Package 'autovi'

Description

This is the class of auto visual inference, inherited from bandicoot::BASE. It is an environment with S3 class bandicoot_oop.

Usage

auto_vi(
  fitted_model,
  keras_model = NULL,
  data = NULL,
  node_index = 1L,
  env = new.env(parent = parent.frame()),
  init_call = sys.call()
)

residual_checker(
  fitted_model,
  keras_model_name = "vss_phn_32",
  data = NULL,
  node_index = 1L,
  env = new.env(parent = parent.frame()),
  init_call = sys.call()
)

Arguments

Value

An instance environment.

Functions

• auto_vi(): Class constructor, same as AUTO_VI$instantiate().

• residual_checker(): Class constructor, same as AUTO_VI$instantiate(), but uses the keras_model_name argument rather than keras_model.

Class information

Parent classes

• Direct:

  • bandicoot::BASE

New attributes

• C:

  • AUTO_VI$check_result

New methods

• A:

  • AUTO_VI$auxiliary()

• B:

  • AUTO_VI$boot_method()

  • AUTO_VI$boot_vss()

• C:

  • AUTO_VI$check()

• F:

  • AUTO_VI$feature_pca()

  • AUTO_VI$feature_pca_plot()

• G:

  • AUTO_VI$get_data()

  • AUTO_VI$get_fitted_and_resid()

• I:

  • AUTO_VI$..init..()

• L:

  • AUTO_VI$lineup_check()

  • AUTO_VI$likelihood_ratio()

• N:

  • AUTO_VI$null_method()

  • AUTO_VI$null_vss()

• P:

  • AUTO_VI$p_value()

  • AUTO_VI$plot_pair()

  • AUTO_VI$plot_lineup()

  • AUTO_VI$plot_resid()

• R:

  • AUTO_VI$rotate_resid()

• S:

  • AUTO_VI$save_plot()

  • AUTO_VI$..str..()

  • AUTO_VI$summary()

  • AUTO_VI$summary_density_plot()

  • AUTO_VI$summary_plot()

  • AUTO_VI$summary_rank_plot()

• V:

  • AUTO_VI$vss()

Description

This function will be called after an instance is built. User input will be stored in the environment.

Usage

AUTO_VI$..init..(fitted_model, keras_model = NULL, data = NULL, node_index = 1L)

Arguments

Value

Return the object itself.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi

Description

This function returns a string representation of the object.

Usage

AUTO_VI$..str..()

Value

A string.

Examples

AUTO_VI$..str..()

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$..str..()

Description

This function computes auxiliary variables including monotonic measure (measure_monotonic), sparse measure (measure_sparse), splines measure (measure_splines), striped measure (measure_striped), and the number of observation (n). Scagnostics are computed using cassowaryr::sc_monotonic(), cassowaryr::sc_sparse2(), cassowaryr::sc_splines(), and cassowaryr::sc_striped().

If you wish to calculate additional auxiliary variables for your keras model, please override this method. Ensure that it accepts a data frame with columns named .fitted and .resid as input and returns a single row tibble.

Usage

AUTO_VI$auxiliary(data = seflf$get_fitted_and_resid())

Arguments

Value

A tibble.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$auxiliary()

Description

This default method gets bootstrapped residuals from a fitted linear model by sampling the observations with replacement then refit the model. User needs to override this method if a different bootstrapping scheme is needed.

Usage

AUTO_VI$boot_method(
  fitted_model = self$fitted_model,
  data = self$get_data()
)

Arguments

Value

A tibble with two columns .fitted and .resid.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
null_resid <- my_vi$boot_method()
my_vi$plot_resid(null_resid)

Description

This function bootstrap the data and refits the model by using AUTO_VI$boot_method(), then predicts the visual signal strength of the bootstrapped residual plots.

Usage

AUTO_VI$boot_vss(
  draws = 100L,
  fitted_model = self$fitted_model,
  keras_model = self$keras_model,
  data = self$get_data(),
  node_index = 1L,
  keep_boot_data = FALSE,
  keep_boot_plot = FALSE,
  extract_feature_from_layer = NULL
)

Arguments

Value

A tibble.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$boot_vss()
}

Description

This function conducts a visual inference check with a computer vision model. The result will be stored in self$check_result.

Usage

AUTO_VI$check(
  null_draws = 100L,
  boot_draws = 100L,
  fitted_model = self$fitted_model,
  keras_model = self$keras_model,
  null_method = self$null_method,
  data = self$get_data(),
  node_index = self$node_index,
  keep_data = FALSE,
  keep_plot = FALSE,
  extract_feature_from_layer = NULL
)

Arguments

Value

Return the object itself.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$check()
  myvi
}

Description

A list, will be initialized after the method AUTO_VI$check() is run.

Description

This function conducts principal component analysis for features extracted from keras model.

Usage

AUTO_VI$feature_pca(
  feature = self$check_result$observed,
  null_feature = self$check_result$null,
  boot_feature = self$check_result$boot,
  center = TRUE,
  scale = TRUE,
  pattern = "^f_.*$"
)

Arguments

Details

Features need to be extracted while running the method AUTO_VI$check() and AUTO_VI$lineup_check() by providing the argument extract_feature_from_layer. Features with zero variance will be ignored from the analysis. See also stats::prcomp(). By default, features are assumed to follow the naming convention "f_(index)", where index is from one to the number of features.

Value

A tibble of the raw features and the rotated features with attributes sdev and rotation representing the standard deviation of the principal components and the rotation matrix respectively.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$lineup_check(extract_feature_from_layer = "global_max_pooling2d")
  myvi$feature_pca()
}

Description

This function draws a summary Plot for principal component analysis conducted on extracted features

Usage

AUTO_VI$feature_pca_plot(
  feature_pca = self$feature_pca(),
  x = PC1,
  y = PC2,
  col_by_set = TRUE)

Arguments

Details

By default, it will visualize PC2 vs PC1. User can choose to visualize other principal components.

Value

A ggplot.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$lineup_check(extract_feature_from_layer = "global_max_pooling2d")
  myvi$feature_pca_plot()
}

Description

This function gets the data out of a model object by using stats::model.frame() if self$data is NULL.

Usage

AUTO_VI$get_data(fitted_model = self$fitted_model)

Arguments

Value

A tibble.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$get_data()

Description

This function gets fitted values and residuals out of a model object by using stats::fitted() and stats::resid().

Usage

AUTO_VI$get_fitted_and_resid(fitted_model = self$fitted_model)

Arguments

Value

A tibble.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$get_fitted_and_resid()

Description

This function estimates the likelihood of observing the visual signal strength in terms of the bootstrapped distribution and the simulated null distribution, and computes the ratio between these two likelihood.

Usage

AUTO_VI$likelihood_ratio(
  vss = self$check_result$observed$vss,
  dist_1 = self$check_result$boot$vss,
  dist_2 = self$check_result$null$vss
)

Arguments

Value

A named vector with three elements likelihood_1, likelihood_2 and likelihood_ratio.

Examples

dist_1 <- rnorm(100, 0, 1)
dist_2 <- rnorm(100, 1, 1)
AUTO_VI$likelihood_ratio(0, dist_1, dist_2)

Description

This function conducts a visual inference lineup check with a computer vision model. The result will be stored in self$check_result.

Usage

AUTO_VI$lineup_check(
  lineup_size = 20L,
  fitted_model = self$fitted_model,
  keras_model = self$keras_model,
  null_method = self$null_method,
  data = self$get_data(),
  node_index = self$node_index,
  extract_feature_from_layer = NULL
)

Arguments

Value

Return the object itself.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$lineup_check()
  myvi
}

Description

This default method gets rotated residuals from a fitted linear model using AUTO_VI$rotate_resid. User needs to override this method if the fitted model is not a linear regression model.

Usage

AUTO_VI$null_method(fitted_model = self$fitted_model)

Arguments

Value

A tibble with two columns .fitted and .resid.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
null_resid <- my_vi$null_method()
my_vi$plot_resid(null_resid)

Description

This function simulates null plots from the null hypothesis distribution, and predicts the visual signal strength.

Usage

AUTO_VI$null_vss(
  draws = 100L,
  fitted_model = self$fitted_model,
  keras_model = self$keras_model,
  null_method = self$null_method,
  node_index = self$node_index,
  keep_null_data = FALSE,
  keep_null_plot = FALSE,
  extract_feature_from_layer = NULL
)

Arguments

Value

A tibble.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$null_vss()
}

Description

This function computes the p-value of observing the visual signal strength of the original residual plot based on the null distribution.

Usage

AUTO_VI$p_value(
  vss = self$check_result$observed$vss,
  null_dist = self$check_result$null$vss
)

Arguments

Value

A numeric value representing the desired p-value.

Examples

vss <- 1
null_dist <- rnorm(100, 0, 1)
AUTO_VI$p_value(vss, null_dist)

Description

This function draws a lineup of standard residual plots consisting of a true residual plot and several null residual plots.

Usage

AUTO_VI$plot_lineup(
  lineup_size = 20L,
  data = self$get_fitted_and_resid(),
  null_method = self$null_method,
  theme = ggplot2::theme_light(),
  alpha = 1,
  size = 0.5,
  stroke = 0.5,
  remove_axis = TRUE,
  remove_legend = TRUE,
  remove_grid_line = TRUE,
  add_zero_line = TRUE,
  remove_facet_label = FALSE,
  display_answer = TRUE
)

Arguments

Value

A ggplot.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$plot_lineup()

Description

This function draws a pair of standard residual plots consisting of a true residual plot and a null residual plot.

Usage

AUTO_VI$plot_pair(
  data = self$get_fitted_and_resid(),
  null_data = self$null_method(),
  theme = ggplot2::theme_light(),
  alpha = 1,
  size = 0.5,
  stroke = 0.5,
  remove_axis = TRUE,
  remove_legend = TRUE,
  remove_grid_line = TRUE,
  add_zero_line = TRUE
)

Arguments

Value

A ggplot.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$plot_pair()

Description

This function draws a standard residual plot.

Usage

AUTO_VI$plot_resid(
  data = self$get_fitted_and_resid(),
  theme = ggplot2::theme_light(base_size = 11/5),
  alpha = 1,
  size = 0.5,
  stroke = 0.5,
  remove_axis = TRUE,
  remove_legend = TRUE,
  remove_grid_line = TRUE,
  add_zero_line = TRUE
)

Arguments

Value

A ggplot.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
my_vi$plot_resid()

Description

This function gets rotated residuals from a fitted linear model. The rotated residuals are generated by first regressing random noises on the original regressors, then multiply the obtained residuals by original RSS divided by the current RSS. The results are the rotated residuals.

Usage

AUTO_VI$rotate_resid(fitted_model = self$fitted_mod)

Arguments

Value

A tibble with two columns .fitted and .resid.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
rotated_resid <- my_vi$rotate_resid()
my_vi$plot_resid(rotated_resid)

Description

This is the default method of saving plot(s). It will use save_plot() to save the ggplot to a 420 (width) * 525 (height) PNG file. If the trained images are generated differently, one can override this method using bandicoot::register_method().

Usage

AUTO_VI$save_plot(p, path = NULL)

Arguments

Value

The image path.

Examples

my_vi <- auto_vi(fitted_model = lm(speed ~ dist, data = cars))
p <- my_vi$plot_resid()
my_vi$save_plot(p)

Description

The AUTO_VI$..str..() method provides a string representation of the object. If a check is performed, the string will contain some simple statistics of the check result. This method does this same thing as AUTO_VI$..str..(), but it returns an AUTO_VI_SUMMARY object which stores those statistics, such as sample quantiles of the distribution of null visual signal strength, in the object.

Usage

AUTO_VI$summary()

Value

An AUTO_VI_SUMMARY object.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$check()
  myvi_summary <- myvi$summary()
  print(myvi_summary)
  names(myvi_summary)
}

Description

This function draws a summary density plot for the result.

Usage

AUTO_VI$summary_plot(
  vss = self$check_result$observed$vss,
  null_dist = self$check_result$null$vss,
  boot_dist = self$check_result$boot$vss,
  p_value = self$check_result$p_value,
  likelihood_ratio = self$check_result$likelihood_ratio,
  density_alpha = 0.6
)

Arguments

Value

A ggplot.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$check()
  myvi$summary_density_plot()
}

Description

This function draws a summary plot for the result.

Usage

AUTO_VI$summary_plot(type = "auto", ...)

Arguments

Value

A ggplot.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$lineup_check()
  myvi$summary_plot()
}

Description

This function draws a summary rank plot for the result.

Usage

AUTO_VI$summary_plot(
  vss = self$check_result$observed$vss,
  null_dist = self$check_result$null$vss,
  p_value = self$check_result$p_value
)

Arguments

Value

A ggplot.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$lineup_check()
  myvi$summary_rank_plot()
}

Description

This function predicts the visual signal strength.

Usage

AUTO_VI$vss(
  p = self$plot_resid(),
  auxiliary = NULL,
  keras_model = self$keras_model,
  node_index = self$node_index,
  extract_feature_from_layer = NULL
)

Arguments

Value

A tibble. The first column is vss which is the prediction, the rest of the columns are features extracted from a layer.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  myvi <- auto_vi(lm(dist ~ speed, data = cars), keras_model)

  myvi$vss()
}

Description

This function checks if a python library is available. If the library can not be found by the importlib.util.find_spec method, then an error will be throw.

Usage

check_python_library_available(lib_name)

Arguments

Value

No return. Called for side-effect.

Examples

try(check_python_library_available("numpy"))

Description

This functions download the keras model from the TengMCing/autovi_data Github repo using download.file() and load the model.

Usage

get_keras_model(model_name, format = "npz")

Arguments

Details

Note that the "SavedModel" and "keras" formats are not supported in tensorflow versions above 2.15, as reticulate::import("tensorflow")$keras$models$load_model encounters issues when loading models saved with the Keras 2 API. Instead, using the "npz" format allows for rebuilding the model from scratch and loading the weights from a ".npz" file, offering a more reliable alternative.

Value

A keras model.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) keras_model$summary()

Description

This is the class of keras wrapper, inherited from bandicoot::BASE. It is an environment with S3 class bandicoot_oop.

Usage

keras_wrapper(
  keras_model = NULL,
  node_index = 1L,
  env = new.env(parent = parent.frame()),
  init_call = sys.call()
)

Arguments

Value

An instance environment.

Functions

• keras_wrapper(): Class constructor, same as KERAS_WRAPPER$instantiate().

Class information

Parent classes

• Direct:

  • bandicoot::BASE

New methods

• G:

  • KERAS_WRAPPER$get_input_height()

  • KERAS_WRAPPER$get_input_width()

• I:

  • KERAS_WRAPPER$image_to_array()

  • KERAS_WRAPPER$..init..()

• L:

  • KERAS_WRAPPER$list_layer_name()

• P:

  • KERAS_WRAPPER$predict()

• S:

  • KERAS_WRAPPER$..str..()

Description

This function will be called after an instance is built. User input will be stored in the environment.

Usage

KERAS_WRAPPER$..init..(keras_mod = NULL, node_index = 1L)

Arguments

Value

Return the object itself.

Examples

keras_wrapper()

Description

This function returns a string representation of the object.

Usage

KERAS_WRAPPER$..str..()

Value

A string.

Examples

KERAS_WRAPPER$..str..()

wrapper <- keras_wrapper()
wrapper$..str..()

Description

This function get the input image height (the input shape is (batch_size, height, width, channels)) of a keras model.

Usage

KERAS_WRAPPER$get_input_height(keras_model = self$keras_model)

Arguments

Value

An integer.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  keras_wrapper(keras_model)$get_input_height()
}

Description

This function get the input image width (the input shape is (batch_size, height, width, channels)) of a keras model.

Usage

KERAS_WRAPPER$get_input_width(keras_model = self$keras_model)

Arguments

Value

An integer.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  keras_wrapper(keras_model)$get_input_width()
}

Description

This function loads an image from file and convert it to a numpy array.

Usage

KERAS_WRAPPER$image_to_array(
  path,
  height = self$get_input_height(),
  width = self$get_input_width()
)

Arguments

Value

A numpy array.

Examples

p <- ggplot2::ggplot(cars) + ggplot2::geom_point(ggplot2::aes(dist, speed))
path <- save_plot(p)
result <- try(KERAS_WRAPPER$image_to_array(path, 32L, 32L))
if (!inherits(result, "try-error")) {
  result
}

Description

This function list all layer names of a keras model.

Usage

KERAS_WRAPPER$list_layer_name(keras_model = self$keras_model)

Arguments

Value

A vector of strings.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  keras_wrapper(keras_model)$list_layer_name()
}

Description

This function predicts the visual signal strength using the provided keras model, input array and optional auxiliary input array.

Usage

KERAS_WRAPPER$predict(
  input_array,
  auxiliary = NULL,
  keras_model = self$keras_model,
  node_index = self$node_index,
  extract_featrue_from_layer = NULL
)

Arguments

Value

A tibble. The first column is vss which is the prediction, the rest of the columns are features extracted from a layer.

Examples

keras_model <- try(get_keras_model("vss_phn_32"))
if (!inherits(keras_model, "try-error")) {
  wrapper <- keras_wrapper(keras_model)

  # Provide one 32 * 32 RGB image and one vector of length 5 as input
  wrapper$predict(input_array = array(255, dim = c(1, 32, 32, 3)),
                  auxiliary = matrix(1, ncol = 5))
}

Description

This function gets a table of available pre-trained computer vision models for predicting visual signal strength.

Usage

list_keras_model()

Value

A tibble of available model names and paths.

Examples

list_keras_model()

Description

This function removes a plot from a provided path.

Usage

remove_plot(path, check_ext = TRUE)

Arguments

Value

No return. Called for side-effect.

Examples

p <- ggplot2::ggplot(cars) + ggplot2::geom_point(ggplot2::aes(dist, speed))
path <- save_plot(p)
remove_plot(path)

Description

This function save a plot of a list of plots to provided path(s).

Usage

save_plot(p, path = NULL, width = 7/5, height = 7/4, ...)

Arguments

Value

The image path(s).

Examples

p <- ggplot2::ggplot(cars) + ggplot2::geom_point(ggplot2::aes(dist, speed))
save_plot(p)