Merge pull request #62 from AstroAI-Lab/all-contributors

update contributors and docs

Author: TobiBu

.all-contributorsrc

@@ -48,6 +48,34 @@
         "talk",
         "mentoring"
       ]
+    },
+    {
+      "login": "lorenzobranca",
+      "name": "Lorenzo Branca",
+      "avatar_url": "https://avatars.githubusercontent.com/u/57775402?v=4",
+      "profile": "https://github.com/lorenzobranca",
+      "contributions": [
+        "data",
+        "ideas",
+        "research",
+        "test",
+        "talk"
+      ]
+    },
+    {
+        "login": "Immi000",
+        "name": "Immanuel Sulzer",
+        "avatar_url": "https://avatars.githubusercontent.com/u/100942429?v=4",
+        "profile": "https://github.com/Immi000",
+        "contributions": [
+          "code",
+          "content",
+          "data",
+          "design",
+          "doc",
+          "infra",
+          "userTesting"
+        ]
     }
   ]
 }

docs/source/guides/running-benchmarks/modalities.md

@@ -37,7 +37,7 @@ Interpolation studies remove every _n_-th timestep, forcing the surrogate to rec
 align: center
 alt: Interpolation modality example
 ---
-Interpolation MAE over time for several interval widths. Wider gaps create bigger spikes but also highlight which surrogates remain stable.
+Interpolation MAE over time for several interval widths. When the distance between time steps increases, one can often observe error spikes between them. 
 ```
 
 ## Extrapolation
@@ -61,11 +61,12 @@ Sparse training reduces the number of observations before fitting, emulating lim
 align: center
 alt: Sparse modality example
 ---
-Down-sampling trajectories shows how MAE changes with fewer observations; FCNN tends to degrade earlier than the latent models.
+Reducing the numer of samples (sets of trajectories) shows how MAE changes with fewer observations, revealing how efficiently the model extracts information from samples, and equivalently informing whether more training data could help improve this surrogate's performance.
 ```
 
 ## Batch scaling
 
 Batch scaling sweeps different batch sizes and records how accuracy/timing behave. This is useful to identify sweet spots for throughput without impacting convergence too heavily. Combine the results with the `timing` evaluation to compare throughput across surrogates.
 
 See the :doc:`configuration reference </reference/configuration>` for the exact YAML schema and defaults.
+

docs/source/index.rst

@@ -12,7 +12,7 @@ CODES Benchmark
 * **Extend the Stack** — :doc:`guides/extending-benchmark` shows how to add datasets or surrogates without rewriting orchestration glue.
 * **API Reference** — :doc:`api-reference` explains how the generated package docs are organized and links to each module.
 
-Looking for a bird’s-eye view first? Start with the **User Guide**. Already configuring experiments or integrating your own model? Skip ahead to the **API Reference**. Either way, the sidebar mirrors the sections below so you are one click away from the next step.
+Looking for a bird’s-eye view first? Start with the **User Guide**. Already configuring experiments or integrating your own model? Skip ahead to the **API Reference**. The sidebar mirrors the sections below so you are one click away from the next step.
 
 .. toctree::
    :maxdepth: 2