Type: Package
Title: Deploy 'TensorFlow' Models
Version: 0.6.1
Maintainer: Daniel Falbel <daniel@rstudio.com>
Description: Tools to deploy 'TensorFlow' https://www.tensorflow.org/ models across multiple services. Currently, it provides a local server for testing 'cloudml' compatible services.
License: Apache License 2.0
Encoding: UTF-8
LazyData: true
Imports: httpuv, httr, jsonlite, magrittr, reticulate, swagger, tensorflow
Suggests: cloudml, knitr, pixels, processx, testthat, yaml, stringr
RoxygenNote: 6.1.1
VignetteBuilder: knitr
NeedsCompilation: no
Packaged: 2019-06-13 18:26:35 UTC; dfalbel
Author: Javier Luraschi [aut, ctb], Daniel Falbel [cre, ctb], RStudio [cph]
Repository: CRAN
Date/Publication: 2019-06-14 16:30:03 UTC

Load a SavedModel

Description

Loads a SavedModel using the given TensorFlow session and returns the model's graph.

Usage

load_savedmodel(sess = NULL, model_dir = NULL)

Arguments

sess

The TensorFlow session. NULL if using Eager execution.

model_dir

The path to the exported model, as a string. Defaults to a "savedmodel" path or the latest training run.

Details

Loading a model improves performance over multiple predict_savedmodel() calls.

See Also

export_savedmodel(), predict_savedmodel()

Examples

## Not run: 
# start session
sess <- tensorflow::tf$Session()

# preload an existing model into a TensorFlow session
graph <- tfdeploy::load_savedmodel(
  sess,
  system.file("models/tensorflow-mnist", package = "tfdeploy")
)

# perform prediction based on a pre-loaded model
tfdeploy::predict_savedmodel(
  list(rep(9, 784)),
  graph
)

# close session
sess$close()

## End(Not run)

Predict using a SavedModel

Description

Runs a prediction over a saved model file, web API or graph object.

Usage

predict_savedmodel(instances, model, ...)

Arguments

instances

A list of prediction instances to be passed as input tensors to the service. Even for single predictions, a list with one entry is expected.

model

The model as a local path, a REST url or graph object.

A local path can be exported using export_savedmodel(), a REST URL can be created using serve_savedmodel() and a graph object loaded using load_savedmodel().

A type parameter can be specified to explicitly choose the type model performing the prediction. Valid values are export, webapi and graph.

...

See predict_savedmodel.export_prediction(), predict_savedmodel.graph_prediction(), predict_savedmodel.webapi_prediction() for additional options.

#' @section Implementations:

See Also

export_savedmodel(), serve_savedmodel(), load_savedmodel()

Examples

## Not run: 
# perform prediction based on an existing model
tfdeploy::predict_savedmodel(
  list(rep(9, 784)),
  system.file("models/tensorflow-mnist", package = "tfdeploy")
)

## End(Not run)

Predict using an Exported SavedModel

Description

Performs a prediction using a locally exported SavedModel.

Usage

## S3 method for class 'export_prediction'
predict_savedmodel(instances, model,
  signature_name = "serving_default", ...)

Arguments

instances

A list of prediction instances to be passed as input tensors to the service. Even for single predictions, a list with one entry is expected.

model

The model as a local path, a REST url or graph object.

A local path can be exported using export_savedmodel(), a REST URL can be created using serve_savedmodel() and a graph object loaded using load_savedmodel().

A type parameter can be specified to explicitly choose the type model performing the prediction. Valid values are export, webapi and graph.

signature_name

The named entry point to use in the model for prediction.

...

See predict_savedmodel.export_prediction(), predict_savedmodel.graph_prediction(), predict_savedmodel.webapi_prediction() for additional options.

#' @section Implementations:

Predict using a Loaded SavedModel

Description

Performs a prediction using a SavedModel model already loaded using load_savedmodel().

Usage

## S3 method for class 'graph_prediction'
predict_savedmodel(instances, model, sess,
  signature_name = "serving_default", ...)

Arguments

instances

A list of prediction instances to be passed as input tensors to the service. Even for single predictions, a list with one entry is expected.

model

The model as a local path, a REST url or graph object.

A local path can be exported using export_savedmodel(), a REST URL can be created using serve_savedmodel() and a graph object loaded using load_savedmodel().

A type parameter can be specified to explicitly choose the type model performing the prediction. Valid values are export, webapi and graph.

sess

The active TensorFlow session.

signature_name

The named entry point to use in the model for prediction.

...

See predict_savedmodel.export_prediction(), predict_savedmodel.graph_prediction(), predict_savedmodel.webapi_prediction() for additional options.

#' @section Implementations:

Predict using a Web API

Description

Performs a prediction using a Web API providing a SavedModel.

Usage

## S3 method for class 'webapi_prediction'
predict_savedmodel(instances, model, ...)

Arguments

instances

A list of prediction instances to be passed as input tensors to the service. Even for single predictions, a list with one entry is expected.

model

The model as a local path, a REST url or graph object.

A local path can be exported using export_savedmodel(), a REST URL can be created using serve_savedmodel() and a graph object loaded using load_savedmodel().

A type parameter can be specified to explicitly choose the type model performing the prediction. Valid values are export, webapi and graph.

...

See predict_savedmodel.export_prediction(), predict_savedmodel.graph_prediction(), predict_savedmodel.webapi_prediction() for additional options.

#' @section Implementations:

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

magrittr

%>%

tensorflow

export_savedmodel, view_savedmodel

Serve a SavedModel

Description

Serve a TensorFlow SavedModel as a local web api.

Usage

serve_savedmodel(model_dir, host = "127.0.0.1", port = 8089,
  daemonized = FALSE, browse = !daemonized)

Arguments

model_dir

The path to the exported model, as a string.

host

Address to use to serve model, as a string.

port

Port to use to serve model, as numeric.

daemonized

Makes 'httpuv' server daemonized so R interactive sessions are not blocked to handle requests. To terminate a daemonized server, call 'httpuv::stopDaemonizedServer()' with the handle returned from this call.

browse

Launch browser with serving landing page?

See Also

export_savedmodel()

Examples

## Not run: 
# serve an existing model over a web interface
tfdeploy::serve_savedmodel(
  system.file("models/tensorflow-mnist", package = "tfdeploy")
)

## End(Not run)