Clean up dataset visualisations, add lv_parametric dataset variants

Author: robin-janssen

config.yaml

@@ -1,36 +1,36 @@
 # Global settings for the benchmark
-training_id: "primordial_params1"
+training_id: "primordial_params_others"
 surrogates: ["MultiONet", "FullyConnected", "LatentPoly"] # ["MultiONet", "FullyConnected", "LatentNeuralODE", "LatentPoly"]
 batch_size: [8192, 8192, 512] # [8192, 8192, 512, 512]
-epochs: [15000, 15000, 15000] # [15000, 15000, 15000, 15000]
+epochs: [20000, 7500, 15000] # [20000, 7500, 20000, 15000]
 dataset: 
   name: "primordial_parametric"
   log10_transform: True
   log10_transform_params: False
   normalise: "minmax" # "minmax" # "standardise", "minmax", "disable"
   use_optimal_params: True
-  tolerance: 1e-15
+  tolerance: 1e-20
   subset_factor: 1
   log_timesteps: True
-devices: ["cuda:3"]
+devices: ["cuda:1", "cuda:2", "cuda:4", "cuda:5", "cuda:7", "cuda:9"]
 seed: 42
 verbose: False
 
 # Models to train
 interpolation: 
-  enabled: False
+  enabled: True
   intervals: [2, 3, 4, 5, 6, 7, 8, 9, 10]
 extrapolation:
-  enabled: False
+  enabled: True
   cutoffs: [50, 60, 70, 80, 90]
 sparse: 
-  enabled: False
+  enabled: True
   factors: [2, 4, 8, 16, 32]
 batch_scaling:
-  enabled: False
+  enabled: True
   sizes: [64, 512, 1024, 4096]
 uncertainty: 
-  enabled: False
+  enabled: True
   ensemble_size: 5  # Number of models for deep ensemble
 
 # Evaluations during benchmark

datasets/_data_analysis/analyse_all_datasets.py

@@ -33,9 +33,10 @@ def main():
     Main function to analyse the dataset. It checks the dataset and loads the data.
     """
     datasets = [
+        "lv_parametric",
+        "lv_parametric_no_params",
         "primordial_parametric",
         "primordial",
-        "lv_parametric",
         "simple_reaction",
         "osutest2",
         "coupled_oscillators",
@@ -122,7 +123,7 @@ def main():
                 full_data,
                 labels,
                 max_quantities=10,
-                threshold=2,
+                threshold=4,
                 max_faulty=5,
                 quantities_per_plot=qpp,
             )

datasets/_data_analysis/dataset_dict.py

@@ -54,18 +54,23 @@
     "primordial": {
         "log": True,
         "qpp": 5,
-        "tol": 1e-30,
+        "tol": 1e-20,
         "log_time": True,
     },
     "lv_parametric": {
         "log": True,
         "qpp": 3,
-        "tol": 1e-30,
+        "tol": 1e-20,
+    },
+    "lv_parametric_no_params": {
+        "log": True,
+        "qpp": 3,
+        "tol": 1e-20,
     },
     "primordial_parametric": {
         "log": True,
         "qpp": 5,
-        "tol": 1e-30,
+        "tol": 1e-20,
         "log_time": True,
     },
 }

datasets/data_sources.yaml

@@ -8,3 +8,5 @@ branca_norad: "https://zenodo.org/records/14041124/files/data.hdf5?download=1"
 branca_large_kyr: "https://zenodo.org/records/14204650/files/data.hdf5?download=1"
 branca_large_myr: "https://zenodo.org/records/14243267/files/data.hdf5?download=1"
 coupled_oscillators: "https://zenodo.org/records/14717175/files/data.hdf5?download=1"
+lv_parametric: https://zenodo.org/records/15412214/files/data.hdf5?download=1
+lv_parametric_no_params: https://zenodo.org/records/15412297/files/data.hdf5?download=1