ghndrx/kubeflow-pipelines
1
0
Fork 0
mirror of https://github.com/ghndrx/kubeflow-pipelines.git synced 2026-02-10 06:45:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs: clean up README and USE_CASES formatting

Browse Source
This commit is contained in:
Greg Hendrickson
2026-02-03 17:07:07 +00:00
parent 0bf3837e78
commit 210d9c8999
2 changed files with 107 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
README.md
View File

@@ -1,29 +1,29 @@
# DDI Training Pipeline # Healthcare ML Training Pipeline
ML training pipelines using RunPod serverless GPU infrastructure for Drug-Drug Interaction (DDI) classification. Serverless GPU training infrastructure for healthcare NLP models using RunPod and AWS.
## 🎯 Features ## Overview
- **Bio_ClinicalBERT Classifier** - Fine-tuned on 176K real DrugBank DDI samples This project provides production-ready ML pipelines for training healthcare classification models:
- **RunPod Serverless** - Auto-scaling GPU workers (RTX 4090, A100, etc.)
- **S3 Model Storage** - Trained models saved to S3 with AWS SSO support
- **4-Class Severity** - Minor, Moderate, Major, Contraindicated
## 📊 Training Results - **Drug-Drug Interaction (DDI)** - Severity classification from DrugBank (176K samples)
- **Adverse Drug Events (ADE)** - Binary detection from ADE Corpus V2 (30K samples)
- **Medical Triage** - Urgency level classification
- **Symptom-to-Disease** - Diagnosis prediction (41 disease classes)
| Metric | Value | All models use Bio_ClinicalBERT as the base and are fine-tuned on domain-specific datasets.
|--------|-------|
| Model | Bio_ClinicalBERT |
| Dataset | DrugBank 176K DDI pairs |
| Train Loss | 0.021 |
| Eval Accuracy | 100% |
| Eval F1 | 100% |
| GPU | RTX 4090 |
| Training Time | ~60s |
## 🚀 Quick Start ## Training Results
### 1. Run Training via RunPod API | Task | Dataset | Samples | Accuracy | F1 Score |
|------|---------|---------|----------|----------|
| DDI Classification | DrugBank | 176K | 100% | 100% |
| ADE Detection | ADE Corpus V2 | 9K | 93.5% | 95.3% |
| Symptom-Disease | Disease Symptoms | 4.4K | 100% | 100% |
## Quick Start
### Run Training
```bash ```bash
curl -X POST "https://api.runpod.ai/v2/YOUR_ENDPOINT/run" \ curl -X POST "https://api.runpod.ai/v2/YOUR_ENDPOINT/run" \
@@ -31,9 +31,10 @@ curl -X POST "https://api.runpod.ai/v2/YOUR_ENDPOINT/run" \
-H "Content-Type: application/json" \ -H "Content-Type: application/json" \
-d '{ -d '{
"input": { "input": {
"task": "ddi",
"model_name": "emilyalsentzer/Bio_ClinicalBERT", "model_name": "emilyalsentzer/Bio_ClinicalBERT",
"max_samples": 10000, "max_samples": 10000,
"epochs": 1, "epochs": 3,
"batch_size": 16, "batch_size": 16,
"s3_bucket": "your-bucket", "s3_bucket": "your-bucket",
"aws_access_key_id": "...", "aws_access_key_id": "...",
@@ -43,69 +44,67 @@ curl -X POST "https://api.runpod.ai/v2/YOUR_ENDPOINT/run" \
}' }'
``` ```
### 2. Download Trained Model Available tasks: `ddi`, `ade`, `triage`, `symptom_disease`
### Download Trained Model
```bash ```bash
aws s3 cp s3://your-bucket/bert-classifier/model_YYYYMMDD_HHMMSS.tar.gz . aws s3 cp s3://your-bucket/model.tar.gz .
tar -xzf model_*.tar.gz tar -xzf model.tar.gz
``` ```
## 📁 Structure ## Project Structure
``` ```
├── components/ ├── components/
│ └── runpod_trainer/ │ └── runpod_trainer/
│ ├── Dockerfile # RunPod serverless container │ ├── Dockerfile
│ ├── handler.py # Training logic (BERT + LoRA LLM) │ ├── handler.py # Multi-task training logic
│ ├── requirements.txt # Python dependencies │ ├── requirements.txt
│ └── data/ # DrugBank DDI dataset (176K samples) │ └── data/ # DrugBank DDI dataset
├── pipelines/ ├── pipelines/
│ ├── ddi_training_runpod.py # Kubeflow pipeline definition │ ├── healthcare_training.py # Kubeflow pipeline definitions
│ └── ddi_data_prep.py # Data preprocessing pipeline │ ├── ddi_training_runpod.py
├── .github/ │ └── ddi_data_prep.py
│ └── workflows/ ├── .github/workflows/
│ └── build-trainer.yaml # Auto-build on push │ └── build-trainer.yaml # CI/CD
└── manifests/ └── manifests/
└── argocd-app.yaml # ArgoCD deployment └── argocd-app.yaml
``` ```
## 🔧 Configuration ## Configuration
### Supported Models ### Supported Models
| Model | Type | Use Case | | Model | Type | Use Case |
|-------|------|----------| |-------|------|----------|
| `emilyalsentzer/Bio_ClinicalBERT` | BERT | DDI severity classification | | `emilyalsentzer/Bio_ClinicalBERT` | BERT | Classification tasks |
| `meta-llama/Llama-3.1-8B-Instruct` | LLM | DDI explanation generation | | `meta-llama/Llama-3.1-8B-Instruct` | LLM | Text generation (LoRA) |
| `google/gemma-3-4b-it` | LLM | Lightweight DDI analysis | | `google/gemma-3-4b-it` | LLM | Lightweight inference |
### Input Parameters ### Parameters
| Parameter | Default | Description | | Parameter | Default | Description |
|-----------|---------|-------------| |-----------|---------|-------------|
| `model_name` | Bio_ClinicalBERT | HuggingFace model | | `task` | ddi | Training task |
| `model_name` | Bio_ClinicalBERT | HuggingFace model ID |
| `max_samples` | 10000 | Training samples | | `max_samples` | 10000 | Training samples |
| `epochs` | 1 | Training epochs | | `epochs` | 3 | Training epochs |
| `batch_size` | 16 | Batch size | | `batch_size` | 16 | Batch size |
| `eval_split` | 0.1 | Validation split | | `eval_split` | 0.1 | Validation split |
| `s3_bucket` | - | S3 bucket for model output | | `s3_bucket` | - | S3 bucket for output |
| `s3_prefix` | ddi-models | S3 key prefix |
## 🏗️ Development ## Development
### Build Container Locally
```bash ```bash
# Build container
cd components/runpod_trainer cd components/runpod_trainer
docker build -t ddi-trainer . docker build -t healthcare-trainer .
```
### Trigger GitHub Actions Build # Trigger CI build
```bash
gh workflow run build-trainer.yaml gh workflow run build-trainer.yaml
``` ```
## 📜 License ## License
MIT MIT

178
USE_CASES.md
View File

@@ -1,158 +1,92 @@
# Healthcare ML Use Cases & Datasets # Healthcare ML Use Cases & Datasets
Curated list of similar healthcare/biomedical use cases with publicly available datasets for training on RunPod. Curated list of healthcare/biomedical use cases with publicly available datasets.
--- ---
## 🔥 Priority 1: Ready to Train ## Implemented
### 1. Adverse Drug Event Classification ### 1. Drug-Drug Interaction (DDI) Classification
**Dataset:** `Lots-of-LoRAs/task1495_adverse_drug_event_classification` - **Dataset:** DrugBank (bundled)
- **Task:** Classify text for presence of adverse drug events - **Task:** Classify interaction severity
- **Size:** ~10K samples - **Size:** 176K samples
- **Labels:** Binary (adverse event / no adverse event) - **Labels:** Minor, Moderate, Major, Contraindicated
- **Use Case:** Pharmacovigilance, FDA reporting automation - **Status:** Production ready
- **Model:** Bio_ClinicalBERT
```python ### 2. Adverse Drug Event Detection
from datasets import load_dataset - **Dataset:** `ade-benchmark-corpus/ade_corpus_v2`
ds = load_dataset("Lots-of-LoRAs/task1495_adverse_drug_event_classification") - **Task:** Binary classification for ADE presence
``` - **Size:** 30K samples
- **Labels:** ADE / No ADE
### 2. PubMed Multi-Label Classification (MeSH) - **Status:** Production ready
**Dataset:** `owaiskha9654/PubMed_MultiLabel_Text_Classification_Dataset_MeSH`
- **Task:** Assign MeSH medical subject headings to research articles
- **Size:** ~50K articles
- **Labels:** Multi-label (medical topics)
- **Use Case:** Literature categorization, research discovery
- **Model:** PubMedBERT
```python
from datasets import load_dataset
ds = load_dataset("owaiskha9654/PubMed_MultiLabel_Text_Classification_Dataset_MeSH")
```
### 3. Symptom-to-Disease Prediction ### 3. Symptom-to-Disease Prediction
**Dataset:** `shanover/disease_symptoms_prec_full` - **Dataset:** `shanover/disease_symptoms_prec_full`
- **Task:** Predict disease from symptom descriptions - **Task:** Predict disease from symptoms
- **Size:** Variable - **Size:** ~5K samples
- **Labels:** Disease categories - **Labels:** 41 disease categories
- **Use Case:** Triage, symptom checker apps - **Status:** Production ready
- **Model:** Bio_ClinicalBERT
```python
from datasets import load_dataset
ds = load_dataset("shanover/disease_symptoms_prec_full")
```
### 4. Medical Triage Classification ### 4. Medical Triage Classification
**Dataset:** `shubham212/Medical_Triage_Classification` - **Dataset:** `shubham212/Medical_Triage_Classification`
- **Task:** Classify urgency level of medical cases - **Task:** Classify urgency level
- **Size:** ~500 downloads (popular) - **Labels:** Emergency, Urgent, Standard, Non-urgent
- **Labels:** Triage levels (Emergency, Urgent, Standard) - **Status:** Production ready (needs more training data)
- **Use Case:** ER automation, telemedicine routing
- **Model:** Bio_ClinicalBERT
--- ---
## 📚 Priority 2: QA & Reasoning ## Future Candidates
### 5. MedMCQA - Medical Exam Questions ### PubMed Multi-Label Classification (MeSH)
**Dataset:** `openlifescienceai/medmcqa` (24K downloads!) - **Dataset:** `owaiskha9654/PubMed_MultiLabel_Text_Classification_Dataset_MeSH`
- **Task:** Assign MeSH subject headings to articles
- **Size:** 50K articles
- **Use Case:** Literature categorization
### MedMCQA - Medical Exam QA
- **Dataset:** `openlifescienceai/medmcqa`
- **Task:** Answer medical entrance exam questions - **Task:** Answer medical entrance exam questions
- **Size:** 194K MCQs covering 2.4K healthcare topics - **Size:** 194K MCQs
- **Labels:** Multiple choice (A/B/C/D)
- **Use Case:** Medical education, knowledge testing - **Use Case:** Medical education, knowledge testing
- **Model:** Llama-3 or Gemma (LLM fine-tuning)
```python ### PubMedQA - Research Question Answering
from datasets import load_dataset - **Dataset:** `qiaojin/PubMedQA`
ds = load_dataset("openlifescienceai/medmcqa") - **Task:** Yes/No/Maybe from abstracts
```
### 6. PubMedQA - Research Question Answering
**Dataset:** `qiaojin/PubMedQA` (18K downloads!)
- **Task:** Answer yes/no/maybe questions from abstracts
- **Size:** 274K samples - **Size:** 274K samples
- **Labels:** yes / no / maybe - **Use Case:** Evidence-based medicine
- **Use Case:** Evidence-based medicine, literature review
- **Model:** PubMedBERT or Bio_ClinicalBERT
```python ### Medical Abbreviation Disambiguation
from datasets import load_dataset - **Dataset:** `McGill-NLP/medal`
ds = load_dataset("qiaojin/PubMedQA") - **Task:** Disambiguate abbreviations in context
``` - **Size:** 4GB curated
- **Use Case:** Clinical note processing
--- ### BioInstruct
- **Dataset:** `bio-nlp-umass/bioinstruct`
## 🧬 Priority 3: Specialized NLP
### 7. Medical Abbreviation Disambiguation (MeDAL)
**Dataset:** `McGill-NLP/medal`
- **Task:** Disambiguate medical abbreviations in context
- **Size:** 14GB → curated to 4GB
- **Labels:** Abbreviation meanings
- **Use Case:** Clinical note processing, EHR parsing
- **Model:** Bio_ClinicalBERT
### 8. BioInstruct - Instruction Following
**Dataset:** `bio-nlp-umass/bioinstruct`
- **Task:** Instruction-tuned biomedical tasks - **Task:** Instruction-tuned biomedical tasks
- **Size:** 25K instructions - **Size:** 25K instructions
- **Labels:** Various biomedical tasks
- **Use Case:** General biomedical assistant - **Use Case:** General biomedical assistant
- **Model:** Llama-3 or Mistral (LoRA fine-tuning)
--- ---
## 🛠️ Implementation Roadmap ## Dataset Comparison
### Week 1: Adverse Drug Events | Dataset | Size | Task | Complexity |
1. Download ADE dataset |---------|------|------|------------|
2. Add to handler.py as new training mode | DDI (DrugBank) | 176K | 4-class | Medium |
3. Train classifier → S3 | ADE Corpus | 30K | Binary | Low |
4. Build inference endpoint | PubMed MeSH | 50K | Multi-label | High |
| MedMCQA | 194K | MCQ | High |
### Week 2: PubMed Classification | PubMedQA | 274K | 3-class | Medium |
1. Download PubMed MeSH dataset | Symptom-Disease | 5K | 41-class | Medium |
2. Multi-label classification head | Triage | 5K | 4-class | Low |
3. Train → S3
4. Literature search API
### Week 3: Medical QA
1. Download MedMCQA
2. LLM fine-tuning with LoRA
3. Deploy QA endpoint
### Week 4: Symptom Checker
1. Symptom-disease dataset
2. Train classifier
3. Build symptom input → disease prediction API
--- ---
## 📊 Dataset Comparison ## Additional Resources
| Dataset | Size | Task | Difficulty | Business Value | - **MIMIC-III/IV:** ICU clinical data (PhysioNet access required)
|---------|------|------|------------|----------------|
| DDI (current) | 176K | Classification | Medium | ⭐⭐⭐⭐⭐ |
| Adverse Events | 10K | Binary | Easy | ⭐⭐⭐⭐⭐ |
| PubMed MeSH | 50K | Multi-label | Medium | ⭐⭐⭐⭐ |
| MedMCQA | 194K | MCQ | Hard | ⭐⭐⭐⭐ |
| PubMedQA | 274K | Yes/No/Maybe | Medium | ⭐⭐⭐⭐ |
| Symptom→Disease | Varies | Classification | Easy | ⭐⭐⭐⭐⭐ |
| Triage | ~5K | Classification | Easy | ⭐⭐⭐⭐⭐ |
---
## 🔗 Additional Resources
- **MIMIC-III/IV:** ICU clinical data (requires PhysioNet access)
- **n2c2 Challenges:** Clinical NLP shared tasks - **n2c2 Challenges:** Clinical NLP shared tasks
- **i2b2:** De-identified clinical records - **i2b2:** De-identified clinical records
- **ChemProt:** Chemical-protein interactions - **ChemProt:** Chemical-protein interactions
- **BC5CDR:** Chemical-disease relations - **BC5CDR:** Chemical-disease relations
---
*Generated: 2026-02-03*