pgsearch
Parallel grid search.
Usage:
To use pgsearch, you need to wrap your model in a specific class (with whatever name you like) with a run method which receives only one argument, i.e., the process id. Keep everything you need in the initialization of the model and pass all arguments through a dict.
You can use a dict of parameters lists to generate all possible combinations of parameters:
from pgsearch import GridSearcher
parameter_dict = {
'pa': [1, 3, 3, 1, 2, 3, 3, 1, 2, 3, 3, 1, 2, 3],
'pb': [9, 1, 1, 2, 3, 3, 1, 2, 3],
'pd': [9, 1, 1, 2, 3, 3, 1, 2, 3],
'pe':[np.array([1,2,3])]
}
class Model:
def __init__(self, args):
self.args = args
pass
def run(self, x):
z = self.args['pa']+self.args['pb']+np.sum(self.args['pe'])
return x+z # include everything you need here from one trial.
gs = GridSearcher(Model, parameter_dict, processes=13, verbose=True, interval=0.1)
res = gs.search(save=False)
print(res[0])Or use a list of parameters dict directly:
class Model:
def __init__(self, args):
self.args = args
pass
def run(self, x):
z = self.args['pa']+self.args['pb']+np.sum(self.args['pe'])
return x+z # include everything you need here from one trial.
parameters = [{'pa': [i],'pb': [i+1],'pd': [i*2],'pe':[np.array([1,2,3])]} for i in range(300)]
gs = GridSearcher(Model, parameters, processes=13, verbose=True, interval=0.1, preprocess=False)
res = gs.search(save=False)
print(res[0])
