<!-- These are examples of badges you might want to add to your README:
please update the URLs accordingly
[](https://anaconda.org/conda-forge/temporai)
[](https://pepy.tech/project/temporai)
-->
<!-- exclude_docs -->
[](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial04_prediction.ipynb)
[](https://temporai.readthedocs.io/en/latest/?badge=latest)
[](https://www.python.org/downloads/release/python-370/)
[](https://pypi.org/project/temporai/)
[](https://pepy.tech/project/temporai)
[](https://github.com/vanderschaarlab/temporai/tree/main/LICENSE.txt)
[](https://github.com/vanderschaarlab/temporai/actions/workflows/test.yml)
[](https://github.com/vanderschaarlab/temporai/actions/workflows/test.yml)
[](https://codecov.io/gh/vanderschaarlab/temporai)
[](https://arxiv.org/abs/2301.12260)
[](https://join.slack.com/t/vanderschaarlab/shared_invite/zt-1u2rmhw06-sHS5nQDMN3Ka2Zer6sAU6Q)
[](https://www.vanderschaar-lab.com/)
<!-- exclude_docs_end -->
# <img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/TemporAI_Logo_Icon.png' height=25> TemporAI
<!-- exclude_docs -->
> **⚗️ Status:** This project is still in *alpha*, and the API may change without warning.
<!-- exclude_docs_end -->
<!-- include_docs
:::{important}
**Status:** This project is still in *alpha*, and the API may change without warning.
:::
include_docs_end -->
## 📃 Overview
*TemporAI* is a Machine Learning-centric time-series library for medicine. The tasks that are currently of focus in TemporAI are: time-to-event (survival) analysis with time-series data, treatment effects (causal inference) over time, and time-series prediction. Data preprocessing methods, including missing value imputation for static and temporal covariates, are provided. AutoML tools for hyperparameter tuning and pipeline selection are also available.
### How is TemporAI unique?
* **🏥 Medicine-first:** Focused on use cases for medicine and healthcare, such as temporal treatment effects, survival analysis over time, imputation methods, models with built-in and post-hoc interpretability, ... See [methods](https://github.com/vanderschaarlab/temporai/tree/main/#-methods).
* **🏗️ Fast prototyping:** A plugin design allowing for on-the-fly integration of new methods by the users.
* **🚀 From research to practice:** Relevant novel models from research community adapted for practical use.
* **🌍 A healthcare ecosystem vision:** A range of interactive demonstration apps, new medical problem settings, interpretability tools, data-centric tools etc. are planned.
### Key concepts
<div align="center">
<!-- exclude_docs -->
<img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/Conceptual.png' alt="key concepts">
<!-- exclude_docs_end -->
<!-- include_docs
<img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/Conceptual.png' alt="key concepts">
include_docs_end -->
</div>
## 🚀 Installation
### Instal with `pip`
From [the Python Package Index (PyPI)](https://pypi.org/):
```bash
$ pip install temporai
```
Or from source:
```bash
$ git clone https://github.com/vanderschaarlab/temporai.git
$ cd temporai
$ pip install .
```
### Install in a [conda](https://docs.conda.io/en/latest/) environment
While have not yet published TemporAI on `conda-forge`, you can still install TemporAI in your conda environment using `pip` as follows:
Create and activate conda environment as normal:
```bash
$ conda create -n <my_environment>
$ conda activate <my_environment>
```
Then install inside your `conda` environment with pip:
```bash
$ pip install temporai
```
## 💥 Sample Usage
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* List the available plugins
include_docs_end -->
* List the available plugins
```python
from tempor import plugin_loader
print(plugin_loader.list())
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Use a time-to-event (survival) analysis model
include_docs_end -->
* Use a time-to-event (survival) analysis model
```python
from tempor import plugin_loader
# Load a time-to-event dataset:
dataset = plugin_loader.get("time_to_event.pbc", plugin_type="datasource").load()
# Initialize the model:
model = plugin_loader.get("time_to_event.dynamic_deephit")
# Train:
model.fit(dataset)
# Make risk predictions:
prediction = model.predict(dataset, horizons=[0.25, 0.50, 0.75])
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Use a temporal treatment effects model
include_docs_end -->
* Use a temporal treatment effects model
```python
import numpy as np
from tempor import plugin_loader
# Load a dataset with temporal treatments and outcomes:
dataset = plugin_loader.get(
"treatments.temporal.dummy_treatments",
plugin_type="datasource",
temporal_covariates_missing_prob=0.0,
temporal_treatments_n_features=1,
temporal_treatments_n_categories=2,
).load()
# Initialize the model:
model = plugin_loader.get("treatments.temporal.regression.crn_regressor", epochs=20)
# Train:
model.fit(dataset)
# Define target variable horizons for each sample:
horizons = [
tc.time_indexes()[0][len(tc.time_indexes()[0]) // 2 :] for tc in dataset.time_series
]
# Define treatment scenarios for each sample:
treatment_scenarios = [
[np.asarray([1] * len(h)), np.asarray([0] * len(h))] for h in horizons
]
# Predict counterfactuals:
counterfactuals = model.predict_counterfactuals(
dataset,
horizons=horizons,
treatment_scenarios=treatment_scenarios,
)
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Use a missing data imputer
include_docs_end -->
* Use a missing data imputer
```python
from tempor import plugin_loader
dataset = plugin_loader.get(
"prediction.one_off.sine", plugin_type="datasource", with_missing=True
).load()
static_data_n_missing = dataset.static.dataframe().isna().sum().sum()
temporal_data_n_missing = dataset.time_series.dataframe().isna().sum().sum()
print(static_data_n_missing, temporal_data_n_missing)
assert static_data_n_missing > 0
assert temporal_data_n_missing > 0
# Initialize the model:
model = plugin_loader.get("preprocessing.imputation.temporal.bfill")
# Train:
model.fit(dataset)
# Impute:
imputed = model.transform(dataset)
temporal_data_n_missing = imputed.time_series.dataframe().isna().sum().sum()
print(static_data_n_missing, temporal_data_n_missing)
assert temporal_data_n_missing == 0
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Use a one-off classifier (prediction)
include_docs_end -->
* Use a one-off classifier (prediction)
```python
from tempor import plugin_loader
dataset = plugin_loader.get("prediction.one_off.sine", plugin_type="datasource").load()
# Initialize the model:
model = plugin_loader.get("prediction.one_off.classification.nn_classifier", n_iter=50)
# Train:
model.fit(dataset)
# Predict:
prediction = model.predict(dataset)
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Use a temporal regressor (forecasting)
include_docs_end -->
* Use a temporal regressor (forecasting)
```python
from tempor import plugin_loader
# Load a dataset with temporal targets.
dataset = plugin_loader.get(
"prediction.temporal.dummy_prediction",
plugin_type="datasource",
temporal_covariates_missing_prob=0.0,
).load()
# Initialize the model:
model = plugin_loader.get("prediction.temporal.regression.seq2seq_regressor", epochs=10)
# Train:
model.fit(dataset)
# Predict:
prediction = model.predict(dataset, n_future_steps=5)
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Benchmark models, time-to-event task
include_docs_end -->
* Benchmark models, time-to-event task
```python
from tempor.benchmarks import benchmark_models
from tempor import plugin_loader
from tempor.methods.pipeline import pipeline
testcases = [
(
"pipeline1",
pipeline(
[
"preprocessing.scaling.temporal.ts_minmax_scaler",
"time_to_event.dynamic_deephit",
]
)({"ts_coxph": {"n_iter": 100}}),
),
(
"plugin1",
plugin_loader.get("time_to_event.dynamic_deephit", n_iter=100),
),
(
"plugin2",
plugin_loader.get("time_to_event.ts_coxph", n_iter=100),
),
]
dataset = plugin_loader.get("time_to_event.pbc", plugin_type="datasource").load()
aggr_score, per_test_score = benchmark_models(
task_type="time_to_event",
tests=testcases,
data=dataset,
n_splits=2,
random_state=0,
horizons=[2.0, 4.0, 6.0],
)
print(aggr_score)
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* Serialization
include_docs_end -->
* Serialization
```python
from tempor.utils.serialization import load, save
from tempor import plugin_loader
# Initialize the model:
model = plugin_loader.get("prediction.one_off.classification.nn_classifier", n_iter=50)
buff = save(model) # Save model to bytes.
reloaded = load(buff) # Reload model.
# `save_to_file`, `load_from_file` also available in the serialization module.
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
* AutoML - search for the best pipeline for your task
include_docs_end -->
* AutoML - search for the best pipeline for your task
```python
from tempor.automl.seeker import PipelineSeeker
dataset = plugin_loader.get("prediction.one_off.sine", plugin_type="datasource").load()
# Specify the AutoML pipeline seeker for the task of your choice, providing candidate methods,
# metric, preprocessing steps etc.
seeker = PipelineSeeker(
study_name="my_automl_study",
task_type="prediction.one_off.classification",
estimator_names=[
"cde_classifier",
"ode_classifier",
"nn_classifier",
],
metric="aucroc",
dataset=dataset,
return_top_k=3,
num_iter=100,
tuner_type="bayesian",
static_imputers=["static_tabular_imputer"],
static_scalers=[],
temporal_imputers=["ffill", "bfill"],
temporal_scalers=["ts_minmax_scaler"],
)
# The search will return the best pipelines.
best_pipelines, best_scores = seeker.search() # doctest: +SKIP
```
<!-- exclude_docs -->
<!-- exclude_docs_end -->
## 📖 Tutorials
### Data
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial01_data_format.ipynb) - [Data Format](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial01_data_format.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial02_datasets.ipynb) - [Datasets](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial02_datasets.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial03_datasources.ipynb) - [Data Loaders](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial03_datasources.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial04_data_splitting.ipynb) - [Data Splitting](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial04_data_splitting.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial05_other_data_formats.ipynb) - [Other Data Formats](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial05_other_data_formats.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial06_mimic_use_case.ipynb) - [MIMIC Use Case](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial06_mimic_use_case.ipynb)
### User Guide
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial01_plugins.ipynb) - [Plugins](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial01_plugins.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial02_imputation.ipynb) - [Imputation](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial02_imputation.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial03_scaling.ipynb) - [Scaling](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial03_scaling.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial04_prediction.ipynb) - [Prediction](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial04_prediction.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial05_time_to_event.ipynb) - [Time-to-event Analysis](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial05_time_to_event.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial06_treatments.ipynb) - [Treatment Effects](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial06_treatments.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial07_pipeline.ipynb) - [Pipeline](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial07_pipeline.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial08_benchmarks.ipynb) - [Benchmarks](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial08_benchmarks.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial09_automl.ipynb) - [AutoML](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial09_automl.ipynb)
### Extending TemporAI
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial01_custom_method.ipynb) - [Writing a Custom Method Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial01_custom_method.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial02_testing_custom_method.ipynb) - [Testing a Custom Method Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial02_testing_custom_method.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial03_custom_datasource.ipynb) - [Writing a Custom Data Source Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial03_custom_datasource.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial04_custom_metric.ipynb) - [Writing a Custom Metric Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial04_custom_metric.ipynb)
- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial05_custom_dataformat.ipynb) - [Writing a Custom Data Format](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial05_custom_dataformat.ipynb)
<!-- exclude_docs -->
## 📘 Documentation
See the full project documentation [here](https://temporai.readthedocs.io/en/latest/).
<!-- exclude_docs_end -->
<!--- Reusable --->
[van der Schaar Lab]: https://www.vanderschaar-lab.com/
[docs]: https://temporai.readthedocs.io/en/latest/
<!-- exclude_docs -->
[docs/user_guide]: https://temporai.readthedocs.io/en/latest/user_guide/index.html
<!-- exclude_docs_end -->
## 🌍 TemporAI Ecosystem (*Experimental*)
We provide additional tools in the TemporAI ecosystem, which are in active development, and are currently (very) experimental. Suggestions and contributions are welcome!
These include:
- [`temporai-clinic`](https://github.com/vanderschaarlab/temporai-clinic): A web app tool for interacting and visualising TemporAI models, data, and predictions.
- [`temporai-mivdp`](https://github.com/vanderschaarlab/temporai-mivdp): A [MIMIC-IV-Data-Pipeline](https://github.com/healthylaife/MIMIC-IV-Data-Pipeline) adaptation for TemporAI.
<!-- include_docs
{#methods}include_docs_end -->
## 🔑 Methods
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
### Time-to-Event (survival) analysis over time
include_docs_end -->
### Time-to-Event (survival) analysis over time
Risk estimation given event data (category: `time_to_event`)
| Name | Description| Reference |
| --- | --- | --- |
| `dynamic_deephit` | Dynamic-DeepHit incorporates the available longitudinal data comprising various repeated measurements (rather than only the last available measurements) in order to issue dynamically updated survival predictions | [Paper](https://pubmed.ncbi.nlm.nih.gov/30951460/) |
| `ts_coxph` | Create embeddings from the time series and use a CoxPH model for predicting the survival function| --- |
| `ts_xgb` | Create embeddings from the time series and use a SurvivalXGBoost model for predicting the survival function| --- |
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
### Treatment effects
include_docs_end -->
### Treatment effects
#### One-off
Treatment effects estimation where treatments are a one-off event.
<!--
* Classification on the outcomes (category: `treatments.one_off.classification`)
-->
* Regression on the outcomes (category: `treatments.one_off.regression`)
| Name | Description| Reference |
| --- | --- | --- |
| `synctwin_regressor` | SyncTwin is a treatment effect estimation method tailored for observational studies with longitudinal data, applied to the LIP setting: Longitudinal, Irregular and Point treatment. | [Paper](https://proceedings.neurips.cc/paper/2021/hash/19485224d128528da1602ca47383f078-Abstract.html) |
#### Temporal
Treatment effects estimation where treatments are temporal (time series).
* Classification on the outcomes (category: `treatments.temporal.classification`)
| Name | Description| Reference |
| --- | --- | --- |
| `crn_classifier` | The Counterfactual Recurrent Network (CRN), a sequence-to-sequence model that leverages the available patient observational data to estimate treatment effects over time. | [Paper](https://arxiv.org/abs/2002.04083) |
* Regression on the outcomes (category: `treatments.temporal.regression`)
| Name | Description| Reference |
| --- | --- | --- |
| `crn_regressor` | The Counterfactual Recurrent Network (CRN), a sequence-to-sequence model that leverages the available patient observational data to estimate treatment effects over time. | [Paper](https://arxiv.org/abs/2002.04083) |
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
### Prediction
include_docs_end -->
### Prediction
#### One-off
Prediction where targets are static.
* Classification (category: `prediction.one_off.classification`)
| Name | Description| Reference |
| --- | --- | --- |
| `nn_classifier` | Neural-net based classifier. Supports multiple recurrent models, like RNN, LSTM, Transformer etc. | --- |
| `ode_classifier` | Classifier based on ordinary differential equation (ODE) solvers. | --- |
| `cde_classifier` | Classifier based Neural Controlled Differential Equations for Irregular Time Series. | [Paper](https://arxiv.org/abs/2005.08926) |
| `laplace_ode_classifier` | Classifier based Inverse Laplace Transform (ILT) algorithms implemented in PyTorch. | [Paper](https://arxiv.org/abs/2206.04843) |
* Regression (category: `prediction.one_off.regression`)
| Name | Description| Reference |
| --- | --- | --- |
| `nn_regressor` | Neural-net based regressor. Supports multiple recurrent models, like RNN, LSTM, Transformer etc. | --- |
| `ode_regressor` | Regressor based on ordinary differential equation (ODE) solvers. | --- |
| `cde_regressor` | Regressor based Neural Controlled Differential Equations for Irregular Time Series. | [Paper](https://arxiv.org/abs/2005.08926)
| `laplace_ode_regressor` | Regressor based Inverse Laplace Transform (ILT) algorithms implemented in PyTorch. | [Paper](https://arxiv.org/abs/2206.04843) |
#### Temporal
Prediction where targets are temporal (time series).
* Classification (category: `prediction.temporal.classification`)
| Name | Description| Reference |
| --- | --- | --- |
| `seq2seq_classifier` | Seq2Seq prediction, classification | --- |
* Regression (category: `prediction.temporal.regression`)
| Name | Description| Reference |
| --- | --- | --- |
| `seq2seq_regressor` | Seq2Seq prediction, regression | --- |
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- exclude_docs -->
<!-- exclude_docs_end -->
<!-- include_docs
### Preprocessing
include_docs_end -->
### Preprocessing
#### Feature Encoding
* Static data (category: `preprocessing.encoding.static`)
| Name | Description| Reference |
| --- | --- | --- |
| `static_onehot_encoder` | One-hot encode categorical static features | --- |
* Temporal data (category: `preprocessing.encoding.temporal`)
| Name | Description| Reference |
| --- | --- | --- |
| `ts_onehot_encoder` | One-hot encode categorical time series features | --- |
#### Imputation
* Static data (category: `preprocessing.imputation.static`)
| Name | Description| Reference |
| --- | --- | --- |
| `static_tabular_imputer` | Use any method from [HyperImpute](https://github.com/vanderschaarlab/hyperimpute) (HyperImpute, Mean, Median, Most-frequent, MissForest, ICE, MICE, SoftImpute, EM, Sinkhorn, GAIN, MIRACLE, MIWAE) to impute the static data | [Paper](https://arxiv.org/abs/2206.07769) |
* Temporal data (category: `preprocessing.imputation.temporal`)
| Name | Description| Reference |
| --- | --- | --- |
| `ffill` | Propagate last valid observation forward to next valid | --- |
| `bfill` | Use next valid observation to fill gap | --- |
| `ts_tabular_imputer` | Use any method from [HyperImpute](https://github.com/vanderschaarlab/hyperimpute) (HyperImpute, Mean, Median, Most-frequent, MissForest, ICE, MICE, SoftImpute, EM, Sinkhorn, GAIN, MIRACLE, MIWAE) to impute the time series data | [Paper](https://arxiv.org/abs/2206.07769) |
#### Scaling
* Static data (category: `preprocessing.scaling.static`)
| Name | Description| Reference |
| --- | --- | --- |
| `static_standard_scaler` | Scale the static features using a StandardScaler | --- |
| `static_minmax_scaler` | Scale the static features using a MinMaxScaler | --- |
* Temporal data (category: `preprocessing.scaling.temporal`)
| Name | Description| Reference |
| --- | --- | --- |
| `ts_standard_scaler` | Scale the temporal features using a StandardScaler | --- |
| `ts_minmax_scaler` | Scale the temporal features using a MinMaxScaler | --- |
<!-- exclude_docs -->
<!-- exclude_docs_end -->
## 🔨 Tests and Development
Install the testing dependencies using:
```bash
pip install .[testing]
```
The tests can be executed using:
```bash
pytest -vsx
```
For local development, we recommend that you should install the `[dev]` extra, which includes `[testing]` and some additional dependencies:
```bash
pip install .[dev]
```
For development and contribution to TemporAI, see:
* 📓 [Extending TemporAI tutorials](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/)
* 📃 [Contribution guide](https://github.com/vanderschaarlab/temporai/tree/main/CONTRIBUTING.md)
* 👩💻 [Developer's guide](https://github.com/vanderschaarlab/temporai/tree/main/docs/dev_guide.md)
## ✍️ Citing
If you use this code, please cite the associated paper:
```
@article{saveliev2023temporai,
title={TemporAI: Facilitating Machine Learning Innovation in Time Domain Tasks for Medicine},
author={Saveliev, Evgeny S and van der Schaar, Mihaela},
journal={arXiv preprint arXiv:2301.12260},
year={2023}
}
```
Raw data
{
"_id": null,
"home_page": "https://github.com/vanderschaarlab/temporai",
"name": "temporai",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.7",
"maintainer_email": "",
"keywords": "",
"author": "Evgeny Saveliev",
"author_email": "e.s.saveliev@gmail.com",
"download_url": "",
"platform": "any",
"description": "<!-- These are examples of badges you might want to add to your README:\r\n please update the URLs accordingly\r\n\r\n[](https://anaconda.org/conda-forge/temporai)\r\n[](https://pepy.tech/project/temporai)\r\n-->\r\n\r\n<!-- exclude_docs -->\r\n[](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial04_prediction.ipynb)\r\n[](https://temporai.readthedocs.io/en/latest/?badge=latest)\r\n\r\n[](https://www.python.org/downloads/release/python-370/)\r\n[](https://pypi.org/project/temporai/)\r\n[](https://pepy.tech/project/temporai)\r\n[](https://github.com/vanderschaarlab/temporai/tree/main/LICENSE.txt)\r\n\r\n[](https://github.com/vanderschaarlab/temporai/actions/workflows/test.yml)\r\n[](https://github.com/vanderschaarlab/temporai/actions/workflows/test.yml)\r\n[](https://codecov.io/gh/vanderschaarlab/temporai)\r\n\r\n[](https://arxiv.org/abs/2301.12260)\r\n[](https://join.slack.com/t/vanderschaarlab/shared_invite/zt-1u2rmhw06-sHS5nQDMN3Ka2Zer6sAU6Q)\r\n[](https://www.vanderschaar-lab.com/)\r\n<!-- exclude_docs_end -->\r\n\r\n# <img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/TemporAI_Logo_Icon.png' height=25> TemporAI\r\n\r\n<!-- exclude_docs -->\r\n> **\u2697\ufe0f Status:** This project is still in *alpha*, and the API may change without warning. \r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n:::{important}\r\n**Status:** This project is still in *alpha*, and the API may change without warning. \r\n:::\r\ninclude_docs_end -->\r\n\r\n\r\n## \ud83d\udcc3 Overview\r\n\r\n*TemporAI* is a Machine Learning-centric time-series library for medicine. The tasks that are currently of focus in TemporAI are: time-to-event (survival) analysis with time-series data, treatment effects (causal inference) over time, and time-series prediction. Data preprocessing methods, including missing value imputation for static and temporal covariates, are provided. AutoML tools for hyperparameter tuning and pipeline selection are also available.\r\n\r\n### How is TemporAI unique?\r\n\r\n* **\ud83c\udfe5 Medicine-first:** Focused on use cases for medicine and healthcare, such as temporal treatment effects, survival analysis over time, imputation methods, models with built-in and post-hoc interpretability, ... See [methods](https://github.com/vanderschaarlab/temporai/tree/main/#-methods).\r\n* **\ud83c\udfd7\ufe0f Fast prototyping:** A plugin design allowing for on-the-fly integration of new methods by the users.\r\n* **\ud83d\ude80 From research to practice:** Relevant novel models from research community adapted for practical use.\r\n* **\ud83c\udf0d A healthcare ecosystem vision:** A range of interactive demonstration apps, new medical problem settings, interpretability tools, data-centric tools etc. are planned.\r\n\r\n### Key concepts\r\n\r\n<div align=\"center\">\r\n\r\n<!-- exclude_docs -->\r\n<img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/Conceptual.png' alt=\"key concepts\">\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n<img src='https://raw.githubusercontent.com/vanderschaarlab/temporai/main/docs/assets/Conceptual.png' alt=\"key concepts\">\r\ninclude_docs_end -->\r\n\r\n</div>\r\n\r\n\r\n\r\n## \ud83d\ude80 Installation\r\n\r\n### Instal with `pip`\r\n\r\nFrom [the Python Package Index (PyPI)](https://pypi.org/):\r\n```bash\r\n$ pip install temporai\r\n```\r\n\r\nOr from source:\r\n```bash\r\n$ git clone https://github.com/vanderschaarlab/temporai.git\r\n$ cd temporai\r\n$ pip install .\r\n```\r\n\r\n### Install in a [conda](https://docs.conda.io/en/latest/) environment\r\n\r\nWhile have not yet published TemporAI on `conda-forge`, you can still install TemporAI in your conda environment using `pip` as follows:\r\n\r\nCreate and activate conda environment as normal:\r\n```bash\r\n$ conda create -n <my_environment>\r\n$ conda activate <my_environment>\r\n```\r\n\r\nThen install inside your `conda` environment with pip:\r\n```bash\r\n$ pip install temporai\r\n```\r\n\r\n\r\n## \ud83d\udca5 Sample Usage\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* List the available plugins\r\ninclude_docs_end -->\r\n\r\n* List the available plugins\r\n\r\n\r\n```python\r\nfrom tempor import plugin_loader\r\n\r\nprint(plugin_loader.list())\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Use a time-to-event (survival) analysis model\r\ninclude_docs_end -->\r\n\r\n* Use a time-to-event (survival) analysis model\r\n\r\n\r\n```python\r\nfrom tempor import plugin_loader\r\n\r\n# Load a time-to-event dataset:\r\ndataset = plugin_loader.get(\"time_to_event.pbc\", plugin_type=\"datasource\").load()\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"time_to_event.dynamic_deephit\")\r\n\r\n# Train:\r\nmodel.fit(dataset)\r\n\r\n# Make risk predictions:\r\nprediction = model.predict(dataset, horizons=[0.25, 0.50, 0.75])\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Use a temporal treatment effects model\r\ninclude_docs_end -->\r\n\r\n* Use a temporal treatment effects model\r\n\r\n\r\n```python\r\nimport numpy as np\r\n\r\nfrom tempor import plugin_loader\r\n\r\n# Load a dataset with temporal treatments and outcomes:\r\ndataset = plugin_loader.get(\r\n \"treatments.temporal.dummy_treatments\",\r\n plugin_type=\"datasource\",\r\n temporal_covariates_missing_prob=0.0,\r\n temporal_treatments_n_features=1,\r\n temporal_treatments_n_categories=2,\r\n).load()\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"treatments.temporal.regression.crn_regressor\", epochs=20)\r\n\r\n# Train:\r\nmodel.fit(dataset)\r\n\r\n# Define target variable horizons for each sample:\r\nhorizons = [\r\n tc.time_indexes()[0][len(tc.time_indexes()[0]) // 2 :] for tc in dataset.time_series\r\n]\r\n\r\n# Define treatment scenarios for each sample:\r\ntreatment_scenarios = [\r\n [np.asarray([1] * len(h)), np.asarray([0] * len(h))] for h in horizons\r\n]\r\n\r\n# Predict counterfactuals:\r\ncounterfactuals = model.predict_counterfactuals(\r\n dataset,\r\n horizons=horizons,\r\n treatment_scenarios=treatment_scenarios,\r\n)\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Use a missing data imputer\r\ninclude_docs_end -->\r\n\r\n* Use a missing data imputer\r\n\r\n\r\n```python\r\nfrom tempor import plugin_loader\r\n\r\ndataset = plugin_loader.get(\r\n \"prediction.one_off.sine\", plugin_type=\"datasource\", with_missing=True\r\n).load()\r\nstatic_data_n_missing = dataset.static.dataframe().isna().sum().sum()\r\ntemporal_data_n_missing = dataset.time_series.dataframe().isna().sum().sum()\r\n\r\nprint(static_data_n_missing, temporal_data_n_missing)\r\nassert static_data_n_missing > 0\r\nassert temporal_data_n_missing > 0\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"preprocessing.imputation.temporal.bfill\")\r\n\r\n# Train:\r\nmodel.fit(dataset)\r\n\r\n# Impute:\r\nimputed = model.transform(dataset)\r\ntemporal_data_n_missing = imputed.time_series.dataframe().isna().sum().sum()\r\n\r\nprint(static_data_n_missing, temporal_data_n_missing)\r\nassert temporal_data_n_missing == 0\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Use a one-off classifier (prediction)\r\ninclude_docs_end -->\r\n\r\n* Use a one-off classifier (prediction)\r\n\r\n\r\n```python\r\nfrom tempor import plugin_loader\r\n\r\ndataset = plugin_loader.get(\"prediction.one_off.sine\", plugin_type=\"datasource\").load()\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"prediction.one_off.classification.nn_classifier\", n_iter=50)\r\n\r\n# Train:\r\nmodel.fit(dataset)\r\n\r\n# Predict:\r\nprediction = model.predict(dataset)\r\n```\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Use a temporal regressor (forecasting)\r\ninclude_docs_end -->\r\n\r\n* Use a temporal regressor (forecasting)\r\n\r\n\r\n```python\r\nfrom tempor import plugin_loader\r\n\r\n# Load a dataset with temporal targets.\r\ndataset = plugin_loader.get(\r\n \"prediction.temporal.dummy_prediction\",\r\n plugin_type=\"datasource\",\r\n temporal_covariates_missing_prob=0.0,\r\n).load()\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"prediction.temporal.regression.seq2seq_regressor\", epochs=10)\r\n\r\n# Train:\r\nmodel.fit(dataset)\r\n\r\n# Predict:\r\nprediction = model.predict(dataset, n_future_steps=5)\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Benchmark models, time-to-event task\r\ninclude_docs_end -->\r\n\r\n* Benchmark models, time-to-event task\r\n\r\n\r\n```python\r\nfrom tempor.benchmarks import benchmark_models\r\nfrom tempor import plugin_loader\r\nfrom tempor.methods.pipeline import pipeline\r\n\r\ntestcases = [\r\n (\r\n \"pipeline1\",\r\n pipeline(\r\n [\r\n \"preprocessing.scaling.temporal.ts_minmax_scaler\",\r\n \"time_to_event.dynamic_deephit\",\r\n ]\r\n )({\"ts_coxph\": {\"n_iter\": 100}}),\r\n ),\r\n (\r\n \"plugin1\",\r\n plugin_loader.get(\"time_to_event.dynamic_deephit\", n_iter=100),\r\n ),\r\n (\r\n \"plugin2\",\r\n plugin_loader.get(\"time_to_event.ts_coxph\", n_iter=100),\r\n ),\r\n]\r\ndataset = plugin_loader.get(\"time_to_event.pbc\", plugin_type=\"datasource\").load()\r\n\r\naggr_score, per_test_score = benchmark_models(\r\n task_type=\"time_to_event\",\r\n tests=testcases,\r\n data=dataset,\r\n n_splits=2,\r\n random_state=0,\r\n horizons=[2.0, 4.0, 6.0],\r\n)\r\n\r\nprint(aggr_score)\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* Serialization\r\ninclude_docs_end -->\r\n\r\n* Serialization\r\n\r\n\r\n```python\r\nfrom tempor.utils.serialization import load, save\r\nfrom tempor import plugin_loader\r\n\r\n# Initialize the model:\r\nmodel = plugin_loader.get(\"prediction.one_off.classification.nn_classifier\", n_iter=50)\r\n\r\nbuff = save(model) # Save model to bytes.\r\nreloaded = load(buff) # Reload model.\r\n\r\n# `save_to_file`, `load_from_file` also available in the serialization module.\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n* AutoML - search for the best pipeline for your task\r\ninclude_docs_end -->\r\n\r\n* AutoML - search for the best pipeline for your task\r\n\r\n\r\n```python\r\nfrom tempor.automl.seeker import PipelineSeeker\r\n\r\ndataset = plugin_loader.get(\"prediction.one_off.sine\", plugin_type=\"datasource\").load()\r\n\r\n# Specify the AutoML pipeline seeker for the task of your choice, providing candidate methods,\r\n# metric, preprocessing steps etc.\r\nseeker = PipelineSeeker(\r\n study_name=\"my_automl_study\",\r\n task_type=\"prediction.one_off.classification\",\r\n estimator_names=[\r\n \"cde_classifier\",\r\n \"ode_classifier\",\r\n \"nn_classifier\",\r\n ],\r\n metric=\"aucroc\",\r\n dataset=dataset,\r\n return_top_k=3,\r\n num_iter=100,\r\n tuner_type=\"bayesian\",\r\n static_imputers=[\"static_tabular_imputer\"],\r\n static_scalers=[],\r\n temporal_imputers=[\"ffill\", \"bfill\"],\r\n temporal_scalers=[\"ts_minmax_scaler\"],\r\n)\r\n\r\n# The search will return the best pipelines.\r\nbest_pipelines, best_scores = seeker.search() # doctest: +SKIP\r\n```\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n\r\n\r\n## \ud83d\udcd6 Tutorials\r\n\r\n### Data\r\n\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial01_data_format.ipynb) - [Data Format](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial01_data_format.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial02_datasets.ipynb) - [Datasets](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial02_datasets.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial03_datasources.ipynb) - [Data Loaders](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial03_datasources.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial04_data_splitting.ipynb) - [Data Splitting](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial04_data_splitting.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial05_other_data_formats.ipynb) - [Other Data Formats](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial05_other_data_formats.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/data/tutorial06_mimic_use_case.ipynb) - [MIMIC Use Case](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/data/tutorial06_mimic_use_case.ipynb)\r\n\r\n### User Guide\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial01_plugins.ipynb) - [Plugins](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial01_plugins.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial02_imputation.ipynb) - [Imputation](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial02_imputation.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial03_scaling.ipynb) - [Scaling](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial03_scaling.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial04_prediction.ipynb) - [Prediction](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial04_prediction.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial05_time_to_event.ipynb) - [Time-to-event Analysis](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial05_time_to_event.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial06_treatments.ipynb) - [Treatment Effects](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial06_treatments.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial07_pipeline.ipynb) - [Pipeline](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial07_pipeline.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial08_benchmarks.ipynb) - [Benchmarks](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial08_benchmarks.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/usage/tutorial09_automl.ipynb) - [AutoML](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/usage/tutorial09_automl.ipynb)\r\n\r\n### Extending TemporAI\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial01_custom_method.ipynb) - [Writing a Custom Method Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial01_custom_method.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial02_testing_custom_method.ipynb) - [Testing a Custom Method Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial02_testing_custom_method.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial03_custom_datasource.ipynb) - [Writing a Custom Data Source Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial03_custom_datasource.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial04_custom_metric.ipynb) - [Writing a Custom Metric Plugin](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial04_custom_metric.ipynb)\r\n- [](https://colab.research.google.com/github/vanderschaarlab/temporai/blob/main/tutorials/extending/tutorial05_custom_dataformat.ipynb) - [Writing a Custom Data Format](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/tutorial05_custom_dataformat.ipynb)\r\n\r\n\r\n\r\n<!-- exclude_docs -->\r\n## \ud83d\udcd8 Documentation\r\n\r\nSee the full project documentation [here](https://temporai.readthedocs.io/en/latest/).\r\n<!-- exclude_docs_end -->\r\n\r\n\r\n\r\n<!--- Reusable --->\r\n [van der Schaar Lab]: https://www.vanderschaar-lab.com/\r\n [docs]: https://temporai.readthedocs.io/en/latest/\r\n<!-- exclude_docs -->\r\n [docs/user_guide]: https://temporai.readthedocs.io/en/latest/user_guide/index.html\r\n<!-- exclude_docs_end -->\r\n\r\n\r\n\r\n## \ud83c\udf0d TemporAI Ecosystem (*Experimental*)\r\n\r\nWe provide additional tools in the TemporAI ecosystem, which are in active development, and are currently (very) experimental. Suggestions and contributions are welcome!\r\n\r\nThese include:\r\n- [`temporai-clinic`](https://github.com/vanderschaarlab/temporai-clinic): A web app tool for interacting and visualising TemporAI models, data, and predictions.\r\n- [`temporai-mivdp`](https://github.com/vanderschaarlab/temporai-mivdp): A [MIMIC-IV-Data-Pipeline](https://github.com/healthylaife/MIMIC-IV-Data-Pipeline) adaptation for TemporAI.\r\n\r\n\r\n\r\n<!-- include_docs\r\n{#methods}include_docs_end -->\r\n## \ud83d\udd11 Methods\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n### Time-to-Event (survival) analysis over time\r\ninclude_docs_end -->\r\n\r\n### Time-to-Event (survival) analysis over time\r\n\r\n\r\nRisk estimation given event data (category: `time_to_event`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `dynamic_deephit` | Dynamic-DeepHit incorporates the available longitudinal data comprising various repeated measurements (rather than only the last available measurements) in order to issue dynamically updated survival predictions | [Paper](https://pubmed.ncbi.nlm.nih.gov/30951460/) |\r\n| `ts_coxph` | Create embeddings from the time series and use a CoxPH model for predicting the survival function| --- |\r\n| `ts_xgb` | Create embeddings from the time series and use a SurvivalXGBoost model for predicting the survival function| --- |\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n### Treatment effects\r\ninclude_docs_end -->\r\n\r\n### Treatment effects\r\n\r\n\r\n#### One-off\r\nTreatment effects estimation where treatments are a one-off event.\r\n\r\n<!--\r\n* Classification on the outcomes (category: `treatments.one_off.classification`)\r\n-->\r\n\r\n* Regression on the outcomes (category: `treatments.one_off.regression`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `synctwin_regressor` | SyncTwin is a treatment effect estimation method tailored for observational studies with longitudinal data, applied to the LIP setting: Longitudinal, Irregular and Point treatment. | [Paper](https://proceedings.neurips.cc/paper/2021/hash/19485224d128528da1602ca47383f078-Abstract.html) |\r\n\r\n#### Temporal\r\nTreatment effects estimation where treatments are temporal (time series).\r\n\r\n* Classification on the outcomes (category: `treatments.temporal.classification`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `crn_classifier` | The Counterfactual Recurrent Network (CRN), a sequence-to-sequence model that leverages the available patient observational data to estimate treatment effects over time. | [Paper](https://arxiv.org/abs/2002.04083) |\r\n\r\n* Regression on the outcomes (category: `treatments.temporal.regression`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `crn_regressor` | The Counterfactual Recurrent Network (CRN), a sequence-to-sequence model that leverages the available patient observational data to estimate treatment effects over time. | [Paper](https://arxiv.org/abs/2002.04083) |\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n### Prediction\r\ninclude_docs_end -->\r\n\r\n### Prediction\r\n\r\n\r\n#### One-off\r\nPrediction where targets are static.\r\n\r\n* Classification (category: `prediction.one_off.classification`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `nn_classifier` | Neural-net based classifier. Supports multiple recurrent models, like RNN, LSTM, Transformer etc. | --- |\r\n| `ode_classifier` | Classifier based on ordinary differential equation (ODE) solvers. | --- |\r\n| `cde_classifier` | Classifier based Neural Controlled Differential Equations for Irregular Time Series. | [Paper](https://arxiv.org/abs/2005.08926) |\r\n| `laplace_ode_classifier` | Classifier based Inverse Laplace Transform (ILT) algorithms implemented in PyTorch. | [Paper](https://arxiv.org/abs/2206.04843) |\r\n\r\n* Regression (category: `prediction.one_off.regression`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `nn_regressor` | Neural-net based regressor. Supports multiple recurrent models, like RNN, LSTM, Transformer etc. | --- |\r\n| `ode_regressor` | Regressor based on ordinary differential equation (ODE) solvers. | --- |\r\n| `cde_regressor` | Regressor based Neural Controlled Differential Equations for Irregular Time Series. | [Paper](https://arxiv.org/abs/2005.08926)\r\n| `laplace_ode_regressor` | Regressor based Inverse Laplace Transform (ILT) algorithms implemented in PyTorch. | [Paper](https://arxiv.org/abs/2206.04843) |\r\n\r\n#### Temporal\r\nPrediction where targets are temporal (time series).\r\n\r\n* Classification (category: `prediction.temporal.classification`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `seq2seq_classifier` | Seq2Seq prediction, classification | --- |\r\n\r\n* Regression (category: `prediction.temporal.regression`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `seq2seq_regressor` | Seq2Seq prediction, regression | --- |\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n<!-- include_docs\r\n### Preprocessing\r\ninclude_docs_end -->\r\n\r\n### Preprocessing\r\n\r\n\r\n#### Feature Encoding\r\n\r\n* Static data (category: `preprocessing.encoding.static`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `static_onehot_encoder` | One-hot encode categorical static features | --- |\r\n\r\n* Temporal data (category: `preprocessing.encoding.temporal`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `ts_onehot_encoder` | One-hot encode categorical time series features | --- |\r\n\r\n#### Imputation\r\n\r\n* Static data (category: `preprocessing.imputation.static`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `static_tabular_imputer` | Use any method from [HyperImpute](https://github.com/vanderschaarlab/hyperimpute) (HyperImpute, Mean, Median, Most-frequent, MissForest, ICE, MICE, SoftImpute, EM, Sinkhorn, GAIN, MIRACLE, MIWAE) to impute the static data | [Paper](https://arxiv.org/abs/2206.07769) |\r\n\r\n* Temporal data (category: `preprocessing.imputation.temporal`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `ffill` | Propagate last valid observation forward to next valid | --- |\r\n| `bfill` | Use next valid observation to fill gap | --- |\r\n| `ts_tabular_imputer` | Use any method from [HyperImpute](https://github.com/vanderschaarlab/hyperimpute) (HyperImpute, Mean, Median, Most-frequent, MissForest, ICE, MICE, SoftImpute, EM, Sinkhorn, GAIN, MIRACLE, MIWAE) to impute the time series data | [Paper](https://arxiv.org/abs/2206.07769) |\r\n\r\n\r\n#### Scaling\r\n\r\n* Static data (category: `preprocessing.scaling.static`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `static_standard_scaler` | Scale the static features using a StandardScaler | --- |\r\n| `static_minmax_scaler` | Scale the static features using a MinMaxScaler | --- |\r\n\r\n* Temporal data (category: `preprocessing.scaling.temporal`)\r\n\r\n| Name | Description| Reference |\r\n| --- | --- | --- |\r\n| `ts_standard_scaler` | Scale the temporal features using a StandardScaler | --- |\r\n| `ts_minmax_scaler` | Scale the temporal features using a MinMaxScaler | --- |\r\n\r\n\r\n<!-- exclude_docs -->\r\n<!-- exclude_docs_end -->\r\n\r\n\r\n\r\n## \ud83d\udd28 Tests and Development\r\n\r\nInstall the testing dependencies using:\r\n```bash\r\npip install .[testing]\r\n```\r\nThe tests can be executed using:\r\n```bash\r\npytest -vsx\r\n```\r\n\r\nFor local development, we recommend that you should install the `[dev]` extra, which includes `[testing]` and some additional dependencies:\r\n```bash\r\npip install .[dev]\r\n```\r\n\r\nFor development and contribution to TemporAI, see:\r\n* \ud83d\udcd3 [Extending TemporAI tutorials](https://github.com/vanderschaarlab/temporai/tree/main/tutorials/extending/)\r\n* \ud83d\udcc3 [Contribution guide](https://github.com/vanderschaarlab/temporai/tree/main/CONTRIBUTING.md)\r\n* \ud83d\udc69\u200d\ud83d\udcbb [Developer's guide](https://github.com/vanderschaarlab/temporai/tree/main/docs/dev_guide.md)\r\n\r\n\r\n\r\n## \u270d\ufe0f Citing\r\n\r\nIf you use this code, please cite the associated paper:\r\n```\r\n@article{saveliev2023temporai,\r\n title={TemporAI: Facilitating Machine Learning Innovation in Time Domain Tasks for Medicine},\r\n author={Saveliev, Evgeny S and van der Schaar, Mihaela},\r\n journal={arXiv preprint arXiv:2301.12260},\r\n year={2023}\r\n}\r\n```\r\n",
"bugtrack_url": null,
"license": "Apache-2.0 license",
"summary": "TemporAI: ML-centric Toolkit for Medical Time Series",
"version": "0.0.3",
"project_urls": {
"Changelog": "https://github.com/vanderschaarlab/temporai/releases",
"Documentation": "https://temporai.readthedocs.io/en/latest/",
"Homepage": "https://github.com/vanderschaarlab/temporai",
"Source": "https://github.com/vanderschaarlab/temporai",
"Tracker": "https://github.com/vanderschaarlab/temporai/issues"
},
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "64def58349da04815bfb0466a0b6330af4e2497c06c9a0ebdb7aa00eda00e2b8",
"md5": "2d12b68a5552d3e85b32b0aaaba82430",
"sha256": "f05f88fa44a40724e57c1d29101869ba7453aae7374048f39decde8a96876516"
},
"downloads": -1,
"filename": "temporai-0.0.3-py3-none-any.whl",
"has_sig": false,
"md5_digest": "2d12b68a5552d3e85b32b0aaaba82430",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.7",
"size": 236555,
"upload_time": "2023-12-07T23:05:06",
"upload_time_iso_8601": "2023-12-07T23:05:06.258166Z",
"url": "https://files.pythonhosted.org/packages/64/de/f58349da04815bfb0466a0b6330af4e2497c06c9a0ebdb7aa00eda00e2b8/temporai-0.0.3-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-12-07 23:05:06",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "vanderschaarlab",
"github_project": "temporai",
"travis_ci": false,
"coveralls": true,
"github_actions": true,
"tox": true,
"lcname": "temporai"
}
JSON
