jaechang-hits/hypogenic-hypothesis-generation

HypoGeniC Hypothesis Generation

Overview

HypoGeniC automates scientific hypothesis generation and testing using LLMs on tabular datasets. Given labeled data (e.g., deception detection, AI-content identification), it generates testable hypotheses, iteratively refines them against validation performance, and runs inference to classify new samples. It supports three approaches: purely data-driven (HypoGeniC), literature-integrated (HypoRefine), and mechanistic union of both.

When to Use

• Generating testable hypotheses from labeled observational datasets without prior theory
• Systematically testing multiple competing hypotheses on empirical data
• Combining insights from research papers with data-driven pattern discovery
• Accelerating hypothesis ideation in domains like deception detection, content analysis, mental health indicators
• Benchmarking LLM-based hypothesis generation methods against few-shot baselines
• For manual hypothesis formulation frameworks, use hypothesis-generation knowhow
• For general-purpose ML classification without hypothesis interpretability, use scikit-learn-machine-learning

Prerequisites

• Python packages: hypogenic
• Optional: Redis server (port 6832) for LLM response caching; GROBID for PDF literature processing
• API keys: OpenAI, Anthropic, or compatible LLM API key in environment
• Data: Labeled JSON datasets in HypoGeniC format (see Key Concepts)

pip install hypogenic

# Optional: clone example datasets
git clone https://github.com/ChicagoHAI/HypoGeniC-datasets.git ./data
git clone https://github.com/ChicagoHAI/Hypothesis-agent-datasets.git ./data_lit

Quick Start

from hypogenic import BaseTask
import re

# Custom label extractor (must match dataset label format)
def extract_label(text: str) -> str:
    match = re.search(r'final answer:\s+(.*)', text, re.IGNORECASE)
    return match.group(1).strip() if match else text.strip()

# 1. Load task from config
task = BaseTask(
    config_path="./data/your_task/config.yaml",
    extract_label=extract_label
)

# 2. Generate hypotheses (data-driven)
task.generate_hypotheses(
    method="hypogenic",
    num_hypotheses=20,
    output_path="./output/hypotheses.json"
)

# 3. Run inference on test set
results = task.inference(
    hypothesis_bank="./output/hypotheses.json",
    test_data="./data/your_task/your_task_test.json"
)
print(f"Accuracy: {results['accuracy']:.3f}")

Workflow

Step 1: Prepare Dataset

Create train/val/test JSON files with text features and labels.

import json

# Dataset: each key maps to a list of equal length
dataset = {
    "headline_1": [
        "What Up, Comet? You Just Got *PROBED*",
        "Scientists Made a Breakthrough in Quantum Computing"
    ],
    "headline_2": [
        "Scientists Were Holding Their Breath Today. Here's Why.",
        "New Quantum Computer Achieves Milestone"
    ],
    "label": [
        "Headline 2 has more clicks than Headline 1",
        "Headline 1 has more clicks than Headline 2"
    ]
}

# All lists must have equal length; labels must match extract_label output
for split in ["train", "val", "test"]:
    with open(f"my_task_{split}.json", "w") as f:
        json.dump(dataset, f, indent=2)
print(f"Created dataset with {len(dataset['label'])} samples")

Step 2: Create Task Configuration

Write a config.yaml defining dataset paths and prompt templates.

# config.yaml structure (write as YAML file)
config = """
task_name: my_task

train_data_path: ./my_task_train.json
val_data_path: ./my_task_val.json
test_data_path: ./my_task_test.json

prompt_templates:
  observations: |
    Feature 1: ${text_features_1}
    Feature 2: ${text_features_2}
    Observation: ${label}

  batched_generation:
    system: "You are a research scientist generating hypotheses."
    user: "Generate ${num_hypotheses} testable hypotheses from these observations."

  inference:
    system: "You are evaluating a hypothesis against data."
    user: "Hypothesis: ${hypothesis}\\nSample: ${sample_text}\\nFinal answer: ${label}"

  is_relevant:
    system: "Check hypothesis relevance."
    user: "Is this hypothesis relevant? ${hypothesis}"
"""

with open("config.yaml", "w") as f:
    f.write(config)
print("Configuration written to config.yaml")

Step 3: Implement Label Extraction

Define a custom extract_label function matching your label format.

import re

def extract_label(llm_output: str) -> str:
    """Parse LLM output to extract predicted label.

    Must return labels matching the 'label' field values in the dataset.
    Default: searches for 'final answer: <label>' pattern.
    """
    match = re.search(r'final answer:\s+(.*)', llm_output, re.IGNORECASE)
    if match:
        return match.group(1).strip()
    # Domain-specific fallback
    if "Final prediction:" in llm_output:
        return llm_output.split("Final prediction:")[-1].strip()
    return llm_output.strip()

# Test against expected labels
assert extract_label("Final answer: Headline 1") == "Headline 1"
print("Label extractor validated")

Step 4: Generate Hypotheses (HypoGeniC)

Run data-driven hypothesis generation with iterative refinement.

from hypogenic import BaseTask

task = BaseTask(
    config_path="./config.yaml",
    extract_label=extract_label
)

# Generate hypotheses: initializes from data subset, iteratively refines
task.generate_hypotheses(
    method="hypogenic",       # Data-driven generation
    num_hypotheses=20,        # Target number of hypotheses
    output_path="./output/hypotheses.json"
)
# CLI equivalent:
# hypogenic_generation --config config.yaml --method hypogenic --num_hypotheses 20
print("Hypothesis bank saved to ./output/hypotheses.json")

Step 5: Run Inference

Test generated hypotheses against the test set.

results = task.inference(
    hypothesis_bank="./output/hypotheses.json",
    test_data="./my_task_test.json"
)
print(f"Test accuracy: {results['accuracy']:.3f}")
print(f"Predictions: {results['predictions'][:5]}")
# CLI equivalent:
# hypogenic_inference --config config.yaml --hypotheses output/hypotheses.json

Step 6: Literature-Integrated Generation (HypoRefine)

Combine literature insights with data-driven hypotheses.

# Requires GROBID setup and preprocessed PDFs
# bash ./modules/setup_grobid.sh  # first time
# bash ./modules/run_grobid.sh    # start GROBID service
# python pdf_preprocess.py --task_name my_task

task.generate_hypotheses(
    method="hyporefine",
    num_hypotheses=15,
    literature_path="./literature/my_task/",
    output_path="./output/"
)
# Generates 3 hypothesis banks:
# - HypoRefine (integrated literature+data)
# - Literature-only hypotheses
# - Literature union HypoRefine
print("HypoRefine generation complete: 3 hypothesis banks created")

Step 7: Multi-Hypothesis Inference

Test multiple hypotheses simultaneously for ensemble classification.

from examples.multi_hyp_inference import run_multi_hypothesis_inference

results = run_multi_hypothesis_inference(
    config_path="./config.yaml",
    hypothesis_bank="./output/hypotheses.json",
    test_data="./my_task_test.json"
)
print(f"Multi-hypothesis accuracy: {results['accuracy']:.3f}")

Key Parameters

| Parameter | Default | Range / Options | Effect | |-----------|---------|-----------------|--------| | method | "hypogenic" | "hypogenic", "hyporefine", "union" | Generation strategy | | num_hypotheses | 20 | 5-50 | Number of hypotheses to generate | | batch_size | 5 | 3-10 | Samples per generation batch | | max_iterations | 10 | 1-50 | Refinement iterations | | temperature | 0.7 | 0.0-1.0 | LLM sampling temperature | | confidence_threshold | 0.7 | 0.5-0.95 | Inference confidence cutoff | | num_papers | 10 | 5-30 | Papers for HypoRefine literature extraction | | inference_method | "voting" | "voting", "weighted", "ensemble" | How multiple hypotheses combine predictions |

Key Concepts

Dataset Format

HypoGeniC expects JSON files with parallel lists:

{
  "text_features_1": ["sample_1_feat1", "sample_2_feat1"],
  "text_features_2": ["sample_1_feat2", "sample_2_feat2"],
  "label": ["class_A", "class_B"]
}

• All lists must have equal length
• Feature keys are customizable (review_text, post_content, etc.)
• Labels must match the extract_label() output format exactly
• Three splits required: <TASK>_train.json, <TASK>_val.json, <TASK>_test.json

Three Generation Methods

| Method | Input | Process | Best For | |--------|-------|---------|----------| | HypoGeniC | Data only | Init from subset, iteratively refine on validation | Exploratory research, novel datasets without literature | | HypoRefine | Data + PDFs | Extract literature insights, merge with data patterns, refine both | Extending or validating existing theories | | Union | Literature + HypoGeniC | Mechanistic combination, deduplication | Maximum hypothesis diversity and coverage |

Configuration Template

Minimal required config.yaml structure:

task_name: my_task
train_data_path: ./my_task_train.json
val_data_path: ./my_task_val.json
test_data_path: ./my_task_test.json

model:
  name: "gpt-4"                    # or claude-3, gpt-3.5-turbo
  api_key_env: "OPENAI_API_KEY"
  temperature: 0.7

generation:
  method: "hypogenic"
  num_hypotheses: 20
  batch_size: 5
  max_iterations: 10

cache:
  enabled: true                    # Redis on localhost:6832
  host: "localhost"
  port: 6832

prompt_templates:
  observations: |
    Feature 1: ${text_features_1}
    Observation: ${label}
  batched_generation:
    system: "Generate testable hypotheses."
    user: "Generate ${num_hypotheses} hypotheses."
  inference:
    system: "Evaluate hypothesis against sample."
    user: "Hypothesis: ${hypothesis}\nSample: ${sample_text}"
  is_relevant:
    system: "Check relevance."
    user: "Is ${hypothesis} relevant?"

Common Recipes

Recipe: Custom Task from Scratch

When to use: creating a new classification task with domain-specific data.

import json
from hypogenic import BaseTask

# 1. Prepare data splits
for split_name, data in [("train", train_data), ("val", val_data), ("test", test_data)]:
    with open(f"my_task_{split_name}.json", "w") as f:
        json.dump(data, f)

# 2. Define domain-specific label extractor
def my_extractor(text):
    if "positive" in text.lower():
        return "positive"
    elif "negative" in text.lower():
        return "negative"
    return text.strip()

# 3. Create task and run full pipeline
task = BaseTask(config_path="./my_task/config.yaml", extract_label=my_extractor)
task.generate_hypotheses(method="hypogenic", num_hypotheses=15, output_path="./output/")
results = task.inference(hypothesis_bank="./output/hypotheses.json")
print(f"Custom task accuracy: {results['accuracy']:.3f}")

Recipe: Literature Processing Setup

When to use: setting up GROBID for PDF-to-structured-text conversion before HypoRefine.

# 1. Setup GROBID (first time only)
bash ./modules/setup_grobid.sh

# 2. Place PDFs in literature directory
mkdir -p literature/my_task/raw/
cp papers/*.pdf literature/my_task/raw/

# 3. Start GROBID and process
bash ./modules/run_grobid.sh
cd examples && python pdf_preprocess.py --task_name my_task
# Output: structured text files in literature/my_task/processed/

Recipe: Union Method for Maximum Coverage

When to use: combining literature and data-driven hypotheses for comprehensive coverage.

# Generate literature hypotheses first (via HypoRefine)
task.generate_hypotheses(
    method="hyporefine",
    num_hypotheses=15,
    literature_path="./literature/my_task/",
    output_path="./output/"
)

# Union combines and deduplicates both banks
# CLI alternative:
# hypogenic_generation --config config.yaml --method union \
#   --literature_hypotheses output/lit_hypotheses.json

# Compare all three approaches
for bank in ["hypogenic", "hyporefine", "union"]:
    r = task.inference(hypothesis_bank=f"./output/{bank}_hypotheses.json")
    print(f"{bank}: accuracy={r['accuracy']:.3f}")

Expected Outputs

• hypotheses.json -- hypothesis bank with ranked, testable hypotheses (typically 10-20)
• Inference results with per-sample predictions and overall accuracy
• For HypoRefine: three hypothesis banks (literature-only, integrated, union)
• Reported improvements: ~9% over few-shot baselines, ~16% over literature-only approaches
• 80-84% hypothesis pair diversity (non-redundant insights)

Troubleshooting

| Problem | Cause | Solution | |---------|-------|----------| | ModuleNotFoundError: hypogenic | Package not installed | pip install hypogenic | | Generic/untestable hypotheses | Prompt templates too vague | Add domain-specific context to batched_generation prompt | | Poor inference accuracy | Few training examples or bad label extraction | Increase training data; verify extract_label matches dataset labels | | GROBID PDF processing fails | GROBID service not running | bash ./modules/run_grobid.sh; ensure PDFs are valid papers | | Label extraction mismatches | extract_label output differs from dataset labels | Print both formats and align; test with assert extract_label(sample) == expected | | Redis connection errors | Redis not running or wrong port | Start Redis on port 6832 or set cache.enabled: false | | API rate limit errors | Too many concurrent LLM calls | Reduce batch_size; enable Redis caching to avoid duplicate calls | | Empty hypothesis bank | Config missing required prompt templates | Include all four templates: observations, batched_generation, inference, is_relevant |

Bundled Resources

This entry is self-contained. The original references/config_template.yaml (151 lines) has been consolidated into the Key Concepts "Configuration Template" subsection, retaining the essential YAML structure, model/cache/generation parameters, and prompt template patterns. Omitted from the template: evaluation metrics block, logging configuration, task-specific feature/label metadata descriptions -- these are standard YAML patterns users can add as needed.

Related Skills

• hypothesis-generation -- knowhow for manual hypothesis formulation frameworks and scientific method
• scikit-learn-machine-learning -- classical ML for classification when hypothesis interpretability is not needed
• pubmed-database -- literature search to find papers for HypoRefine input

References

• Zhou, Y., Liu, H., Srivastava, T., Mei, H., & Tan, C. (2024). "Hypothesis Generation with Large Language Models." EMNLP Workshop NLP for Science. https://aclanthology.org/2024.nlp4science-1.10/
• Liu, H., Zhou, Y., Li, M., Yuan, C., & Tan, C. (2024). "Literature Meets Data: A Synergistic Approach to Hypothesis Generation." arXiv:2410.17309. https://arxiv.org/abs/2410.17309
• Liu, H., Huang, S., Hu, J., Zhou, Y., & Tan, C. (2025). "HypoBench: Towards Systematic and Principled Benchmarking for Hypothesis Generation." arXiv:2504.11524. https://arxiv.org/abs/2504.11524
• GitHub Repository: https://github.com/ChicagoHAI/hypothesis-generation
• PyPI Package: https://pypi.org/project/hypogenic/