu.best_params_
print blu.best_score_
def getDecisionTreeClassifierParams():
treeParams = {
"max_depth": [2, 4, 6, None],
"max_features": ['auto', 'sqrt', 'log2', None],
"min_impurity_decrease": rfloat(0.0, 0.25),
"min_samples_leaf": rint(1, 10)
}
params = treeParams
params["criterion"] = ["gini", "entropy"]
param_grid = {}
for param, dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval
def getExtraTreesClassifierParams():
treeParams = {"max_depth": [2, 4, 6, 8, None],
"max_features": ['auto', 'sqrt', 'log2', None],
"min_impurity_decrease": rfloat(0.0, 0.25),
"min_samples_leaf": rint(1, 10)}
params = treeParams
#params["bootstrap"] = [False, True]
#params["criterion"] = ["mae", "mse"]
params["n_estimators"] = [100]
param_grid = {}
for param,dist in params.iteritems():
param_grid[param] = convertDistribution(dist)
retval = {"dist": params, "grid": param_grid}
return retval