Title: | Multiple-Instance Learning with Support Vector Machines |
---|---|
Description: | Weakly supervised (WS), multiple instance (MI) data lives in numerous interesting applications such as drug discovery, object detection, and tumor prediction on whole slide images. The 'mildsvm' package provides an easy way to learn from this data by training Support Vector Machine (SVM)-based classifiers. It also contains helpful functions for building and printing multiple instance data frames. The core methods from 'mildsvm' come from the following references: Kent and Yu (2022) <arXiv:2206.14704>; Xiao, Liu, and Hao (2018) <doi:10.1109/TNNLS.2017.2766164>; Muandet et al. (2012) <https://proceedings.neurips.cc/paper/2012/file/9bf31c7ff062936a96d3c8bd1f8f2ff3-Paper.pdf>; Chu and Keerthi (2007) <doi:10.1162/neco.2007.19.3.792>; and Andrews et al. (2003) <https://papers.nips.cc/paper/2232-support-vector-machines-for-multiple-instance-learning.pdf>. Many functions use the 'Gurobi' optimization back-end to improve the optimization problem speed; the 'gurobi' R package and associated software can be downloaded from <https://www.gurobi.com> after obtaining a license. |
Authors: | Sean Kent [aut, cre] , Yifei Liou [aut] |
Maintainer: | Sean Kent <[email protected]> |
License: | MIT + file LICENSE |
Version: | 0.4.0.9000 |
Built: | 2024-11-21 02:49:14 UTC |
Source: | https://github.com/skent259/mildsvm |
as_mi_df()
turns an existing object, such as a data frame, into a MI
data frame, a data frame with 'mi_df'. This is in contrast with
mi_df()
, which builds a MI data frame from individual columns.
as_mi_df( x, bag_label = "bag_label", bag_name = "bag_name", instance_label = "instance_label", ... )
as_mi_df( x, bag_label = "bag_label", bag_name = "bag_name", instance_label = "instance_label", ... )
x |
A data-frame or similar to convert. |
bag_label |
A character (default |
bag_name |
A character (default |
instance_label |
A character (default |
... |
Arguments reserved for other methods. |
A 'mi_df' object. This data.frame-like has columns bag_label
,
bag_name
, and potentially others. It also inherits from the
'tbl_df'
and 'tbl'
classes.
Sean Kent
mi_df()
to build a mi_df
object.
x = data.frame('bag_LABEL' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0)) df <- as_mi_df(x)
x = data.frame('bag_LABEL' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0)) df <- as_mi_df(x)
as_mild_df()
turns an existing object, such as a data frame, into a MILD
data frame, a data frame with 'mild_df'. This is in contrast with
mild_df()
, which builds a MILD data frame from individual columns.
as_mild_df( x, bag_label = "bag_label", bag_name = "bag_name", instance_name = "instance_name", instance_label = "instance_label", ... )
as_mild_df( x, bag_label = "bag_label", bag_name = "bag_name", instance_name = "instance_name", instance_label = "instance_label", ... )
x |
A data-frame or similar to convert. |
bag_label |
A character (default |
bag_name |
A character (default |
instance_name |
A character (default |
instance_label |
A character (default |
... |
Arguments reserved for other methods. |
A 'mild_df' object. This data.frame-like has columns bag_label
,
bag_name
, instance_name
, and potentially others. It also inherits from
the 'tbl_df'
and 'tbl'
classes.
Sean Kent
mild_df()
to build a mild_df
object.
x <- data.frame('bag_LABEL' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'instance_name' = c('bag_1_inst_1', 'bag_1_inst_2', 'bag_2_inst_1'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0)) df <- as_mild_df(x)
x <- data.frame('bag_LABEL' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'instance_name' = c('bag_1_inst_1', 'bag_1_inst_2', 'bag_2_inst_1'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0)) df <- as_mild_df(x)
mild_df
object by bags and instancesFrom a mild_df
object, return a sample that evenly pulls from the unique
bags and unique instances from each bag as much as possible. This is a form
of stratified sampling to avoid randomly sampling many rows from a few bags.
bag_instance_sampling(data, size)
bag_instance_sampling(data, size)
data |
A |
size |
A non-negative integer giving the number of rows to choose from
|
A numeric vector of length size
indicating which rows were sampled.
Sean Kent
mil_data <- generate_mild_df(positive_dist = "mvnormal", nbag = 2, ninst = 2, nsample = 2) rows <- bag_instance_sampling(mil_data, 6) table(mil_data$bag_name[rows]) table(mil_data$instance_name[rows]) rows <- bag_instance_sampling(mil_data, 4) table(mil_data$bag_name[rows]) table(mil_data$instance_name[rows])
mil_data <- generate_mild_df(positive_dist = "mvnormal", nbag = 2, ninst = 2, nsample = 2) rows <- bag_instance_sampling(mil_data, 6) table(mil_data$bag_name[rows]) table(mil_data$instance_name[rows]) rows <- bag_instance_sampling(mil_data, 4) table(mil_data$bag_name[rows]) table(mil_data$instance_name[rows])
Feature maps provide a set of covariates in a transformed space. The
build_fm()
function creates these covariates based on an object that
specifies the feature map and a provided dataset.
build_fm(kfm_fit, new_data, ...) ## S3 method for class 'kfm_exact' build_fm(kfm_fit, new_data, ...) ## S3 method for class 'kfm_nystrom' build_fm(kfm_fit, new_data, ...)
build_fm(kfm_fit, new_data, ...) ## S3 method for class 'kfm_exact' build_fm(kfm_fit, new_data, ...) ## S3 method for class 'kfm_nystrom' build_fm(kfm_fit, new_data, ...)
kfm_fit |
An object from a function in the |
new_data |
The data to generate features from. |
... |
Additional arguments for methods. |
A matrix of covariates in the feature space, with the same number of
rows as new_data
. If new_data
is a mild_df
object, build_fm()
will also return the columns containing 'bag_label', 'bag_name',
'instance_name'.
kfm_exact
: Method for kfm_exact
class.
kfm_nystrom
: Method for kfm_nystrom
class.
Sean Kent
kfm_nystrom()
fit a Nystrom kernel feature map approximation.
kfm_exact()
create an exact kernel feature map.
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_nystrom(df, m = 7, r = 6, kernel = "radial", sigma = 0.05) fm <- build_fm(fit, df) fit <- kfm_exact(kernel = "polynomial", degree = 2, const = 1) fm <- build_fm(fit, df)
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_nystrom(df, m = 7, r = 6, kernel = "radial", sigma = 0.05) fm <- build_fm(fit, df) fit <- kfm_exact(kernel = "polynomial", degree = 2, const = 1) fm <- build_fm(fit, df)
mild_df
data to the instance levelFlatten mild_df
type of data to regular multiple instance data where
each instance is a vector by extracting distribution sample quantiles, mean
and sd.
build_instance_feature( data, qtls = seq(0.05, 0.95, length.out = 10), mean = TRUE, sd = TRUE )
build_instance_feature( data, qtls = seq(0.05, 0.95, length.out = 10), mean = TRUE, sd = TRUE )
data |
A |
qtls |
Quantiles to be extracted from each instance empirical distribution. |
mean |
A logical for whether or not to extract mean. |
sd |
A logical for whether or not to extract standard deviation. |
A summarized data.frame at the instance level.
Yifei Liu
summarize_samples()
for a more general way to make a similar data
frame.
mild_df1 <- generate_mild_df(positive_degree = 3, nbag = 3) df1 <- build_instance_feature(mild_df1, seq(0.05, 0.95, length.out = 10))
mild_df1 <- generate_mild_df(positive_degree = 3, nbag = 3) df1 <- build_instance_feature(mild_df1, seq(0.05, 0.95, length.out = 10))
Formally, this function applies max()
on y
for each level of bags
.
classify_bags(y, bags, condense = TRUE)
classify_bags(y, bags, condense = TRUE)
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
condense |
A logical (default |
a named vector of length length(unique(b))
which gives the
classification for each bag. Names come from bags
.
Sean Kent
y <- c(1, 0, 0, 1, 1, 1, 0, 0, 0) bags <- rep(1:3, each = 3) classify_bags(y, bags) classify_bags(y, bags, condense = FALSE) # works with regular vector too scores <- 1:9 classify_bags(scores, bags)
y <- c(1, 0, 0, 1, 1, 1, 0, 0, 0) bags <- rep(1:3, each = 3) classify_bags(y, bags) classify_bags(y, bags, condense = FALSE) # works with regular vector too scores <- 1:9 classify_bags(scores, bags)
Cross-validation wrapper on the misvm()
function to fit the MI-SVM model
over a variety of specified cost parameters. The optimal cost parameter
is chosen by the best AUC of the cross-fit models. See ?misvm
for
more details on the fitting function.
## Default S3 method: cv_misvm( x, y, bags, cost_seq, n_fold, fold_id, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = 1, nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' cv_misvm(formula, data, cost_seq, n_fold, fold_id, ...) ## S3 method for class 'mi_df' cv_misvm(x, ...)
## Default S3 method: cv_misvm( x, y, bags, cost_seq, n_fold, fold_id, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = 1, nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' cv_misvm(formula, data, cost_seq, n_fold, fold_id, ...) ## S3 method for class 'mi_df' cv_misvm(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each row represents a sample. |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
cost_seq |
A sequence of |
n_fold |
The number of folds (default 5). If this is specified,
|
fold_id |
The ids for the specific the fold for each instance. Care must
be taken to ensure that ids respect the bag structure to avoid information
leakage. If |
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
a formula with specification |
data |
If |
An object of class cv_misvm
. The object contains the following
components:
misvm_fit
: A fit object of class misvm
trained on the full data with
the cross-validated choice of cost parameter. See misvm()
for details.
cost_seq
: the input sequence of cost arguments
cost_aucs
: estimated AUC for the models trained for each cost_seq
parameter. These are the average of the fold models for that cost, excluding
any folds that don't have both levels of y
in the validation set.
best_cost
: The optimal choice of cost parameter, chosen as that which has
the maximum AUC. If there are ties, this will pick the smallest cost with
maximum AUC.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates.
Sean Kent, Yifei Liu
misvm()
for fitting without cross-validation.
set.seed(8) mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, dist = rep("mvnormal", 3), mean = list(rep(1, 10), rep(2, 10)), sd_of_mean = rep(0.1, 3)) df <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) cost_seq <- 2^seq(-5, 7, length.out = 3) # Heuristic method mdl1 <- cv_misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, cost_seq = cost_seq, n_fold = 3, method = "heuristic") mdl2 <- cv_misvm(mi(bag_label, bag_name) ~ X1_mean + X2_mean + X3_mean, data = df, cost_seq = cost_seq, n_fold = 3) if (require(gurobi)) { # solve using the MIP method mdl3 <- cv_misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, cost_seq = cost_seq, n_fold = 3, method = "mip") } predict(mdl1, new_data = df, type = "raw", layer = "bag") # summarize predictions at the bag layer suppressWarnings(library(dplyr)) df %>% bind_cols(predict(mdl2, df, type = "class")) %>% bind_cols(predict(mdl2, df, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
set.seed(8) mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, dist = rep("mvnormal", 3), mean = list(rep(1, 10), rep(2, 10)), sd_of_mean = rep(0.1, 3)) df <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) cost_seq <- 2^seq(-5, 7, length.out = 3) # Heuristic method mdl1 <- cv_misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, cost_seq = cost_seq, n_fold = 3, method = "heuristic") mdl2 <- cv_misvm(mi(bag_label, bag_name) ~ X1_mean + X2_mean + X3_mean, data = df, cost_seq = cost_seq, n_fold = 3) if (require(gurobi)) { # solve using the MIP method mdl3 <- cv_misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, cost_seq = cost_seq, n_fold = 3, method = "mip") } predict(mdl1, new_data = df, type = "raw", layer = "bag") # summarize predictions at the bag layer suppressWarnings(library(dplyr)) df %>% bind_cols(predict(mdl2, df, type = "class")) %>% bind_cols(predict(mdl2, df, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
Specialized print methods for the mi_df
, mild_df
classes. These return
helpful information such as the number of rows, columns, bags, and instances
(for mild_df
objects).
These methods print the data frame based on the underlying subclass. This
allows for additional arguments that can be passed to print.tbl()
when the
subclass is a tibble (tbl_df
, tbl
), documented below.
## S3 method for class 'mi_df' print(x, ...) ## S3 method for class 'mild_df' print(x, ...)
## S3 method for class 'mi_df' print(x, ...) ## S3 method for class 'mild_df' print(x, ...)
x |
Object to format or print. |
... |
Passed to other methods. See |
The following extra arguments are available when x
has subclass tbl
:
n
: Number of rows to show. If NULL
, the default, will print all rows
if less than the print_max
option. Otherwise,
will print as many rows as specified by the print_min
option.
width
: Width of text output to generate. This defaults to NULL
, which
means use the width
option.
max_extra_cols
: Number of extra columns to print abbreviated
information for, if the width is too small for the entire tibble. If
NULL
, the max_extra_cols
option is used. The
previously defined n_extra
argument is soft-deprecated.
max_footer_lines
: Maximum number of footer lines. If NULL
, the
max_footer_lines
option is used.
The object passed in x
, invisibly. Primarily called to print the
object to the console.
data("ordmvnorm") print(as_mi_df(ordmvnorm, instance_label = "inst_label")) print(as_mi_df(ordmvnorm, instance_label = "inst_label"), n = 2)
data("ordmvnorm") print(as_mi_df(ordmvnorm, instance_label = "inst_label")) print(as_mi_df(ordmvnorm, instance_label = "inst_label"), n = 2)
This function samples multiple instance distributional data (a mild_df
object) where each row corresponds to a sample from a given instance
distribution. Instance distributions can be multivariate t and normal, with
mean and variance parameters that can be fixed or sampled based on prior
parameters. These instances are grouped into bags and the bag labels
follow the standard MI assumption.
generate_mild_df( nbag = 50, ninst = 4, nsample = 50, ncov = 10, nimp_pos = 1:ncov, nimp_neg = 1:ncov, positive_prob = 0.2, dist = c("mvt", "mvnormal", "mvnormal"), mean = list(rep(0, length(nimp_pos)), rep(0, length(nimp_neg)), 0), sd_of_mean = c(0.5, 0.5, 0.5), cov = list(diag(1, nrow = length(nimp_pos)), diag(1, nrow = length(nimp_neg)), 1), sample_cov = FALSE, df_wishart_cov = c(length(nimp_pos), length(nimp_neg), ncov - length(nimp_pos)), degree = c(3, NA, NA), positive_bag_prob = NULL, n_noise_inst = NULL, ... )
generate_mild_df( nbag = 50, ninst = 4, nsample = 50, ncov = 10, nimp_pos = 1:ncov, nimp_neg = 1:ncov, positive_prob = 0.2, dist = c("mvt", "mvnormal", "mvnormal"), mean = list(rep(0, length(nimp_pos)), rep(0, length(nimp_neg)), 0), sd_of_mean = c(0.5, 0.5, 0.5), cov = list(diag(1, nrow = length(nimp_pos)), diag(1, nrow = length(nimp_neg)), 1), sample_cov = FALSE, df_wishart_cov = c(length(nimp_pos), length(nimp_neg), ncov - length(nimp_pos)), degree = c(3, NA, NA), positive_bag_prob = NULL, n_noise_inst = NULL, ... )
nbag |
The number of bags (default 50). |
ninst |
The number of instances for each bag (default 4). |
nsample |
The number of samples for each instance (default 50). |
ncov |
The number of total covariates (default 10). |
nimp_pos |
An index of important covariates for positve instances
(default |
nimp_neg |
An index of important covariates for negative instances
(default |
positive_prob |
A numeric value between 0 and 1 indicating the probability of an instance being positive (default 0.2). |
dist |
A vector (length 3) of distributions for the positive, negative, and
remaining instances, respectively. Distributions can be one of
|
mean |
A list (length 3) of mean vectors for the positive, negative, and
remaining distributions. |
sd_of_mean |
A vector (length 3) of standard deviations in sampling the
mean for positive, negative, and remaining distributions, where the prior
is given by |
cov |
A list (length 3) of covariance matrices for the positive,
negative, and remaining distributions. |
sample_cov |
A logical value for whether to sample the covariance for
each distribution. If |
df_wishart_cov |
A vector (length 3) of degrees-of-freedom to use in the Wishart covariance matrix sampling. |
degree |
A vector (length 3) of degrees-of-freedom used when any of
|
positive_bag_prob |
A numeric value between 0 and 1 indicating the
probability of a bag being positive. Must be specified jointly with
|
n_noise_inst |
An integer indicating the number of negative instances in
a positive bag. Must be specified jointly with |
... |
Arguments passed to or from other methods. |
The first consideration to use this function is to determine the number of
bags, instances per bag, and samples per instance using the nbag
, ninst
,
and nsample
arguments. Next, one must consider the number of covariates
ncov
, and how those covariates will differ between instances with positive
and negative labels. Some covariates can be common between the positive and
negative instances, which we call the remainder distribution. Use nimp_pos
and nimp_neg
to specify the index of the important (non-remainder)
covariates in the distributions with positive and negative instance labels.
The structure of how many instances/bags are positive and negative is
determined by positive_prob
or the joint specification of
positive_bag_prob
and n_noise_inst
. In the first case, instances labels
have independent Bernoulli draws based on positive_prob
and bag labels are
determined by the standard MI assumption (i.e. positive if any instance in
the bag is positive). In the second case, bag labels are drawn independently
as Bernoilli with positive_bag_prob
chance of success. Each positive bag
will be given n_noise_inst
values with instance label of 0, and the
remaining with instance label of 1.
The remaining arguments are used to determine the distributions used for the
positive, negative, and remaining features. Each argument will be a vector
of list of length 3 corresponding to these 3 different groups. To create
different distributions, the strategy is to first draw the mean parameter
from Normal(mean
, sd_of_mean
* I) and the covariance parameter from
Wishart(df_wishart_cov
, cov
), with expectation equal to cov
. Then we
can sample i.i.d. draws from the specified distribution (either multivariate
normal or student's t). To ensure that each instance distribution has the
same mean, set sd_of_mean
to 0. To ensure that each instance distribution
has the same covariance, set sample_cov = FALSE
.
The final data.frame will have nsample
* nbag
* ninst
rows and ncov + 3
columns including the bag_label, bag_name, instance_name, and ncov
sampled covariates.
A mild_df
object.
Yifei Liu, Sean Kent
set.seed(8) mild_data <- generate_mild_df(nbag = 7, ninst = 3, nsample = 20, ncov = 2, nimp_pos = 1, dist = rep("mvnormal", 3), mean = list( rep(5, 1), rep(15, 2), 0 )) library(dplyr) distinct(mild_data, bag_label, bag_name, instance_name) split(mild_data[, 4:5], mild_data$instance_name) %>% sapply(colMeans) %>% round(2) %>% t()
set.seed(8) mild_data <- generate_mild_df(nbag = 7, ninst = 3, nsample = 20, ncov = 2, nimp_pos = 1, dist = rep("mvnormal", 3), mean = list( rep(5, 1), rep(15, 2), 0 )) library(dplyr) distinct(mild_data, bag_label, bag_name, instance_name) split(mild_data[, 4:5], mild_data$instance_name) %>% sapply(colMeans) %>% round(2) %>% t()
For some kernels, it is possible to create the exact features from given data. This function stores the information needed to build those exact features.
kfm_exact(kernel = "polynomial", degree = 2, const = 1)
kfm_exact(kernel = "polynomial", degree = 2, const = 1)
kernel |
A character determining the kernel to use. Currently, only
|
degree |
A numeric value (default 2) that provides the degree for
|
const |
A numeric value (default 1) for the constant term when |
Currently, the following kernels are supported:
'polynomial'
, with degree
= d and const
= c
An object of class kfm_exact
with the following components,
returned from the inputs:
kernel
degree
const
Sean Kent
Other kernel feature map functions:
kfm_nystrom()
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_exact(kernel = "polynomial", degree = 2, const = 1) fm <- build_fm(fit, df)
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_exact(kernel = "polynomial", degree = 2, const = 1) fm <- build_fm(fit, df)
Use the Nyström method to fit a feature map that approximates a given kernel.
kfm_nystrom(df, m, r, kernel, sampling, ...) ## Default S3 method: kfm_nystrom( df, m = nrow(df), r = m, kernel = "radial", sampling = "random", ... ) ## S3 method for class 'mild_df' kfm_nystrom( df, m = nrow(df), r = m, kernel = "radial", sampling = "random", ... )
kfm_nystrom(df, m, r, kernel, sampling, ...) ## Default S3 method: kfm_nystrom( df, m = nrow(df), r = m, kernel = "radial", sampling = "random", ... ) ## S3 method for class 'mild_df' kfm_nystrom( df, m = nrow(df), r = m, kernel = "radial", sampling = "random", ... )
df |
An object containing covariates for training. Usually a data.frame or matrix. |
m |
The number of examples from |
r |
The rank of matrix approximation to use. Must be less than or equal
to |
kernel |
A character determining the kernel to use. Currently, only
|
sampling |
A character determining how to sample instances. Default is
|
... |
additional parameters needed for the kernels. See details. |
For the ...
argument, the additional parameters depend on which kernel is
used:
For kernel = 'radial'
, specify sigma
to define kernel bandwidth.
an object of class kfm_nystrom
with the following components:
df_sub
the sub-sampled version of df
dv
pre-multiplication matrix which contains information on the
eigenvalues and eigenvectors of df_sub
method
'nystrom'
kernel
the input parameter kernel
kernel_params
parameters passed to ...
default
: For use on objects of class data.frame
or matrix
.
mild_df
: Ignore the information columns 'bag_label'
,
'bag_name'
, and 'instance_name'
when calculating kernel approximation.
Sean Kent
Williams, C., & Seeger, M. (2001). Using the Nyström Method to Speed Up Kernel Machines. Advances in Neural Information Processing Systems, 13, 682–688.
Kent, S., & Yu, M. (2022). Non-convex SVM for cancer diagnosis based on morphologic features of tumor microenvironment arXiv preprint arXiv:2206.14704
Other kernel feature map functions:
kfm_exact()
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_nystrom(df, m = 7, r = 6, kernel = "radial", sigma = 0.05) fm <- build_fm(fit, df)
df <- data.frame( X1 = c(2, 3, 4, 5, 6, 7, 8), X2 = c(1, 1.2, 1.3, 1.4, 1.1, 7, 1), X3 = rnorm(7) ) fit <- kfm_nystrom(df, m = 7, r = 6, kernel = "radial", sigma = 0.05) fm <- build_fm(fit, df)
Function to calculate the kernel mean embedding for to distributional data sets. It uses the empirical approximation for the integral
for a given
kernel . Currently only supports radial basis function
kernel for fast computation.
## Default S3 method: kme(df, df2 = NULL, sigma = 0.05, ...) ## S3 method for class 'mild_df' kme(df, df2 = NULL, sigma = 0.05, ...)
## Default S3 method: kme(df, df2 = NULL, sigma = 0.05, ...) ## S3 method for class 'mild_df' kme(df, df2 = NULL, sigma = 0.05, ...)
df |
A data.frame of |
df2 |
A data.frame, |
sigma |
The parameter for |
... |
Additional arguments passed to methods. |
If df2 = NULL
, calculate the kernel mean embedding matrix of (df
, df
)
otherwise calculate (df
, df2
)
A matrix of kernel mean embedding at the instance level.
default
: Default S3 method
mild_df
: S3 method for class mild_df
Yifei Liu, Sean Kent
x = data.frame('instance_name' = c('inst_1', 'inst_2', 'inst_1'), 'X1' = c(-0.4, 0.5, 2)) kme(x) mild_df1 <- generate_mild_df(nbag = 10, positive_degree = 3) kme(mild_df1)
x = data.frame('instance_name' = c('inst_1', 'inst_2', 'inst_1'), 'X1' = c(-0.4, 0.5, 2)) kme(x) mild_df1 <- generate_mild_df(nbag = 10, positive_degree = 3) kme(mild_df1)
mi
objectCreate an mi
object, usually used as a response variable in a model
formula.
mi(bag_label, bag_name)
mi(bag_label, bag_name)
bag_label |
The bag label or response, recorded as 0 = negative, 1 = positive. |
bag_name |
A unique bag identifier for each instance. |
An object of class mi
. Currently, no methods are implemented for
this.
Sean Kent
Other multiple instance formula helper functions:
mild()
mil_data <- generate_mild_df(positive_degree = 3, nbag = 10) with(mil_data, head(mi(bag_label, bag_name))) df <- get_all_vars(mi(bag_label, bag_name) ~ X1 + X2, data = mil_data) head(df)
mil_data <- generate_mild_df(positive_degree = 3, nbag = 10) with(mil_data, head(mi(bag_label, bag_name))) df <- get_all_vars(mi(bag_label, bag_name) ~ X1 + X2, data = mil_data) head(df)
mi_df()
constructs a data frame that corresponds to Multiple Instance (MI)
data. A mi_df
object must have two special columns:
bag_label
, determines the label of each bag, typically from c(0, 1)
bag_name
, character or factor that specifies the bag that each sample
belongs to.
mi_df( bag_label = character(), bag_name = character(), ..., instance_label = NULL )
mi_df( bag_label = character(), bag_name = character(), ..., instance_label = NULL )
bag_label |
A |
bag_name |
A |
... |
A set of name-value pairs. These construct the covariates for a
|
instance_label |
A |
We refer to the rows of a mi_df
as instances. Each instance is
contained in a bag, with a corresponding label. Bags will typically have
several instances within them. Instance labels can be provided, but they will
be pulled in as an attribute.
A 'mi_df' object. This data.frame-like has columns bag_label
,
bag_name
, and those specified in ...
. It also inherits from the
'tbl_df'
and 'tbl'
classes.
Sean Kent
as_mi_df()
to convert data.frames to mi_df
s.
mi_df('bag_label' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0))
mi_df('bag_label' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0))
Create a mild
object, usually used as a response variable in a model
formula.
mild(bag_label, bag_name, instance_name)
mild(bag_label, bag_name, instance_name)
bag_label |
The bag label or response, recorded as 0 = negative, 1 = positive. |
bag_name |
A unique bag identifier for each instance. |
instance_name |
A unique instance identifier for each sample. |
An object of class mild
. Currently, no methods are implemented for
this.
Sean Kent
Other multiple instance formula helper functions:
mi()
mil_data <- generate_mild_df(positive_degree = 3, nbag = 10) with(mil_data, head(mild(bag_label, bag_name, instance_name))) df <- get_all_vars(mild(bag_label, bag_name) ~ X1 + X2, data = mil_data) head(df)
mil_data <- generate_mild_df(positive_degree = 3, nbag = 10) with(mil_data, head(mild(bag_label, bag_name, instance_name))) df <- get_all_vars(mild(bag_label, bag_name) ~ X1 + X2, data = mil_data) head(df)
mild_df()
constructs a data frame that corresponds to Multiple Instance
Learning with Distributional Instances (MILD) data. A mild_df
object must
have three special columns:
bag_label
, determines the label of each bag, typically from c(0, 1)
bag_name
, character or factor that specifies the bag that each sample
belongs to.
instance_name
, character or factor that specifies the instance that
each sample belongs to.
mild_df( bag_label = character(), bag_name = character(), instance_name = character(), ..., instance_label = NULL )
mild_df( bag_label = character(), bag_name = character(), instance_name = character(), ..., instance_label = NULL )
bag_label |
A |
bag_name |
A |
instance_name |
A |
... |
A set of name-value pairs. These construct the covariates for a
|
instance_label |
A |
We refer to the rows of a mild_df
as samples, since they are
thought of as draws from the distribution that determines each instance.
Each instance is contained in a bag, with a corresponding label. Instance
labels can be provided, but they will be pulled in as an attribute.
A 'mild_df' object. This data.frame-like has columns bag_label
,
bag_name
, instance_name
, and those specified in ...
. It also inherits
from the 'tbl_df'
and 'tbl'
classes.
Yifei Liu, Sean Kent
as_mild_df()
to convert data.frames to mild_df
s.
generate_mild_df()
for simulating a mild_df
object.
summarize_samples()
for summarizing the mild_df
into a multiple
instance learning data set.
mild_df('bag_label' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'instance_name' = c('bag_1_inst_1', 'bag_1_inst_2', 'bag_2_inst_1'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0))
mild_df('bag_label' = factor(c(1, 1, 0)), 'bag_name' = c(rep('bag_1', 2), 'bag_2'), 'instance_name' = c('bag_1_inst_1', 'bag_1_inst_2', 'bag_2_inst_1'), 'X1' = c(-0.4, 0.5, 2), 'instance_label' = c(0, 1, 0))
This function fits the MIOR model, proposed by Xiao Y, Liu B, and Hao Z (2018) in "Multiple-instance Ordinal Regression". MIOR is a modified SVM framework with parallel, ordered hyperplanes where the error terms are based only on the instance closest to a midpoint between hyperplanes.
## Default S3 method: mior( x, y, bags, cost = 1, cost_eta = 1, method = "qp-heuristic", weights = NULL, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, scale = TRUE, verbose = FALSE, time_limit = 60, option = c("corrected", "xiao")), ... ) ## S3 method for class 'formula' mior(formula, data, ...) ## S3 method for class 'mi_df' mior(x, ...)
## Default S3 method: mior( x, y, bags, cost = 1, cost_eta = 1, method = "qp-heuristic", weights = NULL, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, scale = TRUE, verbose = FALSE, time_limit = 60, option = c("corrected", "xiao")), ... ) ## S3 method for class 'formula' mior(formula, data, ...) ## S3 method for class 'mi_df' mior(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents an instance. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM. If |
cost_eta |
The additional cost parameter in MIOR which controls how far away the first and last separating hyperplanes are relative to other costs. |
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
a formula with specification |
data |
If |
Predictions (see predict.mior()
) are determined by considering the smallest
distance from each point to the midpoint hyperplanes across all instances in
the bag. The prediction corresponds to the hyperplane having such a minimal
distance.
It appears as though an error in Equation (12) persists to the dual form in
(21). A corrected version of this dual formulation can be used with
control$option = 'corrected'
, or the formulation as written can be used
with control$option = 'xiao'
.
An object of class mior
The object contains at least the following
components:
gurobi_fit
: A fit from model optimization that includes relevant
components.
call_type
: A character indicating which method misvm()
was called
with.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
weights
: The calculated weights on the cost
parameter.
repr_inst
: The instances from positive bags that are selected to be
most representative of the positive instances.
n_step
: If method %in% c('heuristic', 'qp-heuristic')
, the total
steps used in the heuristic algorithm.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates.
Sean Kent
Xiao, Y., Liu, B., & Hao, Z. (2017). Multiple-instance ordinal regression. IEEE Transactions on Neural Networks and Learning Systems, 29(9), 4398-4413. doi:10.1109/TNNLS.2017.2766164
predict.misvm()
for prediction on new data.
if (require(gurobi)) { set.seed(8) # make some data n <- 15 X <- rbind( mvtnorm::rmvnorm(n/3, mean = c(4, -2, 0)), mvtnorm::rmvnorm(n/3, mean = c(0, 0, 0)), mvtnorm::rmvnorm(n/3, mean = c(-2, 1, 0)) ) score <- X %*% c(2, -1, 0) y <- as.numeric(cut(score, c(-Inf, quantile(score, probs = 1:2 / 3), Inf))) bags <- 1:length(y) # add in points outside boundaries X <- rbind( X, mvtnorm::rmvnorm(n, mean = c(6, -3, 0)), mvtnorm::rmvnorm(n, mean = c(-6, 3, 0)) ) y <- c(y, rep(-1, 2*n)) bags <- rep(bags, 3) repr <- c(rep(1, n), rep(0, 2*n)) y_bag <- classify_bags(y, bags, condense = FALSE) mdl1 <- mior(X, y_bag, bags) predict(mdl1, X, new_bags = bags) }
if (require(gurobi)) { set.seed(8) # make some data n <- 15 X <- rbind( mvtnorm::rmvnorm(n/3, mean = c(4, -2, 0)), mvtnorm::rmvnorm(n/3, mean = c(0, 0, 0)), mvtnorm::rmvnorm(n/3, mean = c(-2, 1, 0)) ) score <- X %*% c(2, -1, 0) y <- as.numeric(cut(score, c(-Inf, quantile(score, probs = 1:2 / 3), Inf))) bags <- 1:length(y) # add in points outside boundaries X <- rbind( X, mvtnorm::rmvnorm(n, mean = c(6, -3, 0)), mvtnorm::rmvnorm(n, mean = c(-6, 3, 0)) ) y <- c(y, rep(-1, 2*n)) bags <- rep(bags, 3) repr <- c(rep(1, n), rep(0, 2*n)) y_bag <- classify_bags(y, bags, condense = FALSE) mdl1 <- mior(X, y_bag, bags) predict(mdl1, X, new_bags = bags) }
This function fits the MILD-SVM model, which takes a multiple-instance learning with distributions (MILD) data set and fits a modified SVM to it. The MILD-SVM methodology is based on research in progress.
## Default S3 method: mismm( x, y, bags, instances, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "radial", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' mismm(formula, data, ...) ## S3 method for class 'mild_df' mismm(x, ...)
## Default S3 method: mismm( x, y, bags, instances, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "radial", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' mismm(formula, data, ...) ## S3 method for class 'mild_df' mismm(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents a sample. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
instances |
A vector specifying which samples belong to each instance. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM. If |
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
A formula with specification |
data |
If |
Several choices of fitting algorithm are available, including a version of the heuristic algorithm proposed by Andrews et al. (2003) and a novel algorithm that explicitly solves the mixed-integer programming (MIP) problem using the gurobi package optimization back-end.
An object of class mismm
The object contains at least the
following components:
*_fit
: A fit object depending on the method
parameter. If method = 'heuristic'
, this will be a ksvm
fit from the kernlab package. If
method = 'mip'
this will be gurobi_fit
from a model optimization.
call_type
: A character indicating which method misvm()
was called
with.
x
: The training data needed for computing the kernel matrix in
prediction.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
weights
: The calculated weights on the cost
parameter.
sigma
: The radial basis function kernel parameter.
repr_inst
: The instances from positive bags that are selected to be
most representative of the positive instances.
n_step
: If method %in% c('heuristic', 'qp-heuristic')
, the total
steps used in the heuristic algorithm.
useful_inst_idx
: The instances that were selected to represent the bags
in the heuristic fitting.
inst_order
: A character vector that is used to modify the ordering of
input data.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mild_df
: Method for mild_df
objects
Sean Kent, Yifei Liu
Kent, S., & Yu, M. (2022). Non-convex SVM for cancer diagnosis based on morphologic features of tumor microenvironment arXiv preprint arXiv:2206.14704
predict.mismm()
for prediction on new data.
set.seed(8) mil_data <- generate_mild_df(nbag = 15, nsample = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) # Heuristic method mdl1 <- mismm(mil_data) mdl2 <- mismm(mild(bag_label, bag_name, instance_name) ~ X1 + X2 + X3, data = mil_data) # MIP method if (require(gurobi)) { mdl3 <- mismm(mil_data, method = "mip", control = list(nystrom_args = list(m = 10, r = 10))) predict(mdl3, mil_data) } predict(mdl1, new_data = mil_data, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) mil_data %>% bind_cols(predict(mdl2, mil_data, type = "class")) %>% bind_cols(predict(mdl2, mil_data, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
set.seed(8) mil_data <- generate_mild_df(nbag = 15, nsample = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) # Heuristic method mdl1 <- mismm(mil_data) mdl2 <- mismm(mild(bag_label, bag_name, instance_name) ~ X1 + X2 + X3, data = mil_data) # MIP method if (require(gurobi)) { mdl3 <- mismm(mil_data, method = "mip", control = list(nystrom_args = list(m = 10, r = 10))) predict(mdl3, mil_data) } predict(mdl1, new_data = mil_data, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) mil_data %>% bind_cols(predict(mdl2, mil_data, type = "class")) %>% bind_cols(predict(mdl2, mil_data, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
This function fits the MI-SVM model, first proposed by Andrews et al. (2003). It is a variation on the traditional SVM framework that carefully treats data from the multiple instance learning paradigm, where instances are grouped into bags, and a label is only available for each bag.
## Default S3 method: misvm( x, y, bags, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' misvm(formula, data, ...) ## S3 method for class 'mi_df' misvm(x, ...) ## S3 method for class 'mild_df' misvm(x, .fns = list(mean = mean, sd = stats::sd), cor = FALSE, ...)
## Default S3 method: misvm( x, y, bags, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' misvm(formula, data, ...) ## S3 method for class 'mi_df' misvm(x, ...) ## S3 method for class 'mild_df' misvm(x, .fns = list(mean = mean, sd = stats::sd), cor = FALSE, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents an instance. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM. If |
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
a formula with specification |
data |
If |
.fns |
(argument for |
cor |
(argument for |
Several choices of fitting algorithm are available, including a version of the heuristic algorithm proposed by Andrews et al. (2003) and a novel algorithm that explicitly solves the mixed-integer programming (MIP) problem using the gurobi package optimization back-end.
An object of class misvm.
The object contains at least the
following components:
*_fit
: A fit object depending on the method
parameter. If method = 'heuristic'
, this will be an svm
fit from the e1071 package. If
method = 'mip', 'qp-heuristic'
this will be gurobi_fit
from a model
optimization.
call_type
: A character indicating which method misvm()
was called
with.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
weights
: The calculated weights on the cost
parameter.
repr_inst
: The instances from positive bags that are selected to be
most representative of the positive instances.
n_step
: If method %in% c('heuristic', 'qp-heuristic')
, the total
steps used in the heuristic algorithm.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates.
mild_df
: Method for mild_df
objects. Summarize samples to the
instance level based on specified functions, then perform misvm()
on
instance level data.
Sean Kent, Yifei Liu
Andrews, S., Tsochantaridis, I., & Hofmann, T. (2002). Support vector machines for multiple-instance learning. Advances in neural information processing systems, 15.
Kent, S., & Yu, M. (2022). Non-convex SVM for cancer diagnosis based on morphologic features of tumor microenvironment arXiv preprint arXiv:2206.14704
predict.misvm()
for prediction on new data.
cv_misvm()
for cross-validation fitting.
set.seed(8) mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) df <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) # Heuristic method mdl1 <- misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, method = "heuristic") mdl2 <- misvm(mi(bag_label, bag_name) ~ X1_mean + X2_mean + X3_mean, data = df) # MIP method if (require(gurobi)) { mdl3 <- misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, method = "mip") } predict(mdl1, new_data = df, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) df %>% bind_cols(predict(mdl2, df, type = "class")) %>% bind_cols(predict(mdl2, df, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
set.seed(8) mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) df <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) # Heuristic method mdl1 <- misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, method = "heuristic") mdl2 <- misvm(mi(bag_label, bag_name) ~ X1_mean + X2_mean + X3_mean, data = df) # MIP method if (require(gurobi)) { mdl3 <- misvm(x = df[, 4:123], y = df$bag_label, bags = df$bag_name, method = "mip") } predict(mdl1, new_data = df, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) df %>% bind_cols(predict(mdl2, df, type = "class")) %>% bind_cols(predict(mdl2, df, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
This function uses the one-vs-all multiclass classification strategy to fit a series of MI-SVM models for predictions on ordinal outcome data. For an ordinal outcome with K levels, we fit K MI-SVM models to predict an individual level vs not.
## Default S3 method: misvm_orova( x, y, bags, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' misvm_orova(formula, data, ...) ## S3 method for class 'mi_df' misvm_orova(x, ...)
## Default S3 method: misvm_orova( x, y, bags, cost = 1, method = c("heuristic", "mip", "qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), nystrom_args = list(m = nrow(x), r = nrow(x), sampling = "random"), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60, start = FALSE), ... ) ## S3 method for class 'formula' misvm_orova(formula, data, ...) ## S3 method for class 'mi_df' misvm_orova(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents an instance. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM. If |
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
a formula with specification |
data |
If |
An object of class misvm_orova
The object contains at least the
following components:
fits
: a list of misvm
objects with length equal to the number of
classes in y
. See misvm()
for details on the misvm
object.
call_type
: A character indicating which method misvm_orova()
was
called with.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates.
Sean Kent
Andrews, S., Tsochantaridis, I., & Hofmann, T. (2002). Support vector machines for multiple-instance learning. Advances in neural information processing systems, 15.
predict.misvm_orova()
for prediction on new data.
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- misvm_orova(x, y, bags) predict(mdl1, x, new_bags = bags)
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- misvm_orova(x, y, bags) predict(mdl1, x, new_bags = bags)
This function fits a modification of MI-SVM to ordinal outcome data based on the research method proposed by Kent and Yu.
## Default S3 method: omisvm( x, y, bags, cost = 1, h = 1, s = Inf, method = c("qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60), ... ) ## S3 method for class 'formula' omisvm(formula, data, ...) ## S3 method for class 'mi_df' omisvm(x, ...)
## Default S3 method: omisvm( x, y, bags, cost = 1, h = 1, s = Inf, method = c("qp-heuristic"), weights = TRUE, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, type = "C-classification", scale = TRUE, verbose = FALSE, time_limit = 60), ... ) ## S3 method for class 'formula' omisvm(formula, data, ...) ## S3 method for class 'mi_df' omisvm(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents an instance. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
bags |
A vector specifying which instance belongs to each bag. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM. If |
h |
A scalar that controls the trade-off between maximizing the margin and minimizing distance between hyperplanes. |
s |
An integer for how many replication points to add to the dataset. If
|
method |
The algorithm to use in fitting (default |
weights |
named vector, or |
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
a formula with specification |
data |
If |
Currently, the only method available is a heuristic algorithm in linear SVM space. Additional methods should be available shortly.
An object of class omisvm.
The object contains at least the
following components:
*_fit
: A fit object depending on the method
parameter. If method = 'qp-heuristic'
this will be gurobi_fit
from a model optimization.
call_type
: A character indicating which method omisvm()
was called
with.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
weights
: The calculated weights on the cost
parameter.
repr_inst
: The instances from positive bags that are selected to be
most representative of the positive instances.
n_step
: If method == 'qp-heuristic'
, the total steps used in the
heuristic algorithm.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates.
Sean Kent
predict.omisvm()
for prediction on new data.
if (require(gurobi)) { data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- omisvm(x, y, bags, weights = NULL) predict(mdl1, x, new_bags = bags) }
if (require(gurobi)) { data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- omisvm(x, y, bags, weights = NULL) predict(mdl1, x, new_bags = bags) }
A data set that demonstrates the ordinal multiple-instance learning structure with feature columns randomly sampled from a multivariate normal distribution.
ordmvnorm
ordmvnorm
An MI data frame with 1000 rows 8 variables, and 5 bags. Instance
labels can be accessed via attr(ordmvnorm, "instance_label")
.
outcome label at the bag level. This is the maximum of the inst_label
for each bag
indicator of each bag
Variable with mean equal to 2 * inst_label
Variable with mean equal to -1 * inst_label
Variable with mean equal to 1 * inst_label
Variable with mean 0, essentially noise
Variable with mean 0, essentially noise
cv_misvm
objectPredict method for cv_misvm
object
## S3 method for class 'cv_misvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
## S3 method for class 'cv_misvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column '.pred_class'. If type = 'raw'
, the tibble will have
a column '.pred'.
Sean Kent
mil_data <- generate_mild_df( nbag = 10, nsample = 20, positive_degree = 3 ) df1 <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) mdl1 <- cv_misvm(x = df1[, 4:123], y = df1$bag_label, bags = df1$bag_name, cost_seq = 2^(-2:2), n_fold = 3, method = "heuristic") predict(mdl1, new_data = df1, type = "raw", layer = "bag") # summarize predictions at the bag layer suppressWarnings(library(dplyr)) df1 %>% bind_cols(predict(mdl1, df1, type = "class")) %>% bind_cols(predict(mdl1, df1, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
mil_data <- generate_mild_df( nbag = 10, nsample = 20, positive_degree = 3 ) df1 <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) mdl1 <- cv_misvm(x = df1[, 4:123], y = df1$bag_label, bags = df1$bag_name, cost_seq = 2^(-2:2), n_fold = 3, method = "heuristic") predict(mdl1, new_data = df1, type = "raw", layer = "bag") # summarize predictions at the bag layer suppressWarnings(library(dplyr)) df1 %>% bind_cols(predict(mdl1, df1, type = "class")) %>% bind_cols(predict(mdl1, df1, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
mior
objectPredict method for mior
object
## S3 method for class 'mior' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
## S3 method for class 'mior' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameters
new_bags
and new_instances
are not necessary, as long as the names match
the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
a column .pred
.
Sean Kent
mior()
for fitting the mior
object.
if (require(gurobi)) { set.seed(8) # make some data n <- 15 X <- rbind( mvtnorm::rmvnorm(n/3, mean = c(4, -2, 0)), mvtnorm::rmvnorm(n/3, mean = c(0, 0, 0)), mvtnorm::rmvnorm(n/3, mean = c(-2, 1, 0)) ) score <- X %*% c(2, -1, 0) y <- as.numeric(cut(score, c(-Inf, quantile(score, probs = 1:2 / 3), Inf))) bags <- 1:length(y) # add in points outside boundaries X <- rbind( X, mvtnorm::rmvnorm(n, mean = c(6, -3, 0)), mvtnorm::rmvnorm(n, mean = c(-6, 3, 0)) ) y <- c(y, rep(-1, 2*n)) bags <- rep(bags, 3) repr <- c(rep(1, n), rep(0, 2*n)) y_bag <- classify_bags(y, bags, condense = FALSE) mdl1 <- mior(X, y_bag, bags) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y_bag, bags = bags, as.data.frame(X)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% distinct(y, bags, .pred_class, .pred) }
if (require(gurobi)) { set.seed(8) # make some data n <- 15 X <- rbind( mvtnorm::rmvnorm(n/3, mean = c(4, -2, 0)), mvtnorm::rmvnorm(n/3, mean = c(0, 0, 0)), mvtnorm::rmvnorm(n/3, mean = c(-2, 1, 0)) ) score <- X %*% c(2, -1, 0) y <- as.numeric(cut(score, c(-Inf, quantile(score, probs = 1:2 / 3), Inf))) bags <- 1:length(y) # add in points outside boundaries X <- rbind( X, mvtnorm::rmvnorm(n, mean = c(6, -3, 0)), mvtnorm::rmvnorm(n, mean = c(-6, 3, 0)) ) y <- c(y, rep(-1, 2*n)) bags <- rep(bags, 3) repr <- c(rep(1, n), rep(0, 2*n)) y_bag <- classify_bags(y, bags, condense = FALSE) mdl1 <- mior(X, y_bag, bags) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y_bag, bags = bags, as.data.frame(X)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% distinct(y, bags, .pred_class, .pred) }
mismm
objectPredict method for mismm
object
## S3 method for class 'mismm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", new_instances = "instance_name", kernel = NULL, ... )
## S3 method for class 'mismm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", new_instances = "instance_name", kernel = NULL, ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
new_instances |
A character or character vector. Can specify a singular
character that provides the column name for the instance names in
|
kernel |
An optional pre-computed kernel matrix at the instance level or
|
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameters
new_bags
and new_instances
are not necessary, as long as the names match
the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
a column .pred
.
Sean Kent
mismm()
for fitting the mismm
object.
mil_data <- generate_mild_df(nbag = 15, nsample = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) mdl1 <- mismm(mil_data, control = list(sigma = 1/5)) # bag level predictions library(dplyr) mil_data %>% bind_cols(predict(mdl1, mil_data, type = "class")) %>% bind_cols(predict(mdl1, mil_data, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred) # instance level prediction mil_data %>% bind_cols(predict(mdl1, mil_data, type = "class", layer = "instance")) %>% bind_cols(predict(mdl1, mil_data, type = "raw", layer = "instance")) %>% distinct(bag_name, instance_name, bag_label, .pred_class, .pred)
mil_data <- generate_mild_df(nbag = 15, nsample = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) mdl1 <- mismm(mil_data, control = list(sigma = 1/5)) # bag level predictions library(dplyr) mil_data %>% bind_cols(predict(mdl1, mil_data, type = "class")) %>% bind_cols(predict(mdl1, mil_data, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred) # instance level prediction mil_data %>% bind_cols(predict(mdl1, mil_data, type = "class", layer = "instance")) %>% bind_cols(predict(mdl1, mil_data, type = "raw", layer = "instance")) %>% distinct(bag_name, instance_name, bag_label, .pred_class, .pred)
misvm
objectPredict method for misvm
object
## S3 method for class 'misvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
## S3 method for class 'misvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameters
new_bags
and new_instances
are not necessary, as long as the names match
the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
a column .pred
.
Sean Kent
misvm()
for fitting the misvm
object.
cv_misvm()
for fitting the misvm
object with cross-validation.
mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) df1 <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) mdl1 <- misvm(x = df1[, 4:63], y = df1$bag_label, bags = df1$bag_name, method = "heuristic") predict(mdl1, new_data = df1, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) df1 %>% bind_cols(predict(mdl1, df1, type = "class")) %>% bind_cols(predict(mdl1, df1, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
mil_data <- generate_mild_df(nbag = 20, positive_prob = 0.15, sd_of_mean = rep(0.1, 3)) df1 <- build_instance_feature(mil_data, seq(0.05, 0.95, length.out = 10)) mdl1 <- misvm(x = df1[, 4:63], y = df1$bag_label, bags = df1$bag_name, method = "heuristic") predict(mdl1, new_data = df1, type = "raw", layer = "bag") # summarize predictions at the bag layer library(dplyr) df1 %>% bind_cols(predict(mdl1, df1, type = "class")) %>% bind_cols(predict(mdl1, df1, type = "raw")) %>% distinct(bag_name, bag_label, .pred_class, .pred)
misvm_orova
objectPredict method for misvm_orova
object. Predictions use the K fitted MI-SVM
models. For class predictions, we return the class whose MI-SVM model has
the highest raw predicted score. For raw predictions, a full matrix of
predictions is returned, with one column for each model.
## S3 method for class 'misvm_orova' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
## S3 method for class 'misvm_orova' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the
parameters new_bags
and new_instances
are not necessary, as long as the
names match the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
K columns .pred_{class_name}
corresponding to the raw predictions of the
K models.
Sean Kent
misvm_orova()
for fitting the misvm_orova
object.
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- misvm_orova(x, y, bags) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y, bags = bags, as.data.frame(x)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% select(-starts_with("V")) %>% distinct()
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- misvm_orova(x, y, bags) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y, bags = bags, as.data.frame(x)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% select(-starts_with("V")) %>% distinct()
omisvm
objectPredict method for omisvm
object
## S3 method for class 'omisvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
## S3 method for class 'omisvm' predict( object, new_data, type = c("class", "raw"), layer = c("bag", "instance"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameters
new_bags
and new_instances
are not necessary, as long as the names match
the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
a column .pred
.
Sean Kent
omisvm()
for fitting the omisvm
object.
if (require(gurobi)) { data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- omisvm(x, y, bags, weights = NULL) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y, bags = bags, as.data.frame(x)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% distinct(y, bags, .pred_class, .pred) }
if (require(gurobi)) { data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- ordmvnorm$bag_label bags <- ordmvnorm$bag_name mdl1 <- omisvm(x, y, bags, weights = NULL) # summarize predictions at the bag layer library(dplyr) df1 <- bind_cols(y = y, bags = bags, as.data.frame(x)) df1 %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "class")) %>% bind_cols(predict(mdl1, df1, new_bags = bags, type = "raw")) %>% distinct(y, bags, .pred_class, .pred) }
smm
objectPredict method for smm
object
## S3 method for class 'smm' predict( object, new_data, type = c("class", "raw"), layer = "instance", new_instances = "instance_name", new_bags = "bag_name", kernel = NULL, ... )
## S3 method for class 'smm' predict( object, new_data, type = c("class", "raw"), layer = "instance", new_instances = "instance_name", new_bags = "bag_name", kernel = NULL, ... )
object |
an object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_instances |
A character or character vector. Can specify a singular
character that provides the column name for the instance names in
|
new_bags |
A character or character vector. Only relevant when fit with
|
kernel |
An optional pre-computed kernel matrix at the instance level or
|
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameters
new_bags
and new_instances
are not necessary, as long as the names match
the original function call.
tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column named .pred_class
. If type = 'raw'
, the tibble will
have a column name .pred
.
Sean Kent
smm()
for fitting the smm
object.
set.seed(8) n_instances <- 10 n_samples <- 20 y <- rep(c(1, -1), each = n_samples * n_instances / 2) instances <- as.character(rep(1:n_instances, each = n_samples)) x <- data.frame(x1 = rnorm(length(y), mean = 1*(y==1)), x2 = rnorm(length(y), mean = 2*(y==1)), x3 = rnorm(length(y), mean = 3*(y==1))) mdl <- smm(x, y, instances, control = list(sigma = 1/3)) # instance level predictions (training data) suppressWarnings(library(dplyr)) data.frame(instance_name = instances, y = y, x) %>% bind_cols(predict(mdl, type = "raw", new_data = x, new_instances = instances)) %>% bind_cols(predict(mdl, type = "class", new_data = x, new_instances = instances)) %>% distinct(instance_name, y, .pred, .pred_class) # test data new_inst <- rep(c("11", "12"), each = 30) new_y <- rep(c(1, -1), each = 30) new_x <- data.frame(x1 = rnorm(length(new_inst), mean = 1*(new_inst=="11")), x2 = rnorm(length(new_inst), mean = 2*(new_inst=="11")), x3 = rnorm(length(new_inst), mean = 3*(new_inst=="11"))) # instance level predictions (test data) data.frame(instance_name = new_inst, y = new_y, new_x) %>% bind_cols(predict(mdl, type = "raw", new_data = new_x, new_instances = new_inst)) %>% bind_cols(predict(mdl, type = "class", new_data = new_x, new_instances = new_inst)) %>% distinct(instance_name, y, .pred, .pred_class)
set.seed(8) n_instances <- 10 n_samples <- 20 y <- rep(c(1, -1), each = n_samples * n_instances / 2) instances <- as.character(rep(1:n_instances, each = n_samples)) x <- data.frame(x1 = rnorm(length(y), mean = 1*(y==1)), x2 = rnorm(length(y), mean = 2*(y==1)), x3 = rnorm(length(y), mean = 3*(y==1))) mdl <- smm(x, y, instances, control = list(sigma = 1/3)) # instance level predictions (training data) suppressWarnings(library(dplyr)) data.frame(instance_name = instances, y = y, x) %>% bind_cols(predict(mdl, type = "raw", new_data = x, new_instances = instances)) %>% bind_cols(predict(mdl, type = "class", new_data = x, new_instances = instances)) %>% distinct(instance_name, y, .pred, .pred_class) # test data new_inst <- rep(c("11", "12"), each = 30) new_y <- rep(c(1, -1), each = 30) new_x <- data.frame(x1 = rnorm(length(new_inst), mean = 1*(new_inst=="11")), x2 = rnorm(length(new_inst), mean = 2*(new_inst=="11")), x3 = rnorm(length(new_inst), mean = 3*(new_inst=="11"))) # instance level predictions (test data) data.frame(instance_name = new_inst, y = new_y, new_x) %>% bind_cols(predict(mdl, type = "raw", new_data = new_x, new_instances = new_inst)) %>% bind_cols(predict(mdl, type = "class", new_data = new_x, new_instances = new_inst)) %>% distinct(instance_name, y, .pred, .pred_class)
svor_exc
objectPredict method for svor_exc
object
## S3 method for class 'svor_exc' predict( object, new_data, type = c("class", "raw"), layer = c("instance", "bag"), new_bags = "bag_name", ... )
## S3 method for class 'svor_exc' predict( object, new_data, type = c("class", "raw"), layer = c("instance", "bag"), new_bags = "bag_name", ... )
object |
An object of class |
new_data |
A data frame to predict from. This needs to have all of the features that the data was originally fitted with. |
type |
If |
layer |
If |
new_bags |
A character or character vector. Can specify a singular
character that provides the column name for the bag names in |
... |
Arguments passed to or from other methods. |
When the object was fitted using the formula
method, then the parameter
new_bags
is not necessary, as long as the names match
the original function call.
A tibble with nrow(new_data)
rows. If type = 'class'
, the tibble
will have a column .pred_class
. If type = 'raw'
, the tibble will have
a column .pred
.
Sean Kent
svor_exc()
for fitting the svor_exc
object.
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- attr(ordmvnorm, "instance_label") mdl1 <- svor_exc(x, y) predict(mdl1, x) predict(mdl1, x, type = "raw")
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- attr(ordmvnorm, "instance_label") mdl1 <- svor_exc(x, y) predict(mdl1, x) predict(mdl1, x, type = "raw")
Function to carry out support measure machines algorithm which is appropriate
for multiple instance learning. The algorithm calculates the kernel matrix of
different empirical measures using kernel mean embedding. The data set should
be passed in with rows corresponding to samples from a set of instances. SMM
will compute a kernel on the instances and pass that to kernlab::ksvm()
to
train the appropriate SVM model.
## Default S3 method: smm( x, y, instances, cost = 1, weights = TRUE, control = list(kernel = "radial", sigma = if (is.vector(x)) 1 else 1/ncol(x), scale = TRUE), ... ) ## S3 method for class 'formula' smm(formula, data, instances = "instance_name", ...) ## S3 method for class 'mild_df' smm(x, ...)
## Default S3 method: smm( x, y, instances, cost = 1, weights = TRUE, control = list(kernel = "radial", sigma = if (is.vector(x)) 1 else 1/ncol(x), scale = TRUE), ... ) ## S3 method for class 'formula' smm(formula, data, instances = "instance_name", ...) ## S3 method for class 'mild_df' smm(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents a sample. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
instances |
A vector specifying which samples belong to each instance. Can be a string, numeric, of factor. |
cost |
The cost parameter in SVM, fed to the |
weights |
named vector, or |
control |
A list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
A formula with specification |
data |
If |
An object of class smm
The object contains at least the
following components:
ksvm_fit
: A fit of class ksvm
from the kernlab package.
call_type
: A character indicating which method smm()
was called with.
x
: The training data needed for computing the kernel matrix in
prediction.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
sigma
: The radial basis function kernel parameter.
weights
: The calculated weights on the cost
parameter, if applicable.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mild_df
: Method for mild_df
objects. Use the bag_label
as y
at
the instance level, then perform smm()
ignoring the MIL structure.
Sean Kent, Yifei Liu
Muandet, K., Fukumizu, K., Dinuzzo, F., & Schölkopf, B. (2012). Learning from distributions via support measure machines. Advances in neural information processing systems, 25.
predict.smm()
for prediction on new data.
set.seed(8) n_instances <- 10 n_samples <- 20 y <- rep(c(1, -1), each = n_samples * n_instances / 2) instances <- as.character(rep(1:n_instances, each = n_samples)) x <- data.frame(x1 = rnorm(length(y), mean = 1*(y==1)), x2 = rnorm(length(y), mean = 2*(y==1)), x3 = rnorm(length(y), mean = 3*(y==1))) df <- data.frame(instance_name = instances, y = y, x) mdl <- smm(x, y, instances) mdl2 <- smm(y ~ ., data = df) # instance level predictions suppressWarnings(library(dplyr)) df %>% dplyr::bind_cols(predict(mdl, type = "raw", new_data = x, new_instances = instances)) %>% dplyr::bind_cols(predict(mdl, type = "class", new_data = x, new_instances = instances)) %>% dplyr::distinct(instance_name, y, .pred, .pred_class)
set.seed(8) n_instances <- 10 n_samples <- 20 y <- rep(c(1, -1), each = n_samples * n_instances / 2) instances <- as.character(rep(1:n_instances, each = n_samples)) x <- data.frame(x1 = rnorm(length(y), mean = 1*(y==1)), x2 = rnorm(length(y), mean = 2*(y==1)), x3 = rnorm(length(y), mean = 3*(y==1))) df <- data.frame(instance_name = instances, y = y, x) mdl <- smm(x, y, instances) mdl2 <- smm(y ~ ., data = df) # instance level predictions suppressWarnings(library(dplyr)) df %>% dplyr::bind_cols(predict(mdl, type = "raw", new_data = x, new_instances = instances)) %>% dplyr::bind_cols(predict(mdl, type = "class", new_data = x, new_instances = instances)) %>% dplyr::distinct(instance_name, y, .pred, .pred_class)
Summarize a numeric data frame based on specified grouping columns and a list
of functions. This is useful in summarizing a mild_df
object from the
sample level to the instance level.
## Default S3 method: summarize_samples(data, group_cols, .fns = list(mean = mean), cor = FALSE, ...) ## S3 method for class 'mild_df' summarize_samples(data, ...)
## Default S3 method: summarize_samples(data, group_cols, .fns = list(mean = mean), cor = FALSE, ...) ## S3 method for class 'mild_df' summarize_samples(data, ...)
data |
A data.frame, 'mild_df' object, or similar of data to summarize. |
group_cols |
A character vector of column(s) that describe groups to summarize across. |
.fns |
A list of functions (default |
cor |
A logical (default |
... |
Arguments passed to or from other methods. |
A tibble with summarized data. There will be one row for each set
of distinct groups specified by group_cols
. There will be one column for
each of the group_cols
, plus length(.fns)
columns for each of the
features in data
, plus correlation columns if specified.
default
: Method for data.frame-like objects.
mild_df
: Method for mild_df
objects.
Sean Kent
fns <- list(mean = mean, sd = sd) summarize_samples(mtcars, group_cols = c("cyl", "gear"), .fns = fns) summarize_samples(mtcars, group_cols = c("cyl", "gear"), .fns = fns, cor = TRUE)
fns <- list(mean = mean, sd = sd) summarize_samples(mtcars, group_cols = c("cyl", "gear"), .fns = fns) summarize_samples(mtcars, group_cols = c("cyl", "gear"), .fns = fns, cor = TRUE)
This function fits the Support Vector Ordinal Regression with Explicit Constraints based on the research of Chu and Keerthi (2007).
## Default S3 method: svor_exc( x, y, cost = 1, method = c("smo"), weights = NULL, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, scale = TRUE, verbose = FALSE), ... ) ## S3 method for class 'formula' svor_exc(formula, data, ...) ## S3 method for class 'mi_df' svor_exc(x, ...)
## Default S3 method: svor_exc( x, y, cost = 1, method = c("smo"), weights = NULL, control = list(kernel = "linear", sigma = if (is.vector(x)) 1 else 1/ncol(x), max_step = 500, scale = TRUE, verbose = FALSE), ... ) ## S3 method for class 'formula' svor_exc(formula, data, ...) ## S3 method for class 'mi_df' svor_exc(x, ...)
x |
A data.frame, matrix, or similar object of covariates, where each
row represents an instance. If a |
y |
A numeric, character, or factor vector of bag labels for each
instance. Must satisfy |
cost |
The cost parameter in SVM. |
method |
The algorithm to use in fitting (default |
weights |
|
control |
list of additional parameters passed to the method that control computation with the following components:
|
... |
Arguments passed to or from other methods. |
formula |
A formula with specification |
data |
If |
An object of class svor_exc
The object contains at least the
following components:
smo_fit
: A fit object from running the modified ordinal smo algorithm.
call_type
: A character indicating which method svor_exc()
was called
with.
features
: The names of features used in training.
levels
: The levels of y
that are recorded for future prediction.
cost
: The cost parameter from function inputs.
n_step
: The total steps used in the heuristic algorithm.
x_scale
: If scale = TRUE
, the scaling parameters for new predictions.
default
: Method for data.frame-like objects
formula
: Method for passing formula
mi_df
: Method for mi_df
objects, automatically handling bag
names, labels, and all covariates. Use the bag_label
as y
at the
instance level, then perform svor_exc()
ignoring the MIL structure and
bags.
Sean Kent
Chu, W., & Keerthi, S. S. (2007). Support vector ordinal regression. Neural computation, 19(3), 792-815. doi:10.1162/neco.2007.19.3.792
predict.svor_exc()
for prediction on new data.
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- attr(ordmvnorm, "instance_label") mdl1 <- svor_exc(x, y) predict(mdl1, x)
data("ordmvnorm") x <- ordmvnorm[, 3:7] y <- attr(ordmvnorm, "instance_label") mdl1 <- svor_exc(x, y) predict(mdl1, x)