> ## Documentation Index
> Fetch the complete documentation index at: https://docs.graphext.com/llms.txt
> Use this file to discover all available pages before exploring further.
# train_survival
> Train and store a survival model to be loaded at a later point for prediction.
Trains a survival model using the Cox Proportional Hazard model.
The output will always be a new column with the trained model's predictions on the training data,
as well as a saved and named model file that can be used in other projects for prediction of new data.
## Usage
The following examples show how the step can be used in a recipe.
Train a Cox Proportional Hazard survival model
```stan theme={null}
train_survival(ds, {"target": ["event_observed", "duration"]}) -> (ds.predicted_survival, "my-survival-model")
```
Train a CoxPH model with penalization and median survival time prediction
```stan theme={null}
train_survival(ds, {"target": ["event_observed", "time_to_event"], "predictions": {"kind": "median"}, "params": {"penalizer": 0.1}}) -> (ds.predicted_survival, "my-survival-model")
```
General syntax for using the step in a recipe. Shows the inputs and outputs the step is expected to receive and will produce respectively. For futher details see sections below.
```stan theme={null}
train_survival(ds: dataset, {
"param": value,
...
}) -> (predicted: number, model: model_survival[ds])
```
## Inputs & Outputs
The following are the inputs expected by the step and the outputs it produces. These are generally
columns (`ds.first_name`), datasets (`ds` or `ds[["first_name", "last_name"]]`) or models (referenced
by name e.g. `"churn-clf"`).
Should contain the target columns (see `target` parameter below) and the feature columns you wish to use in the model.
Name for output column containing model predictions.
Zip file containing the trained model and associated information.
## Configuration
The following parameters can be used to configure the behaviour of the step by including them in
a json object as the last "input" to the step, i.e. `step(..., {"param": "value", ...}) -> (output)`.
Kind of survival model to train.
"CoxPH" trains a [lifelines Cox Proportional Hazard model](https://lifelines.readthedocs.io/en/latest/Survival%20Regression.html#cox-s-proportional-hazard-model).
Target variables.
Two names, exactly, corresponding to the target columns that contain in the following order:
1. whether the event was observed (boolean) and
2. the time (duration) to event or censoring (number).
Configure the kind of predictions to return.
Kind of prediction.
`median` returns the median survival time. `percentile` returns the survival time at
the given percentile. `expectation` returns the expected survival time.
`survival_function` returns the whole survival function (one series per sample).
Values must be one of the following:
* `median`
* `percentile`
* `expectation`
* `survival_function`
Percentile when `kind` is set to `percentile`
Values must be in the following range:
```javascript theme={null}
0 ≤ percentile ≤ 1
```
Points in time to predict.
Configures at which points to predict when `kind` is set to `survival_function`.
Either an explicit array of durations, or an object specifying a duration step size and
maximum duration.
Array of times/durations.
Will predict the survival function at each of the durations. E.g. `[1, 2, 3, 4, 5]`.
Each item in array.
Step size.
The step size in the enumeration of durations. E.g. `1` will predict the survival
function at each integer duration.
Values must be in the following range:
```javascript theme={null}
0 < step < inf
```
Maximum duration.
If not provided, or `null`, the maximum duration in the dataset is used.
Model parameters.
Level in the confidence intervals.
Penalizer strength.
Attach an L2 penalizer to the size of the coefficients during regression.
This improves stability of the estimates and controls for high correlation between covariates.
Values must be in the following range:
```javascript theme={null}
0.0 ≤ penalizer < inf
```
L1 vs L2 penalty ratio.
Specify what ratio to assign to a L1 vs L2 penalty (ridge vs lasso). Same as scikit-learn
convention.
Values must be in the following range:
```javascript theme={null}
0.0 ≤ l1_ratio ≤ 1.0
```
Columns to use in stratification.
This is useful if a categorical covariate does not obey the proportional hazard assumption.
Each item in array.
How the fitter should estimate the baseline.
Values must be one of the following:
* `breslow`
* `spline`
* `piecewise`