3layer kan

tasks/models/frugal_2025-01-26/frugal_kan_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f5652f3be0033b5a249ded449c1f7c40dd2f32d649b07ca7c2f6158a4b57cb5
+size 1710980

tasks/models/frugal_2025-01-27/frugal_kan_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f520cff8b9531981e16a8b009b6a55fb8ca98573fc4d3dc6806df60b07a49c2
+size 1710980

tasks/run.py

@@ -9,6 +9,7 @@ import yaml
 import datetime
 import json
 import numpy as np
+from collections import OrderedDict
 
 # local_rank = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 current_date = datetime.date.today().strftime("%Y-%m-%d")
@@ -56,6 +57,13 @@ model = CNNKan(model_args, conformer_args, kan_args.get_dict())
 # model.kan.speed()
 # model = KanEncoder(kan_args.get_dict())
 model = model.to(local_rank)
+state_dict = torch.load(data_args.checkpoint_path, map_location=torch.device('cpu'))
+new_state_dict = OrderedDict()
+for key, value in state_dict.items():
+    if key.startswith('module.'):
+        key = key[7:]
+    new_state_dict[key] = value
+missing, unexpected = model.load_state_dict(new_state_dict)
 # model = DDP(model, device_ids=[local_rank], output_device=local_rank)
 num_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
 print(f"Number of parameters: {num_params}")

tasks/utils/config.yaml

@@ -12,7 +12,7 @@ Data:
   max_days_lc: 270
   lc_freq: 0.0208
   create_umap: True
-  checkpoint_path: 'tasks/models/frugal_2025-01-21/frugal_kan_2.pth'
+  checkpoint_path: 'tasks/models/frugal_2025-01-27/frugal_kan_2.pth'
 
 CNNEncoder:
   # Model
@@ -31,7 +31,7 @@ CNNEncoder:
   avg_output: False
 
 KAN:
-  layers_hidden: [1125,32,8,8,1]
+  layers_hidden: [1125,32,8,1]
   grid_min:  -1.2
   grid_max:  1.2
   num_grids:  8