proportional_hazards_glmnet.md
In parsnip: A Common API to Modeling and Analysis Functions

For this engine, there is a single mode: censored regression

This model has 2 tuning parameters:

penalty: Amount of Regularization (type: double, default: see below)
mixture: Proportion of Lasso Penalty (type: double, default: 1.0)

The penalty parameter has no default and requires a single numeric value. For more details about this, and the glmnet model in general, see [parsnip::glmnet-details]. As for mixture:

mixture = 1 specifies a pure lasso model,
mixture = 0 specifies a ridge regression model, and
0 < mixture < 1 specifies an elastic net model, interpolating lasso and ridge.

The censored extension package is required to fit this model.

library(censored)

proportional_hazards(penalty = double(1), mixture = double(1)) %>% 
  set_engine("glmnet") %>% 
  translate()

## Proportional Hazards Model Specification (censored regression)
## 
## Main Arguments:
##   penalty = 0
##   mixture = double(1)
## 
## Computational engine: glmnet 
## 
## Model fit template:
## censored::coxnet_train(formula = missing_arg(), data = missing_arg(), 
##     weights = missing_arg(), alpha = double(1))

Factor/categorical predictors need to be converted to numeric values (e.g., dummy or indicator variables) for this engine. When using the formula method via \code{\link[=fit.model_spec]{fit()}}, parsnip will convert factor columns to indicators.

Predictors should have the same scale. One way to achieve this is to center and scale each so that each predictor has mean zero and a variance of one. By default, [glmnet::glmnet()] uses the argument standardize = TRUE to center and scale the data.

The model does not fit an intercept.

The model formula (which is required) can include special terms, such as [survival::strata()]. This allows the baseline hazard to differ between groups contained in the function. The column used inside strata() is treated as qualitative no matter its type. This is different than the syntax offered by the [glmnet::glmnet()] package (i.e., [glmnet::stratifySurv()]) which is not recommended here.

For example, in this model, the numeric column rx is used to estimate two different baseline hazards for each value of the column:

library(survival)
library(censored)
library(dplyr)
library(tidyr)

mod <- 
  proportional_hazards(penalty = 0.01) %>% 
  set_engine("glmnet", nlambda = 5) %>% 
  fit(Surv(futime, fustat) ~ age + ecog.ps + strata(rx), data = ovarian)

pred_data <- data.frame(age = c(50, 50), ecog.ps = c(1, 1), rx = c(1, 2))

# Different survival probabilities for different values of 'rx'
predict(mod, pred_data, type = "survival", time = 500) %>% 
  bind_cols(pred_data) %>% 
  unnest(.pred)

## # A tibble: 2 × 5
##   .time .pred_survival   age ecog.ps    rx
##   <dbl>          <dbl> <dbl>   <dbl> <dbl>
## 1   500          0.666    50       1     1
## 2   500          0.769    50       1     2

Note that columns used in the strata() function will also be estimated in the regular portion of the model (i.e., within the linear predictor).

Predictions of type "time" are predictions of the mean survival time.

Since risk regression and parametric survival models are modeling different characteristics (e.g. relative hazard versus event time), their linear predictors will be going in opposite directions.

For example, for parametric models, the linear predictor increases with time. For proportional hazards models the linear predictor decreases with time (since hazard is increasing). As such, the linear predictors for these two quantities will have opposite signs.

tidymodels does not treat different models differently when computing performance metrics. To standardize across model types, the default for proportional hazards models is to have increasing values with time. As a result, the sign of the linear predictor will be the opposite of the value produced by the predict() method in the engine package.

This behavior can be changed by using the increasing argument when calling predict() on a \pkg{parsnip} model object.

This model can utilize case weights during model fitting. To use them, see the documentation in [case_weights] and the examples on tidymodels.org.

The fit() and fit_xy() arguments have arguments called case_weights that expect vectors of case weights.

This model object contains data that are not required to make predictions. When saving the model for the purpose of prediction, the size of the saved object might be substantially reduced by using functions from the butcher package.

References

Simon N, Friedman J, Hastie T, Tibshirani R. 2011. "Regularization Paths for Cox’s Proportional Hazards Model via Coordinate Descent." Journal of Statistical Software, Articles 39 (5): 1–13. \doi{10.18637/jss.v039.i05}.
Hastie T, Tibshirani R, Wainwright M. 2015. Statistical Learning with Sparsity. CRC Press.
Kuhn M, Johnson K. 2013. Applied Predictive Modeling. Springer.