[Feat] data download/generation docs #1

README.md

@@ -40,6 +40,19 @@ If you would like to install all dependencies including optional solvers, please
 
 We recommend exploring [this quickstart notebook](examples/1.quickstart.ipynb) to get started with the `RouteFinder` codebase!
 
+
+### Generating Data
+
+Data may be generated by running the following command:
+
+```bash
+python routefinder/data/generate_data.py
+```
+and will be saved under the `data/` directory.
+
+Note that we provide the original testing data since the data may differ slightly across devices due to PyTorch's random number generator. The distribution will however be the same, so results should be comparable. To ensure full reproducibility and make sure the data is exactly the same, you may use the uploaded files under the `data/` folder.
+
+
 ### Running
 
 The main runner (example here of main baseline) can be called via:
@@ -50,6 +63,8 @@ python run.py experiment=main/rf/rf-100
 You may change the experiment by using the `experiment=YOUR_EXP`, with the path under [`configs/experiment`](configs/experiment) directory.
 
 
+
+
 ## 🚚 Available Environments
 
 <div align="center">

routefinder/data/generate_data.py

@@ -10,12 +10,21 @@
 folder = "data/"
 
 
-def generate(num_loc, num_data, variant, phase="val"):
-    filename = f"{variant}/{phase}/{num_loc}.npz"
+def generate(num_loc, num_data, variant, phase="val", mixed=False):
+    if mixed:
+        # variant mb: find "b", insert "m" before i
+        new_variant = variant[: variant.find("b")] + "m" + variant[variant.find("b") :]
+        filename = f"{new_variant}/{phase}/{num_loc}.npz"
+        backhaul_class = 2
+    else:
+        filename = f"{variant}/{phase}/{num_loc}.npz"
+        backhaul_class = 1
+
     path = os.path.join(folder, filename)
     os.makedirs(os.path.dirname(path), exist_ok=True)
-
-    generator = MTVRPGenerator(num_loc=num_loc, variant_preset=variant)
+    generator = MTVRPGenerator(
+        num_loc=num_loc, variant_preset=variant, backhaul_class=backhaul_class
+    )
     env = MTVRPEnv(generator, check_solution=False)
     td_data = env.generator(num_data)
 
@@ -24,24 +33,33 @@ def generate(num_loc, num_data, variant, phase="val"):
 
 
 def main():
+    # validation has less data for faster training
     for variant in MTVRPGenerator.available_variants():
-        # Validation (less data for faster training)
         generate(50, 128, variant, phase="val")
         generate(100, 128, variant, phase="val")
         generate(200, 128, variant, phase="val")
-
-        # Test
         generate(50, 1000, variant, phase="test")
         generate(100, 1000, variant, phase="test")
         generate(200, 1000, variant, phase="test")
-        generate(500, 128, variant, phase="test")
-        generate(1000, 128, variant, phase="test")
+
+        # mixed variants: if not contains "b", skip
+        if "b" not in variant:
+            continue
+        else:
+            generate(50, 128, variant, phase="val", mixed=True)
+            generate(100, 128, variant, phase="val", mixed=True)
+            generate(200, 128, variant, phase="val", mixed=True)
+            generate(50, 1000, variant, phase="test", mixed=True)
+            generate(100, 1000, variant, phase="test", mixed=True)
+            generate(200, 1000, variant, phase="test", mixed=True)
 
 
 if __name__ == "__main__":
     input(
-        "WARNING: you should not generate the dataset but download it from Github"
-        " since generation results are not reproducible across devices. Press Enter to continue anyways."
+        "Warning: generated data may differ slightly across devices due to PyTorch's random number generator. "
+        "The distribution will however be the same, so results should be comparable. "
+        "To ensure full reproducibility and make sure the data is exactly the same, you may use the uploaded files under the data/ folder."
+        "Note that this will overwrite any existing datasets. Press Enter to confirm."
     )
 
     main()