ml4co-kit 0.3.2__tar.gz → 0.3.3__tar.gz

{ml4co_kit-0.3.2/ml4co_kit.egg-info → ml4co_kit-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ml4co_kit
-Version: 0.3.2
+Version: 0.3.3
 Summary: ml4co-kit provides convenient dataset generators for the combinatorial optimization problem
 Home-page: https://github.com/Thinklab-SJTU/ML4CO-Kit
 Author: SJTU-ReThinkLab

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/__init__.py

@@ -101,5 +101,5 @@ if found_pytorch_lightning is not None:
     from .learning import BaseEnv, BaseModel, Checkpoint, Logger, Trainer
 
 
-__version__ = "0.3.2"
+__version__ = "0.3.3"
 __author__ = "SJTU-ReThinkLab"

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mcl/local_search/rlsa.py

@@ -9,7 +9,7 @@ def mcl_rlsa_local_search(
     init_sol: np.ndarray,
     graph: np.ndarray,
     rlsa_init_type: str = "uniform",
-    rlsa_kth_dim: Union[str, int] = 0,
+    rlsa_kth_dim: Union[str, int] = "both",
     rlsa_tau: float = 0.01, 
     rlsa_d: int = 2, 
     rlsa_k: int = 1000, 

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mcut/decoder/rlsa.py

@@ -10,7 +10,7 @@ def mcut_rlsa_decoder(
     edge_index: np.ndarray,
     rlsa_kth_dim: Union[str, int] = 0,
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
     rlsa_device: str = "cpu", 

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mcut/local_search/rlsa.py

@@ -9,9 +9,9 @@ def mcut_rlsa_local_search(
     init_sol: np.ndarray,
     graph: np.ndarray,
     edge_index: np.ndarray,
-    rlsa_kth_dim: Union[str, int] = 0,
+    rlsa_kth_dim: Union[str, int] = "both",
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
     rlsa_device: str = "cpu", 

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mis/decoder/rlsa.py

@@ -10,7 +10,7 @@ def mis_rlsa_decoder(
     rlsa_init_type: str = "uniform",
     rlsa_kth_dim: Union[str, int] = 0,
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
     rlsa_alpha: float = 0.3,

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mis/local_search/rlsa.py

@@ -9,9 +9,9 @@ def mis_rlsa_local_search(
     init_sol: np.ndarray,
     graph: np.ndarray,
     rlsa_init_type: str = "uniform",
-    rlsa_kth_dim: Union[str, int] = 0,
+    rlsa_kth_dim: Union[str, int] = "both",
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
     rlsa_alpha: float = 0.3,
@@ -31,10 +31,10 @@ def mis_rlsa_local_search(
     # initial solutions
     x = init_sol.repeat(rlsa_k, 1).to(rlsa_device).float()
     if rlsa_init_type == "gaussian":
-        x[1:] = rlsa_alpha * torch.randn_like(x[1:])
+        x[1:] = (1-rlsa_alpha) * x[1:] + rlsa_alpha * torch.randn_like(x[1:])
         x = torch.clip(x, 0, 1).float()
     elif rlsa_init_type == "uniform":
-        x[1:] = rlsa_alpha * torch.randint_like(x[1:], high=2)
+        x[1:] = (1-rlsa_alpha) * x[1:] + rlsa_alpha * torch.randint_like(x[1:], high=2)
     else:
         raise NotImplementedError(
             "Only ``gaussian`` and ``uniform`` distributions are supported!"

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mvc/decoder/rlsa.py

@@ -1,18 +1,19 @@
 import torch
 import numpy as np
 from torch import Tensor
-from typing import Tuple
+from typing import Tuple, Union
 from ml4co_kit.learning.utils import to_numpy, to_tensor
 
 
 def mvc_rlsa_decoder(
     graph: np.ndarray,
     rlsa_init_type: str = "uniform",
+    rlsa_kth_dim: Union[str, int] = 0,
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
-    rlsa_alpha: float = 0.2,
+    rlsa_alpha: float = 0.3,
     rlsa_beta: float = 1.02,
     rlsa_device: str = "cpu", 
     seed: int = 1234
@@ -48,13 +49,16 @@ def mvc_rlsa_decoder(
     
     # SA
     for epoch in range(rlsa_t):
+        # kth_dim
+        kth_dim = epoch % 2 if rlsa_kth_dim == "both" else rlsa_kth_dim
+        
         # temperature
         tau = rlsa_tau * (1 - epoch / rlsa_k)
 
         # sampling
         delta = grad * (2 * x - 1) / 2
         k = torch.randint(2, rlsa_d + 1, size=(1,)).item()
-        term2 = -torch.kthvalue(-delta, k, dim=1, keepdim=True).values
+        term2 = -torch.kthvalue(-delta, k, dim=kth_dim, keepdim=True).values
         flip_prob = torch.sigmoid((delta - term2) / tau)
         rr = torch.rand_like(x.data.float())
         x = torch.where(rr < flip_prob, 1 - x, x)

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/algorithm/mvc/local_search/rlsa.py

@@ -9,9 +9,9 @@ def mvc_rlsa_local_search(
     init_sol: np.ndarray,
     graph: np.ndarray,
     rlsa_init_type: str = "uniform",
-    rlsa_kth_dim: Union[str, int] = 0,
+    rlsa_kth_dim: Union[str, int] = "both",
     rlsa_tau: float = 0.01, 
-    rlsa_d: int = 2, 
+    rlsa_d: int = 5, 
     rlsa_k: int = 1000, 
     rlsa_t: int = 1000, 
     rlsa_alpha: float = 0.3,

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/solver/cvrp/base.py

@@ -300,8 +300,8 @@ class CVRPSolver(SolverBase):
         for idx in range(self.points.shape[0]):
             cur_points = self.points[idx]
             cur_depots = self.depots[idx]
-            max_value = np.max(cur_points)
-            min_value = np.min(cur_points)
+            max_value = max(np.max(cur_points), np.max(cur_depots))
+            min_value = min(np.min(cur_points), np.min(cur_depots))
             cur_points = (cur_points - min_value) / (max_value - min_value)
             cur_depots = (cur_depots - min_value) / (max_value - min_value)
             self.points[idx] = cur_points

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/solver/mis/base.py

@@ -426,7 +426,7 @@ class MISSolver(SolverBase):
                     self_loop = self_loop.reshape(-1, 1).repeat(2, axis=1)
                     edge_index = np.concatenate([self_loop, edge_index.T], axis=0)
                     nx_graph: nx.Graph = nx.from_edgelist(edge_index)
-                    nx_graph = nx_graph.remove_edges_from(nx.selfloop_edges(nx_graph))
+                    nx_graph.remove_edges_from(nx.selfloop_edges(nx_graph))
                 else:
                     nx_graph = nx.from_edgelist(edge_index.T)
 

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3}/ml4co_kit/utils/graph/mvc.py

@@ -93,7 +93,9 @@ class MVCGraphData(GraphData):
         self._check_nodes_label(ref=ref)
         sol = self.ref_nodes_label if ref else self.nodes_label
         if sol is not None:
-            src_dst = sol[self.edge_index[0]] + sol[self.edge_index[1]]
+            mask = self.edge_index[0] != self.edge_index[1]
+            edge_index = self.edge_index[:, mask]
+            src_dst = sol[edge_index[0]] + sol[edge_index[1]]
             if src_dst.min() == 0:
                 raise ValueError(
                     "The solution does not conform to constraint!"

{ml4co_kit-0.3.2 → ml4co_kit-0.3.3/ml4co_kit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ml4co_kit
-Version: 0.3.2
+Version: 0.3.3
 Summary: ml4co-kit provides convenient dataset generators for the combinatorial optimization problem
 Home-page: https://github.com/Thinklab-SJTU/ML4CO-Kit
 Author: SJTU-ReThinkLab