Esempio n. 1
0
    def __init__(self, rules, random_explore):
        self.random_explore = random_explore
        self.lock = threading.RLock()
        self.rules = rules
        self.build_grid()
        self.all_bas = []
        self.ras = {"T": [], "SG": []}
        for teacher in range(rules["nb_teachers"]):
            self.ras["T"].append(RA("T", teacher, rules, self, self.random_explore))
            self.all_bas.extend(self.ras["T"][-1].bas)

        for st_group in range(rules["nb_st_groups"]):
            self.ras["SG"].append(RA("SG", st_group, rules, self, self.random_explore))
            self.all_bas.extend(self.ras["SG"][-1].bas)

        # self.step_count = 0
        self.all_ras = self.ras["T"] + self.ras["SG"]
Esempio n. 2
0
    def _test_avg(self):
        dataset = nc_rna_reader.toNumpy()
        train_set_size = 200

        X_train_full, y_train_full, X_test_full, y_test_full = dataset
        X_train, y_train = self.get_sub_set_with_size([X_train_full, y_train_full], train_set_size)
        X_test, y_test = self.get_sub_set_with_size([X_test_full, y_test_full], 10000)

        train_set = (X_train, y_train)
        test_set_original = (X_test, y_test)

        clf_class = LinearSVC

        for split_r in numpy.arange(0.1, 1.0, 0.1):
            ra = RA(clf_class, ac_method="ac", subsample_count=200, split_r=split_r)
            ra.fit(train_set)
            err = self.compute_avg_error(ra, test_set_original)
            print "%f\t%f" % (split_r, err)
Esempio n. 3
0
    def _test_avg(self):
        dataset = rcv1_binary_reader.toNumpy()
        train_set_size = 300

        X_train_full, y_train_full, X_test_full, y_test_full = dataset
        X_train, y_train = self.get_sub_set_with_size([X_train_full, y_train_full], train_set_size)
        X_test, y_test = self.get_sub_set_with_size([X_test_full, y_test_full], 10000)

        train_set = (X_train, y_train)
        test_set_original = (X_test, y_test)

        clf_class = LogisticRegression

        for split_r in numpy.arange(0.1, 1.0, 0.1):
            ra = RA(clf_class, ac_method = 'ac', subsample_count = 200, split_r=split_r)
            ra.fit(train_set)
            err = self.compute_avg_error(ra, test_set_original)
            print split_r, err