make PR + option out tmle3shift dep

tlverse · Mar 13, 2021 · 00bbeb4 · 00bbeb4
1 parent b0d2c22
commit 00bbeb4
Show file tree

Hide file tree

Showing 95 changed files with 3,892 additions and 1,703 deletions.
diff --git a/.travis.yml b/.travis.yml
@@ -36,6 +36,7 @@ r_github_packages:
   - r-lib/sessioninfo
   - r-lib/covr
   - tlverse/sl3
+  - tlverse/tmle3shift
 
 after_success:
   - travis_wait 80 Rscript -e 'covr::codecov()'

diff --git a/DESCRIPTION b/DESCRIPTION
@@ -15,10 +15,10 @@ Description: A general framework supporting the implementation of targeted
     target parameters through a unified interface. The goal is that the exposed
     framework be as general as the mathematical framework upon which it draws.
 Depends:
-    R (>= 3.2.2),
+    R (>= 3.6.0),
     graphics
 Imports:
-    sl3 (>= 1.3.7),
+    sl3 (>= 1.4.2),
     delayed,
     data.table,
     assertthat,
@@ -27,7 +27,6 @@ Imports:
     methods,
     ggplot2,
     stats,
-    Rsolnp,
     foreach,
     mvtnorm,
     magrittr,
@@ -40,15 +39,14 @@ Suggests:
     tableone,
     speedglm,
     rmarkdown,
+    Rsolnp,
     nnls,
     tmle,
-    tmle3shift,
     future,
     future.apply,
     xgboost
 Remotes:
-    github::tlverse/sl3,
-    github::tlverse/tmle3shift
+    github::tlverse/sl3
 License: GPL-3
 URL: https://tlverse.org/tmle3
 BugReports: https://github.com/tlverse/tmle3/issues

diff --git a/R/Param_MSM.R b/R/Param_MSM.R
@@ -53,10 +53,10 @@ Param_MSM <- R6Class(
   inherit = Param_base,
   public = list(
     initialize = function(observed_likelihood, strata_variable, strata_name = "V",
-                              msm = "A + V", weight = "Cond.Prob.", weight_ub = 1 / 0.025,
-                              continuous_treatment = FALSE, treatment_values = NULL, n_samples = 30,
-                              ...,
-                              covariate_node = "W", treatment_node = "A", outcome_node = "Y") {
+                          msm = "A + V", weight = "Cond.Prob.", weight_ub = 1 / 0.025,
+                          continuous_treatment = FALSE, treatment_values = NULL, n_samples = 30,
+                          ...,
+                          covariate_node = "W", treatment_node = "A", outcome_node = "Y") {
       super$initialize(observed_likelihood, ..., outcome_node = outcome_node)
       private$.strata_variable <- strata_variable
       private$.strata_name <- strata_name

diff --git a/R/Param_delta.R b/R/Param_delta.R
@@ -18,7 +18,7 @@ Param_delta <- R6Class(
   inherit = Param_base,
   public = list(
     initialize = function(observed_likelihood, delta_param, parent_parameters,
-                              ..., outcome_node = NA) {
+                          ..., outcome_node = NA) {
       super$initialize(observed_likelihood, ..., outcome_node = outcome_node)
       private$.delta_param <- delta_param
       private$.parent_parameters <- parent_parameters

diff --git a/R/tmle3_Fit.R b/R/tmle3_Fit.R
@@ -46,7 +46,7 @@ tmle3_Fit <- R6Class(
       print(self$summary)
     },
     set_timings = function(start_time, task_time, likelihood_time, params_time,
-                               fit_time) {
+                           fit_time) {
       timings <- list(
         make_tmle_task = task_time - start_time,
         fit_likelihood = likelihood_time - task_time,

diff --git a/R/tmle3_Node.R b/R/tmle3_Node.R
@@ -56,7 +56,7 @@ tmle3_Node <- R6Class(
   class = TRUE,
   public = list(
     initialize = function(name, variables, parents = c(),
-                              variable_type = NULL, censoring_node = NULL, scale = FALSE) {
+                          variable_type = NULL, censoring_node = NULL, scale = FALSE) {
       private$.name <- name
       private$.variables <- variables
       private$.parents <- parents

diff --git a/R/tmle3_Spec.R b/R/tmle3_Spec.R
@@ -12,7 +12,7 @@ tmle3_Spec <- R6Class(
   class = TRUE,
   public = list(
     initialize = function(likelihood_override = NULL,
-                              variable_types = NULL, ...) {
+                          variable_types = NULL, ...) {
       private$.options <- list(
         likelihood_override = likelihood_override,
         variable_types = variable_types, ...

diff --git a/R/tmle3_Spec_MSM.R b/R/tmle3_Spec_MSM.R
@@ -12,7 +12,7 @@ tmle3_Spec_MSM <- R6Class(
   inherit = tmle3_Spec,
   public = list(
     initialize = function(msm = "A + V", weight = "Cond.Prob.", weight_ub = 1 / 0.025,
-                              n_samples = 30, ...) {
+                          n_samples = 30, ...) {
       super$initialize(
         msm = msm, weight = weight, weight_ub = weight_ub, n_samples = n_samples, ...
       )

diff --git a/R/tmle3_Update.R b/R/tmle3_Update.R
@@ -51,12 +51,12 @@ tmle3_Update <- R6Class(
   public = list(
     # TODO: change maxit for test
     initialize = function(maxit = 100, cvtmle = TRUE, one_dimensional = FALSE,
-                              constrain_step = FALSE, delta_epsilon = 1e-4,
-                              convergence_type = c("scaled_var", "sample_size"),
-                              fluctuation_type = c("standard", "weighted"),
-                              optim_delta_epsilon = TRUE,
-                              use_best = FALSE,
-                              verbose = FALSE) {
+                          constrain_step = FALSE, delta_epsilon = 1e-4,
+                          convergence_type = c("scaled_var", "sample_size"),
+                          fluctuation_type = c("standard", "weighted"),
+                          optim_delta_epsilon = TRUE,
+                          use_best = FALSE,
+                          verbose = FALSE) {
       private$.maxit <- maxit
       private$.cvtmle <- cvtmle
       private$.one_dimensional <- one_dimensional
@@ -110,9 +110,9 @@ tmle3_Update <- R6Class(
       private$.step_number <- current_step
     },
     generate_submodel_data = function(likelihood, tmle_task,
-                                          fold_number = "full",
-                                          update_node = "Y",
-                                          drop_censored = FALSE) {
+                                      fold_number = "full",
+                                      update_node = "Y",
+                                      drop_censored = FALSE) {
 
       # TODO: change clever covariates to allow only calculating some nodes
       clever_covariates <- lapply(self$tmle_params, function(tmle_param) {

diff --git a/R/tmle3_Update_survival.R b/R/tmle3_Update_survival.R
@@ -47,12 +47,12 @@ tmle3_Update_survival <- R6Class(
   inherit = tmle3_Update,
   public = list(
     initialize = function(maxit = 100, cvtmle = TRUE, one_dimensional = FALSE,
-                              constrain_step = FALSE, delta_epsilon = 1e-4,
-                              convergence_type = c("scaled_var", "sample_size"),
-                              fluctuation_type = c("standard", "weighted"),
-                              use_best = FALSE,
-                              verbose = FALSE,
-                              fit_method = "l2") {
+                          constrain_step = FALSE, delta_epsilon = 1e-4,
+                          convergence_type = c("scaled_var", "sample_size"),
+                          fluctuation_type = c("standard", "weighted"),
+                          use_best = FALSE,
+                          verbose = FALSE,
+                          fit_method = "l2") {
       super$initialize(
         maxit = maxit, cvtmle = cvtmle,
         one_dimensional = one_dimensional,
@@ -74,49 +74,52 @@ tmle3_Update_survival <- R6Class(
         # TODO: check
         # print("l2")
         # mean_eic <- self$get_mean_eic(self$update_fold)
-        epsilon_n <- tryCatch({
-          alpha <- 0
-          norm_func <- self$norm_l2
-          lambda.min.ratio <- 1e-2
+        epsilon_n <- tryCatch(
+          {
+            alpha <- 0
+            norm_func <- self$norm_l2
+            lambda.min.ratio <- 1e-2
 
-          ind <- 1
-          while (ind == 1) {
-            submodel_fit <- glmnet::glmnet(
-              x = submodel_data$H,
-              y = submodel_data$observed,
-              offset = qlogis(submodel_data$initial),
-              family = "binomial",
-              alpha = alpha,
-              standardize = FALSE,
-              intercept = FALSE,
-              lambda.min.ratio = lambda.min.ratio,
-              # nlambda = 2e2
-              nlambda = 1e2
-              # TODO: check
-              # penalty.factor = 1/abs(mean_eic)
-            )
-            norms <- apply(submodel_fit$beta, 2, norm_func)
-            ind <- max(which(norms <= self$delta_epsilon))
-            if (ind > 1) break
+            ind <- 1
+            while (ind == 1) {
+              submodel_fit <- glmnet::glmnet(
+                x = submodel_data$H,
+                y = submodel_data$observed,
+                offset = qlogis(submodel_data$initial),
+                family = "binomial",
+                alpha = alpha,
+                standardize = FALSE,
+                intercept = FALSE,
+                lambda.min.ratio = lambda.min.ratio,
+                # nlambda = 2e2
+                nlambda = 1e2
+                # TODO: check
+                # penalty.factor = 1/abs(mean_eic)
+              )
+              norms <- apply(submodel_fit$beta, 2, norm_func)
+              ind <- max(which(norms <= self$delta_epsilon))
+              if (ind > 1) break
 
-            fit_lambda <- submodel_fit$lambda
+              fit_lambda <- submodel_fit$lambda
 
-            if (fit_lambda == 1) {
-              stop("only one lambda could be fit")
-            }
+              if (fit_lambda == 1) {
+                stop("only one lambda could be fit")
+              }
 
-            # try to estimate what the correct lambda value is and go a bit beyond that
-            norm_ratio <- self$delta_epsilon / norms[2]
-            lambda_guess <- fit_lambda[1] - norm_ratio * (fit_lambda[1] - fit_lambda[2])
-            lambda_min_ratio <- 0.8 * lambda_guess / fit_lambda[1]
-            # lambda.min.ratio <- sort(submodel_fit$lambda, decreasing = TRUE)[2] / max(submodel_fit$lambda)
+              # try to estimate what the correct lambda value is and go a bit beyond that
+              norm_ratio <- self$delta_epsilon / norms[2]
+              lambda_guess <- fit_lambda[1] - norm_ratio * (fit_lambda[1] - fit_lambda[2])
+              lambda_min_ratio <- 0.8 * lambda_guess / fit_lambda[1]
+              # lambda.min.ratio <- sort(submodel_fit$lambda, decreasing = TRUE)[2] / max(submodel_fit$lambda)
+            }
+            epsilon_n <- submodel_fit$beta[, ind]
+          },
+          error = function(e) {
+            # TODO: check
+            print(e)
+            return(rep(0, ncol(submodel_data$H)))
           }
-          epsilon_n <- submodel_fit$beta[, ind]
-        }, error = function(e) {
-          # TODO: check
-          print(e)
-          return(rep(0, ncol(submodel_data$H)))
-        })
+        )
         epsilon <- epsilon_n
         # TODO: check if necessary
         # NOTE: this protects against collinear covariates

diff --git a/README.md b/README.md
@@ -57,6 +57,13 @@ remotes::install_github("tlverse/tmle3")
 
 -----
 
+## Issues
+
+If you encounter any bugs or have any specific feature requests, please
+[file an issue](https://github.com/tlverse/tmle3/issues).
+
+-----
+
 ## Getting Started
 
 The best place to get started is the “Framework Overview” document,
@@ -65,16 +72,36 @@ may be found at <https://tlverse.org/tmle3/articles/framework.html>.
 
 -----
 
-## Issues
+## Contributions
 
-If you encounter any bugs or have any specific feature requests, please
-[file an issue](https://github.com/tlverse/tmle3/issues).
+Contributions are very welcome. Interested contributors should consult
+our [contribution
+guidelines](https://github.com/tlverse/tmle3/blob/master/CONTRIBUTING.md)
+prior to submitting a pull request.
+
+-----
+
+## Citation
+
+After using the `tmle3` R package, please cite the following:
+
+``` 
+    @software{coyle2021tmle3-rpkg,
+      author = {Coyle, Jeremy R},
+      title = {{tmle3}: The Extensible {TMLE} Framework},
+      year = {2021},
+      howpublished = {\url{https://github.com/tlverse/tmle3}},
+      note = {{R} package version 0.2.0},
+      url = {https://doi.org/},
+      doi = {}
+    }
+```
 
 -----
 
 ## License
 
-© 2017-2020 [Jeremy R. Coyle](https://github.com/jeremyrcoyle)
+© 2017-2021 [Jeremy R. Coyle](https://github.com/jeremyrcoyle)
 
 The contents of this repository are distributed under the GPL-3 license.
 See file `LICENSE` for details.

diff --git a/docs/404.html b/docs/404.html