CI

Author: xzhang2523-c

libmoon/auto_tester/test_synthetic.py

@@ -21,9 +21,7 @@
 import argparse
 import numpy as np
 
-
 if __name__ == '__main__':
-
     parser = argparse.ArgumentParser()
     parser.add_argument('--n-epoch', type=int, default=10000)
     parser.add_argument('--step-size', type=float, default=1e-3)
@@ -44,7 +42,6 @@
 
     solver = solver_dict[parser.parse_args().solver_name]
     args = parser.parse_args()
-
     args.method_name = args.solver_name if args.solver_name != 'GradAgg' \
         else '{}_{}'.format(args.solver_name, args.agg_name)
     problem = VLMOP1(n_var=10)
@@ -66,21 +63,17 @@
     ts = time() - ts
     y = res['y']
     fig = plt.figure()
-
     plt.scatter(y[:, 0], y[:, 1], s=100)
     rho_arr = np.linalg.norm(y, axis=1)
     plt.xlabel('$f_1$', fontsize=20)
     plt.ylabel('$f_2$', fontsize=20)
-
     plt.xlim([0,1])
     plt.ylim([0,1])
     plt.gca().set_aspect('equal', adjustable="box")
-
     plt.xticks(fontsize=18)
     plt.yticks(fontsize=18)
     os.makedirs(folder_name, exist_ok=True)
     prefs = prefs.numpy()
-
     prefs_norm = prefs / np.linalg.norm(prefs, axis=1, keepdims=True)
     rho_max = 1.2
     for pref in prefs_norm:
@@ -97,8 +90,6 @@
     plt.xlabel('Epoch', fontsize=20)
     plt.ylabel('Hypervolume', fontsize=20)
 
-
-
     plt.xticks(fontsize=15)
     plt.yticks(fontsize=15)
     plt.tight_layout()

libmoon/auto_tester/test_synthetic.sh

@@ -1,5 +1,9 @@
 nepoch=1000
 
+
+# Online using libmoon/auto_tester/X
+# pycharm using auto_tester/X
+
 for agg in STche
 do
   python libmoon/auto_tester/test_synthetic.py --solver-name GradAgg --agg-name $agg --n-epoch $nepoch

libmoon/solver/gradient/methods/base_solver.py

@@ -93,14 +93,16 @@ def solve(self, problem, x, prefs):
         if self.solver_name == 'UMOD':
             self.pfl_model = PFLModel(n_obj=problem.n_obj)
             self.pfl_optimizer = torch.optim.Adam(self.pfl_model.parameters(), lr=1e-3)
+
         self.n_prob, self.n_obj = prefs.shape[0], prefs.shape[1]
         xs_var = Variable(x, requires_grad=True)
         optimizer = Adam([xs_var], lr=self.step_size)
         ind = HV( ref_point=get_hv_ref(problem.problem_name) )
         hv_arr, y_arr = [], []
         # For UMOD solver, we need to store (pref, y) pairs.
         pref_y_pairs = []
-        for epoch_idx in tqdm(range(self.epoch)):
+
+        for epoch_idx in range(self.epoch):
             fs_var = problem.evaluate(xs_var)
             y_np = fs_var.detach().numpy()
             y_arr.append(y_np)
@@ -195,32 +197,4 @@ def __init__(self, problem, prefs, step_size=1e-3, n_epoch=500, tol=1e-3,
         super().__init__(step_size, n_epoch, tol, core_solver=self.core_solver)
 
     def solve(self, x_init):
-        return super().solve(self.problem, x_init, self.prefs)
-        # x = x_init
-        # x = Variable(x, requires_grad=True)
-        # ind = HV(ref_point=get_hv_ref(self.problem.problem_name))
-        # hv_arr = []
-        # y_arr, x_arr = [], []
-        # prefs = Tensor(self.prefs)
-        # optimizer = SGD([x], lr=self.step_size)
-        # agg_func = get_agg_func(self.agg)
-        # res = {}
-        #
-        # for epoch_idx in tqdm(range(self.n_epoch)):
-        #     y = self.problem.evaluate(x)
-        #     hv_arr.append(ind.do(y.detach().numpy()))
-        #     agg_val = agg_func(y, prefs)
-        #     optimizer.zero_grad()
-        #     torch.sum(agg_val).backward()
-        #     optimizer.step()
-        #     y_arr.append(y.detach().numpy())
-        #
-        #     if 'lbound' in dir(self.problem):
-        #         x.data = torch.clamp(x.data, torch.Tensor(self.problem.lbound) + solution_eps,
-        #                              torch.Tensor(self.problem.ubound) - solution_eps)
-        # res['x'] = x.detach().numpy()
-        # res['y'] = y.detach().numpy()
-        # res['hv_history'] = np.array(hv_arr)
-        # res['y_history'] = np.array(y_arr)
-        # res['x_history'] = np.array(y_arr)
-        # return res
+        return super().solve(self.problem, x_init, self.prefs)

libmoon/solver/gradient/methods/pmgda_solver.py

@@ -1,4 +1,7 @@
 from libmoon.solver.gradient.methods.base_solver import GradBaseSolver
+
+from libmoon.util.constant import root_name
+
 from torch.optim import SGD
 from tqdm import tqdm
 from pymoo.indicators.hv import HV
@@ -28,7 +31,8 @@ def __init__(self, n_var, prefs):
         self.n_prob, self.n_obj = prefs.shape[0], prefs.shape[1]
         self.n_var = n_var
         prefs_np = prefs.cpu().numpy() if type(prefs) == torch.Tensor else prefs
-        self.config_name = 'D:\\pycharm_project\\libmoon\\libmoon\\config\\pmgda.json'
+        self.config_name = os.path.join(root_name, 'config', 'pmgda.json')
+
         json_file = open(self.config_name, 'r')
         self.config = json.load(json_file)
         self.h_eps = self.config['h_eps']
@@ -217,7 +221,6 @@ def get_Jhf(f_arr, pref, return_h=False):
 
 class PMGDASolver(GradBaseSolver):
     # The PGMDA paper: http://arxiv.org/abs/2402.09492.
-
     def __init__(self, problem, prefs, step_size=1e-3, n_epoch=500, tol=1e-3,
                  sigma=0.1, h_tol=1e-3, folder_name=None):
         self.folder_name=folder_name