write to s3 method

NAMESPACE

@@ -7,7 +7,8 @@ export(batch_predict)
 export(default_bucket)
 export(download_file)
 export(install_sagemaker)
-export(s3_path)
+export(read_s3)
+export(s3)
 export(sagemaker_attach_tuner)
 export(sagemaker_delete_endpoint)
 export(sagemaker_deploy_endpoint)
@@ -21,4 +22,5 @@ export(sagemaker_tuning_job_logs)
 export(sagemaker_xgb_container)
 export(sagemaker_xgb_estimator)
 export(upload_file)
+export(write_s3)
 importFrom(magrittr,"%>%")

R/s3.R

@@ -4,35 +4,35 @@ default_bucket <- function() {
 }
 
 #' @export
-upload_file <- function(file, bucket = default_bucket(), key) {
+upload_file <- function(file, s3_path) {
   s3 <- boto3$client('s3')
-  s3$upload_file(file, bucket, key)
+
+  s3_components <- s3_bucket_key_extract(s3_path)
+  s3$upload_file(file, s3_components$bucket, s3_components$key)
 }
 
 
 #' @export
-download_file <- function(s3_path_name, file_name) {
-  # TODO: is it possible to download to R session, without
-  #       going through a file?
-
+download_file <- function(s3_path, file) {
   system(
     paste0(
       "aws s3 cp ",
-      s3_path_name(), " ",
-      file_name
+      s3_path, " ",
+      file
     )
   )
 }
 
 # downloads s3 object into the R session
 # ... passes to readr::read_delim
 #' @export
-read_s3 <- function(s3_path_name, delim = ",", col_names = FALSE, ...) {
-  s3_components <- s3_bucket_key_extract(s3_path_name)
+read_s3 <- function(s3_path, delim = ",", col_names = FALSE, ...) {
+  s3_components <- s3_bucket_key_extract(s3_path)
 
   io <- reticulate::import("io")
   s3 <- boto3$client('s3')
 
+  # https://datasciencechronicles.com.au/2017/11/12/adventures-in-python-1/
   file <- io$BytesIO()
   s3$download_fileobj(s3_components$bucket, s3_components$key, file)
 
@@ -47,13 +47,35 @@ read_s3 <- function(s3_path_name, delim = ",", col_names = FALSE, ...) {
   s3_obj
 }
 
+# ... to format_csv
+#' @export
+write_s3 <- function(x, s3_path, delim = ",", col_names = FALSE, ...) {
+  s3_components <- s3_bucket_key_extract(s3_path)
+
+  io <- reticulate::import("io")
+  builtin <- reticulate::import_builtins()
+  s3 <- boto3$client('s3')
+
+  # https://datasciencechronicles.com.au/2017/11/12/adventures-in-python-1/
+  file <- io$BytesIO(builtin$bytes(
+    readr::format_delim(
+      x, delim = delim, col_names = col_names, ...
+    ),
+    "utf-8"
+  ))
+
+  s3$upload_fileobj(file, s3_components$bucket, s3_components$key)
+}
+
+
+
 
 #' @export
-s3_path <- function(...) {
+s3 <- function(...) {
   path <- file.path(..., fsep = "/") %>%
     paste0("s3://", .)
 
-  class(path) <- c("s3_path", class(path))
+  class(path) <- c("s3", class(path))
   path
 }
 

R/xgboost.R

@@ -59,7 +59,7 @@ sagemaker_estimator <- function(
   role = sagemaker_get_execution_role(),
   train_instance_count = 1L,
   train_instance_type = "ml.m4.xlarge",
-  output_path = s3_path(default_bucket(), "models/"),
+  output_path = s3(default_bucket(), "models/"),
   sagemaker_session = sagemaker$Session()
 
 ) {
@@ -403,8 +403,8 @@ batch_predict <- function(
     logs = FALSE
   )
 
-  # TODO: make s3_path generic so it knows how
-  #       to take a s3_path object and not double
+  # TODO: make s3 generic so it knows how
+  #       to take a s3 object and not double
   #       transform it like "s3://s3://".
   #       I think there are some examples in Shiny
   #       or htmltools.

vignettes/random-cv2.Rmd

@@ -117,6 +117,16 @@ sagemaker_has_endpoint(model)
 read_s3(s3_path(default_bucket(), "abalone-train.csv"))
 ```
 
+```{r}
+sagemaker::abalone %>%
+  write_s3(s3(default_bucket(), "test-upload.csv"))
+```
+
+```{r}
+read_s3(s3(default_bucket(), "test-upload.csv"))
+```
+
+
 
 ```{r}
 tuner <- sagemaker$tuner$HyperparameterTuner$attach("xgboost-191114-2052")