rename tsp_2opt -> tsp_nls

README.md

@@ -22,7 +22,7 @@ Welcome! This repository contains the code implementation of paper [*DeepACO: Ne
 ---
 
 ### Usage
-- **Traveling Salesman Problem (TSP).** Please refer to `tsp/` for vanilla DeepACO and `tsp_2opt/` for DeepACO with NLS on TSP.
+- **Traveling Salesman Problem (TSP).** Please refer to `tsp/` for vanilla DeepACO and `tsp_nls/` for DeepACO with NLS on TSP.
 - **Capacitated Vehicle Routing Problem (CVRP).** Please refer to `cvrp/` for vanilla DeepACO and `cvrp_nls/` for DeepACO with NLS on CVRP.
 - **Orienteering Problem (OP).** Please refer to `op/`.
 - **Prize Collecting Travelling Salesman Problem (PCTSP).** Please refer to `pctsp/`.

cvrp/aco.py

@@ -205,6 +205,7 @@ def check_done(self, visit_mask, actions):
         return (visit_mask[:, 1:] == 0).all() and (actions == 0).all()    
 
     # ======== code for adaptive elitist AS ========
+    # These code are unrelated to DeepACO, and are kept for comparisons.
     def get_subroutes(self, route, end_with_zero = True):
         x = torch.nonzero(route == 0).flatten()
         subroutes = []

cvrp_nls/README.md

@@ -1,3 +1,5 @@
+# DeepACO with NLS for CVRP
+
 We adopted the C++ implementation of the SWAP* algorithm provided by Vidal T. et al [1,2] ([source repo](https://github.com/vidalt/HGS-CVRP)) and added a C++ interface to directly utilize the local search algorithm in the program. We also utilized and modified the Python wrapper provided by [PyHygese](https://github.com/chkwon/PyHygese). We would like to express our gratitude for the contributions made by the aforementioned paper and project. 
 
 To run our program, please follow the steps below to compile the shared library (if the file `./HGS-CVRP-main/build/libhgscvrp.so` we provide isn't compatiable with your setup)

cvrp_nls/aco.py

@@ -287,6 +287,7 @@ def positions_cpu(self):
         return self.positions.cpu().numpy() if self.positions is not None else None
 
     # ======== code for adaptive elitist AS ========
+    # These code are unrelated to DeepACO, and are kept for comparisons.
     def insertion_single(self, route, index):
         # route starts from 0, terminates with 0
         insertion_cost = (((self.distances[p1, index]+self.distances[index,p2]-self.distances[p1, p2]).item(), i) 

tsp_2opt/README.md → tsp_nls/README.md

@@ -1,8 +1,8 @@
-## DeepACO with NLS
+## DeepACO with NLS for TSP
 
 ### Training
 
-The checkpoints will be saved in [`../pretrained/tsp_2opt`](../pretrained/tsp_2opt) with suffix `-best.pt` and `-last.pt` by default.
+The checkpoints will be saved in [`../pretrained/tsp_nls`](../pretrained/tsp_nls) with suffix `-best.pt` and `-last.pt` by default.
 
 TSP200:
 ```raw

tsp_2opt/test.py → tsp_nls/test.py

@@ -69,11 +69,11 @@ def main(n_node, model_file, k_sparse = None, n_ants=48, t_aco = None):
 
     parser = argparse.ArgumentParser()
     parser.add_argument("nodes", type=int, help="Problem scale")
-    parser.add_argument("-m", "--model", type=str, default=None, help="Path to checkpoint file, default to '../pretrained/tsp_2opt/tsp{nodes}.pt'")
+    parser.add_argument("-m", "--model", type=str, default=None, help="Path to checkpoint file, default to '../pretrained/tsp_nls/tsp{nodes}.pt'")
     opt = parser.parse_args()
     n_nodes = opt.nodes
 
-    filepath = opt.model or f'../pretrained/tsp_2opt/tsp{n_nodes}.pt'
+    filepath = opt.model or f'../pretrained/tsp_nls/tsp{n_nodes}.pt'
     if not os.path.isfile(filepath):
         print(f"Checkpoint file '{filepath}' not found!")
         exit(1)

tsp_2opt/train.py → tsp_nls/train.py

@@ -88,7 +88,7 @@ def validation(n_ants, epoch, net, val_dataset):
     return avg_stats
 
 
-def train(n_node, n_ants, steps_per_epoch, epochs, k_sparse = None, batch_size = 3, test_size = None, pretrained = None, savepath = "../pretrained/tsp_2opt"):
+def train(n_node, n_ants, steps_per_epoch, epochs, k_sparse = None, batch_size = 3, test_size = None, pretrained = None, savepath = "../pretrained/tsp_nls"):
     k_sparse = k_sparse or n_node//10
     net = Net().to(device)
     if pretrained:
@@ -121,7 +121,7 @@ def train(n_node, n_ants, steps_per_epoch, epochs, k_sparse = None, batch_size =
 
     print('\ntotal training duration:', sum_time)
 
-    return f'../pretrained/tsp_2opt/tsp{n_node}.pt'
+    return f'../pretrained/tsp_nls/tsp{n_node}.pt'
 
 
 if __name__ == "__main__":
@@ -138,7 +138,7 @@ def train(n_node, n_ants, steps_per_epoch, epochs, k_sparse = None, batch_size =
     parser.add_argument("-s", "--steps", type=int, default=20, help="Steps per epoch")
     parser.add_argument("-e", "--epochs", type=int, default=100, help="Epochs to run")
     parser.add_argument("-t", "--test_size", type=int, default=None, help="Number of instances for testing")
-    parser.add_argument("-o", "--output", type=str, default="../pretrained/tsp_2opt",
+    parser.add_argument("-o", "--output", type=str, default="../pretrained/tsp_nls",
                         help="The directory to store checkpoints")
     opt = parser.parse_args()