Update categories

Author: georg-wolflein

pyproject.toml

@@ -49,6 +49,7 @@ pythonpath = ["."]
 [dependency-groups]
 dev = [
     "ipykernel>=6.29.5",
+    "matplotlib>=3.10.3",
     "pillow>=11.2.1",
     "pytest-asyncio>=0.26.0",
     "types-docker>=7.1.0.20241229",

tasks/flowmap_overfit_scene/task.yaml

@@ -4,7 +4,7 @@ repo:
   url: "https://github.com/dcharatan/flowmap"
   commit: 578a515
 papers: [smith2024flowmap]
-category: 3d_vision
+category: misc  # 3d_vision
 requires: cuda
 description: Overfit FlowMap on an input scene to determine camera extrinsics for each frame in the scene.
 arguments:

tasks/llmaix_extract_text/task.yaml

@@ -10,7 +10,7 @@ repo:
     # You may also leave this empty if the repository does not require any tokens/secrets.
     # HF_TOKEN: "${env:HF_TOKEN}"  # Example environment variable that sets HF_TOKEN to the local environment variable HF_TOKEN.
 papers: [wiest2024llm] # List of papers that this task is based on. Papers are specified by their IDs in the papers.bib file.
-category: pathology # Category of the task.
+category: llms # Category of the task.
 description: Preprocess various input files into a standardized structure - a list of the texts. The list can contain multiple texts in case of .csv and .xlsx input files. In some cases, OCR needs to be applied or even enforced to get the correct text from a pdf (or image) document. # Description of the task. This should be 1-3 sentences in length, explaining to the LLM what this tool should do.
 arguments:
   # List of arguments of the task. Each argument is a dictionary with the keys "name", "description", and "type".

tasks/mopadi_generate_counterfactuals/task.yaml

@@ -8,7 +8,7 @@ repo:
 papers: [zigutyte2024mopadi]
 category: pathology
 requires: cuda
-description: Generate counterfactual explanations for the top 3 tiles per patient by manipulating them with a specific amplitude, such that the predicted class of each counterfactual image flips to the opposite class (i.e., the predicted output for the opposite class exceeds 0.9), while avoiding excessive overmanipulation. Use a pretrained diffusion autoencoder according to the cancer type, combined with a corresponding MIL classifier trained to distinguish biologically meaningful histological patterns. You will be provided with the path to the folder containing images, the clinical table with each patient’s target label values, and the folder containing pre-extracted features
+description: Generate counterfactual explanations for the top 3 tiles per patient by manipulating them with a specific amplitude, such that the predicted class of each counterfactual image flips to the opposite class (i.e., the predicted output for the opposite class exceeds 0.9), while avoiding excessive overmanipulation. Use a pretrained diffusion autoencoder according to the cancer type, combined with a corresponding MIL classifier trained to distinguish biologically meaningful histological patterns. You will be provided with the path to the folder containing images, the clinical table with each patient's target label values, and the folder containing pre-extracted features
 arguments:
   - name: images_dir
     description: Path to the folder containing patient subfolders with image patches

tasks/retfound_feature_vector/task.yaml

@@ -6,7 +6,7 @@ repo:
   env:
     HF_TOKEN: "${env:HF_TOKEN}"  # required for downloading models
 papers: [zhou2023retfound]
-category: imaging
+category: misc  # imaging
 requires: cuda
 description: >-
   Extract the latent feature vector for the given retinal image using the RETFound pretrained RETFound_mae_natureCFP model.

tasks/tabpfn_predict/task.yaml

@@ -4,7 +4,7 @@ repo:
   url: "https://github.com/PriorLabs/TabPFN"
   commit: e8744e4
 papers: [hollmann2025tabpfn]
-category: tabular
+category: misc  # tabular
 requires: cpu
 description: Train a predictor using TabPFN on a tabular dataset. Evaluate the predictor on the test set. Train on CPU.
 arguments: