+ 0
+
+
+
+ Patients & Events
+ pyhealth.data +How a longitudinal record is modeled: build Event and Patient objects and query a history with get_events() filters.
PyHealth Tutorials
+Learn PyHealth in Colab
+Runnable notebooks that walk through each PyHealth module, then put them together into complete clinical ML pipelines. Most run on public or synthetic data, so you can start without credentialed access.
+ +Core API Tutorials
+One notebook per module, in the order data flows through PyHealth: data → datasets → tasks → models → trainer → metrics.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ 1
+
+
+
+ Loading Datasets
+ pyhealth.datasets +Load MIMIC-III and COVID19-CXR, explore events efficiently, and see how to define and contribute your own dataset with a YAML config.
+ +
+ 2
+
+
+
+ Defining Tasks
+ pyhealth.tasks +Write your own BaseTask to turn a dataset into model-ready samples, with medical coding and chest X-ray examples.
+ 3
+
+
+
+ Models: RNN Deep Dive
+ pyhealth.models +The BaseModel contract, synthetic data with create_sample_dataset(), and how RNN and MultimodalRNN work inside.
+ 4
+
+
+
+ Training End to End
+ pyhealth.trainer +Train an RNN for mortality prediction on synthetic MIMIC-III (no credentials needed): splits, early stopping, checkpoints, and inference.
+ +
+ 5
+
+
+
+ Evaluation Metrics
+ pyhealth.metrics +Binary, multiclass, and multilabel metrics, plus fairness metrics such as disparate impact, and how to get raw predictions from the trainer.
+ +
+ 6
+
+
+
+ Tokenizing Codes
+ pyhealth.tokenizer +Build a Vocabulary and Tokenizer, then encode flat code lists (2D) and visit-level histories (3D) with padding and truncation.
+ 7
+
+
+
+ Medical Code Ontologies
+ pyhealth.medcode +Look up ICD, ATC, and CCS codes, walk code hierarchies with InnerMap, and translate between systems with CrossMap.
The videos are recordings from PyHealth Live sessions on the earlier 1.x API. The concepts carry over, but follow the notebooks for current code.
+End-to-End Pipelines
+Complete workflows that load a dataset, define a task, train a model, and evaluate it. Each one runs on public Kaggle data, an open MIMIC demo, or synthetic MIMIC-III, so no credentialed access is needed to try them.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ Chest X-ray Classification
+ Image +Classify chest X-rays from the public COVID-19 Radiography Database, from download through training and evaluation.
+COVID19CXRDatasetTorchvisionModel
+
+ Medical Coding
+ Text +Predict ICD-9 codes from clinical notes in synthetic MIMIC-III, fine-tuning Bio_ClinicalBERT.
+MIMIC3DatasetTransformersModel
+
+ Medical Transcription Classification
+ Text +Classify Kaggle medical transcriptions by specialty with a pretrained transformer.
+MedicalTranscriptionsDatasetTransformersModel
+
+ Mortality Prediction
+ EHR +Predict mortality on the open MIMIC-III demo from diagnosis, procedure, and drug codes across visits.
+MIMIC3DatasetRNN
+
+ Readmission Prediction
+ EHR +Predict 30-day readmission on the public MIMIC-IV demo with Readmission30DaysMIMIC4.
MIMIC4DatasetRNNGo Further
+
+
+
+
+
+
+
+
+ Multimodal & Smart Processors
+Put a HuggingFace tokenizer inside TupleTimeTextProcessor and train MLP, Transformer, and RNN models on mixed codes and clinical text.
Example Scripts by Task
+Mortality, readmission, drug recommendation, sleep staging, chest X-ray, interpretability, and more. Not every script has been updated for PyHealth 2.0 yet.
+ +Getting Data Access
+Which datasets are public, which need PhysioNet credentialing, and how to download each one.
+ +