Added MedConceptsQA Benchmark (#2010)

* Added MedConceptsQA Benchmark

* pre-commit factor

* update group name

* update in naming

* changed name

* Changed mcqa to med_concepts_qa prefix

* Added med_concepts_qa to README.md

* Changed config files according the new format

* Updated README

---------
Co-authored-by: lintangsutawika <lintang@eleuther.ai>

lm_eval/tasks/README.md

@@ -62,6 +62,7 @@
 | [logiqa2](logiqa2/README.md) | Large-scale logical reasoning dataset adapted from the Chinese Civil Service Examination. | English, Chinese |
 | [mathqa](mathqa/README.md) | Question answering tasks involving mathematical reasoning and problem-solving. | English |
 | [mc_taco](mc_taco/README.md) | Question-answer pairs that require temporal commonsense comprehension. | English |
+| [med_concepts_qa](med_concepts_qa/README.md) | Benchmark for evaluating LLMs on their abilities to interpret medical codes and distinguish between medical concept. | English |
 | medmcqa | Medical multiple choice questions assessing detailed medical knowledge. | English |
 | medqa | Multiple choice question answering based on the United States Medical License Exams. | |
 | [mgsm](mgsm/README.md) | Benchmark of multilingual grade-school math problems. | Spanish, French, German, Russian, Chinese, Japanese, Thai, Swahili, Bengali, Telugu |

lm_eval/tasks/med_concepts_qa/README.md

+# MedConceptsQA
+
+### Paper
+
+Title: `MedConceptsQA: Open Source Medical Concepts QA Benchmark`
+
+Abstract: https://arxiv.org/abs/2405.07348
+
+MedConceptsQA is a dedicated open source benchmark for medical concepts question answering. The benchmark comprises of questions of various medical concepts across different vocabularies: diagnoses, procedures, and drugs.
+
+The questions are categorized into three levels of difficulty: easy, medium, and hard.
+
+Our benchmark serves as a valuable resource for evaluating the
+abilities of Large Language Models to interpret medical codes and distinguish
+between medical concepts.
+
+### Citation
+
+```
+@article{shoham2024medconceptsqa,
+  title={MedConceptsQA--Open Source Medical Concepts QA Benchmark},
+  author={Shoham, Ofir Ben and Rappoport, Nadav},
+  journal={arXiv preprint arXiv:2405.07348},
+  year={2024}
+}
+```
+
+### Groups and Tasks
+
+#### Groups
+
+* `med_concepts_qa`: Contains all the QA tasks (diagnosis, procedures ,and drugs).
+
+#### Tasks
+
+
+* `med_concepts_qa_icd9cm` - ICD9-CM (diagnosis codes, ICD9 format) question-answering. This involves providing information, clarifications, and answering questions related to ICD-9-CM (International Classification of Diseases, 9th Revision, Clinical Modification) diagnosis codes.
+
+
+* `med_concepts_qa_icd10cm` - ICD10-CM (diagnosis codes, ICD10 format) question-answering. This involves providing information, clarifications, and answering questions related to ICD-10-CM (International Classification of Diseases, 10th Revision, Clinical Modification) diagnosis codes.
+
+
+* `med_concepts_qa_icd9proc` - ICD9-Proc (procedure codes, ICD9 format) question-answering. This involves providing information, clarifications, and answering questions related to ICD-9-PCS (International Classification of Diseases, 9th Revision, Procedure Coding System) procedure codes.
+
+
+* `med_concepts_qa_icd10proc` - ICD10-Proc (procedure codes, ICD10 format) question-answering. This involves providing information, clarifications, and answering questions related to ICD-10-PCS (International Classification of Diseases, 10th Revision, Procedure Coding System) procedure codes.
+
+
+* `med_concepts_qa_atc` - ATC (Anatomical Therapeutic Chemical Classification System) question-answering. This involves providing information, clarifications, and answering questions related to the ATC classification system, which is used for the classification of drugs and other medical products according to the organ or system on which they act and their therapeutic, pharmacological, and chemical properties.

lm_eval/tasks/med_concepts_qa/_default_template_yaml

+dataset_path: ofir408/MedConceptsQA
+output_type: multiple_choice
+description: "Answer A,B,C,D according to the answer to this multiple choice question.\n"
+fewshot_split: dev
+fewshot_config:
+  sampler: first_n
+num_fewshot: 4
+test_split: test
+doc_to_text: "{{question}}\nAnswer:"
+doc_to_target: answer_id
+doc_to_choice: ['A', 'B', 'C', 'D']
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true

lm_eval/tasks/med_concepts_qa/_generate_configs.py

+from typing import List
+
+import yaml
+
+
+def generate_yaml_content(vocab_name: str, level: str):
+    content = {
+        "dataset_name": f"{vocab_name}_{level}",
+        "tag": f"med_concepts_qa_{vocab_name}_tasks",
+        "include": "_default_template_yaml",
+        "task": f"med_concepts_qa_{vocab_name}_{level}",
+        "task_alias": f"{vocab_name}_{level}",
+    }
+    return content
+
+
+def generate_yaml_files(
+    vocab_names: List[str], levels: List[str], file_name_prefix: str
+):
+    for vocab_name in vocab_names:
+        for level in levels:
+            yaml_content = generate_yaml_content(vocab_name, level)
+            filename = f"{file_name_prefix}_{vocab_name}_{level}.yaml"
+            with open(filename, "w") as yaml_file:
+                yaml.dump(yaml_content, yaml_file, default_flow_style=False)
+            print(f"Done to generated {filename}")
+
+
+if __name__ == "__main__":
+    generate_yaml_files(
+        vocab_names=["icd9cm", "icd10cm", "icd9proc", "icd10proc", "atc"],
+        levels=["easy", "medium", "hard"],
+        file_name_prefix="med_concepts_qa",
+    )

lm_eval/tasks/med_concepts_qa/_med_concepts_qa.yaml

+group: med_concepts_qa
+task:
+  - med_concepts_qa_icd9cm
+  - med_concepts_qa_icd10cm
+  - med_concepts_qa_icd9proc
+  - med_concepts_qa_icd10proc
+  - med_concepts_qa_atc
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean

lm_eval/tasks/med_concepts_qa/_med_concepts_qa_atc.yaml

+group: med_concepts_qa_atc
+task:
+  - med_concepts_qa_atc_tasks
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean
\ No newline at end of file

lm_eval/tasks/med_concepts_qa/_med_concepts_qa_icd10cm.yaml

+group: med_concepts_qa_icd10cm
+task:
+  - med_concepts_qa_icd10cm_tasks
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean

lm_eval/tasks/med_concepts_qa/_med_concepts_qa_icd10proc.yaml

+group: med_concepts_qa_icd10proc
+task:
+  - med_concepts_qa_icd10proc_tasks
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean
\ No newline at end of file

lm_eval/tasks/med_concepts_qa/_med_concepts_qa_icd9cm.yaml

+group: med_concepts_qa_icd9cm
+task:
+  - med_concepts_qa_icd9cm_tasks
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean
\ No newline at end of file

lm_eval/tasks/med_concepts_qa/_med_concepts_qa_icd9proc.yaml

+group: med_concepts_qa_icd9proc
+task:
+  - med_concepts_qa_icd9proc_tasks
+aggregate_metric_list:
+  - metric: acc
+    aggregation: mean
\ No newline at end of file

lm_eval/tasks/med_concepts_qa/med_concepts_qa_atc_easy.yaml

+dataset_name: atc_easy
+include: _default_template_yaml
+tag: med_concepts_qa_atc_tasks
+task: med_concepts_qa_atc_easy
+task_alias: atc_easy

lm_eval/tasks/med_concepts_qa/med_concepts_qa_atc_hard.yaml

+dataset_name: atc_hard
+include: _default_template_yaml
+tag: med_concepts_qa_atc_tasks
+task: med_concepts_qa_atc_hard
+task_alias: atc_hard

lm_eval/tasks/med_concepts_qa/med_concepts_qa_atc_medium.yaml

+dataset_name: atc_medium
+include: _default_template_yaml
+tag: med_concepts_qa_atc_tasks
+task: med_concepts_qa_atc_medium
+task_alias: atc_medium

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10cm_easy.yaml

+dataset_name: icd10cm_easy
+include: _default_template_yaml
+tag: med_concepts_qa_icd10cm_tasks
+task: med_concepts_qa_icd10cm_easy
+task_alias: icd10cm_easy

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10cm_hard.yaml

+dataset_name: icd10cm_hard
+include: _default_template_yaml
+tag: med_concepts_qa_icd10cm_tasks
+task: med_concepts_qa_icd10cm_hard
+task_alias: icd10cm_hard

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10cm_medium.yaml

+dataset_name: icd10cm_medium
+include: _default_template_yaml
+tag: med_concepts_qa_icd10cm_tasks
+task: med_concepts_qa_icd10cm_medium
+task_alias: icd10cm_medium

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10proc_easy.yaml

+dataset_name: icd10proc_easy
+include: _default_template_yaml
+tag: med_concepts_qa_icd10proc_tasks
+task: med_concepts_qa_icd10proc_easy
+task_alias: icd10proc_easy

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10proc_hard.yaml

+dataset_name: icd10proc_hard
+include: _default_template_yaml
+tag: med_concepts_qa_icd10proc_tasks
+task: med_concepts_qa_icd10proc_hard
+task_alias: icd10proc_hard

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd10proc_medium.yaml

+dataset_name: icd10proc_medium
+include: _default_template_yaml
+tag: med_concepts_qa_icd10proc_tasks
+task: med_concepts_qa_icd10proc_medium
+task_alias: icd10proc_medium

lm_eval/tasks/med_concepts_qa/med_concepts_qa_icd9cm_easy.yaml

+dataset_name: icd9cm_easy
+include: _default_template_yaml
+tag: med_concepts_qa_icd9cm_tasks
+task: med_concepts_qa_icd9cm_easy
+task_alias: icd9cm_easy