crfm-helm 0.5.4__py3-none-any.whl → 0.5.6__py3-none-any.whl

helm/benchmark/scenarios/shc_bmt_scenario.py

@@ -0,0 +1,75 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCBMTMedScenario(Scenario):
+    """
+    This benchmark dataset was built from a patient status gold-standard
+    for specific questions asked after a bone marrow transplant has taken place.
+    """
+
+    name = "shc_bmt_med"
+    description = (
+        "BMT-Status is a benchmark composed of clinical notes and associated binary questions"
+        "related to bone marrow transplant (BMT), hematopoietic stem cell transplant (HSCT),"
+        "or hematopoietic cell transplant (HCT) status. The goal is to determine whether the"
+        "patient received a subsequent transplant based on the provided clinical documentation."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    f"Provide an answer to the following question: {question} with the following context: {context} "
+                    ", Answer the question with a 'A' for yes or 'B' for no. Do not provide any additional "
+                    "details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCBMTMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCBMTMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCBMTMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_cdi_scenario.py

@@ -0,0 +1,75 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCCDIMedScenario(Scenario):
+    """
+    This benchmark dataset was built from Clinical Document Integrity (CDI)
+    notes were there are verifications of clinical activities. The idea behind
+    it was to assess an LLM capability to answer these questions from previous notes.
+    """
+
+    name = "shc_cdi_med"
+    description = (
+        "CDI-QA is a benchmark constructed from Clinical Documentation Integrity (CDI)"
+        "notes. It is used to evaluate a model's ability to verify clinical conditions based on"
+        "documented evidence in patient records."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    f"Provide an answer to the following question: {question} with the following context: {context} , "
+                    "Answer the question with either 'A' for yes or 'B' for no. Do not provide any "
+                    "additional details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCCDIMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCCDIMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCCDIMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_conf_scenario.py

@@ -0,0 +1,76 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCCONFMedScenario(Scenario):
+    """
+    Benchmark derived from extracting confidential information from clinical notes.
+    From Evaluation of a Large Language Model to Identify Confidential Content in
+    Adolescent Encounter Notes published at https://jamanetwork.com/journals/jamapediatrics/fullarticle/2814109
+    """
+
+    name = "shc_conf_med"
+    description = (
+        "MedConfInfo is a benchmark comprising clinical notes from adolescent patients. It is"
+        "used to evaluate whether the content contains sensitive protected health information"
+        "(PHI) that should be restricted from parental access, in accordance with adolescent"
+        "confidentiality policies in clinical care."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    f"Provide an answer to the following question: {question} with the following context: {context} "
+                    ", Answer the question with a 'A' for yes or 'B' for no. Do not provide any additional "
+                    "details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCCONFMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCCONFMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCCONFMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_ent_scenario.py

@@ -0,0 +1,77 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCENTMedScenario(Scenario):
+    """
+    This benchmark dataset was built to assess the capabilities "
+    "of an LLM for referral to the Ear, Nose and Throat department.
+    """
+
+    name = "shc_ent_med"
+    description = (
+        "ENT-Referral is a benchmark designed to evaluate whether a patient's clinical note"
+        "supports a referral to an Ear, Nose, and Throat (ENT) specialist. It helps assess"
+        "models' abilities to make referral decisions based on unstructured clinical text."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B", "C"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        counter = 1
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                if row["label"] != "":  # skip rows with character/encoding issues - 79
+                    question = row["prompt"]
+                    context = row["context"]
+                    answer = row["label"]
+                    prompt = (
+                        f"{counter} Provide an answer to the following question: {question} with the following context:"
+                        f" {context} , Answer the question with either 'A' for yes, 'B' for no, or 'C' for no mention."
+                        " Do not provide any additional details or response, just a simple A, B, or C response."
+                    )
+                    data[prompt] = answer
+                    counter = counter + 1
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCENTMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCENTMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCENTMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_gip_scenario.py

@@ -0,0 +1,74 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCGIPMedScenario(Scenario):
+    """
+    This benchmark dataset was built from a patient referral gold-standard set
+    to a specialty clinic to verify the ability of LLMs for patient hospice referral purposes.
+    """
+
+    name = "shc_gip_med"
+    description = (
+        "HospiceReferral is a benchmark that evaluates model performance in identifying"
+        "whether patients are eligible for hospice care based on palliative care clinical notes."
+        "The benchmark focuses on end-of-life care referral decisions."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    f"Provide an answer to the following question: {question} with the following context: {context} "
+                    ", Answer the question with a 'A' for yes or 'B' for no. Do not provide any additional "
+                    "details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCGIPMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCGIPMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCGIPMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_privacy_scenario.py

@@ -0,0 +1,78 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCPRIVACYMedScenario(Scenario):
+    """
+    This dataset features messages sent generated by an LLM from patient clinical notes data.
+    The scenario evaluates the ability of an LLM to determine if any potentially confidential
+    information about the patient was included. From publication: https://doi.org/10.1001/jamapediatrics.2024.4438
+    """
+
+    name = "shc_privacy_med"
+    description = (
+        "PrivacyDetection is a benchmark composed of patient portal messages submitted by"
+        "patients or caregivers. The task is to determine whether the message contains any"
+        "confidential or privacy-leaking information that should be protected."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    "You are reviewing clinical messages in order to determine if any confidential "
+                    f"information was included. Please answer the following question: {question} with the "
+                    f"following context: {context} , Answer the question with a 'A' for yes or 'B' "
+                    "for no. Do not provide any additional details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(
+            self.data_path, msg=f"[SHCPRIVACYMedScenario] Required data file not found: '{self.data_path}'"
+        )
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCPRIVACYMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCPRIVACYMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_proxy_scenario.py

@@ -0,0 +1,76 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCPROXYMedScenario(Scenario):
+    """
+    This dataset features messages sent by proxy users and non proxy users, for evaluation of
+    LLM capabilities to determine the sender. From publication: https://doi.org/10.1001/jamapediatrics.2024.4438
+    """
+
+    name = "shc_proxy_med"
+    description = (
+        "ProxySender is a benchmark composed of patient portal messages received by clinicians."
+        "It evaluates whether the message was sent by the patient or by a proxy user (e.g., parent,"
+        "spouse), which is critical for understanding who is communicating with healthcare"
+        "providers."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    "You are reviewing a clinical messages in order to determine if they have been "
+                    f"sent by a proxy user. Please determine the following: {question} with the "
+                    f"following context: {context} , Answer the question with a 'A' for yes or 'B' "
+                    "for no. Do not provide any additional details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCPROXYMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCPROXYMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCPROXYMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances

helm/benchmark/scenarios/shc_ptbm_scenario.py

@@ -0,0 +1,81 @@
+import sys
+import csv
+from typing import Dict, List
+
+from helm.benchmark.scenarios.scenario import (
+    Input,
+    Scenario,
+    Instance,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Reference,
+    Output,
+)
+from helm.common.general import check_file_exists
+
+csv.field_size_limit(sys.maxsize)
+
+
+class SHCPTBMMedScenario(Scenario):
+    """
+    This dataset contains clinical notes from primary care visit encounters of
+    children ages 4-6 years old with ADHD seen at Stanford's community-based primary
+    care network, Packard Children's Health Alliance, between 2015-2019. In this classification
+    task, the LLM is tasked with classifying whether the note contains clinician recommendation
+    for parent training in behavior management, which is the first-line evidence-based treatment
+    for young children with ADHD. From publication: https://doi.org/10.1093/jamia/ocae001
+    """
+
+    name = "shc_ptbm_med"
+    description = (
+        "ADHD-Behavior is a benchmark that evaluates a model’s ability to detect whether"
+        "a clinician recommends parent training in behavior management, an evidence-based"
+        "first-line treatment for young children diagnosed with ADHD. Each instance includes"
+        "a clinical note from a pediatric visit and a binary classification task."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B"]
+
+    def __init__(self, data_path: str):
+        super().__init__()
+        self.data_path = data_path
+
+    def create_benchmark(self, csv_path) -> Dict[str, str]:
+        data = {}
+        with open(csv_path, "r") as file:
+            reader = csv.DictReader(file)
+            for row in reader:
+                question = row["prompt"]
+                context = row["context"]
+                answer = row["label"]
+                prompt = (
+                    "You are reviewing a clinical note from health records of children with "
+                    "attention deficit hyperactivity disorder (ADHD) and classifying mentions of "
+                    f"behavioral therapy. Provide an answer to the following question: {question} with the "
+                    f"following context: {context} , Answer the question with a 'A' for yes or 'B' "
+                    "for no. Do not provide any additional details or response, just a simple A or B response."
+                )
+                data[prompt] = answer
+        return data
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        check_file_exists(self.data_path, msg=f"[SHCPTBMMedScenario] Required data file not found: '{self.data_path}'")
+        instances: List[Instance] = []
+        benchmark_data = self.create_benchmark(self.data_path)
+
+        for prompt, answer in benchmark_data.items():
+            assert answer in SHCPTBMMedScenario.POSSIBLE_ANSWER_CHOICES
+            references: List[Reference] = [
+                Reference(Output(text=pred_answer), tags=[CORRECT_TAG] if pred_answer == answer else [])
+                for pred_answer in SHCPTBMMedScenario.POSSIBLE_ANSWER_CHOICES
+            ]
+            instances.append(
+                Instance(
+                    input=Input(text=prompt),
+                    references=references,  # [Reference(Output(text=answer), tags=[CORRECT_TAG])],
+                    split=TEST_SPLIT,
+                )
+            )
+
+        return instances