README.md

In ghonk/catlearn.suppls: Catlearn Supplemental Functions

catlearn.suppls

A constellation of functions that work with the catlearn package to facilitate the broad modeling goals of the Learning and Representation in Cognition Laboratory.

Package under active development

The package can be installed using these commands:

install.packages("devtools")
devtools::install_github("ghonk/catlearn.suppls")

catlearn.suppls is intended to be used as a supplement to the catlearn package. It facilitates modeling with the DIVergent Autoencoder architecture (slpDIVA()) and features a suite of functions under development for active research projects.

Demo

First load the packages.

library(catlearn)
library(catlearn.suppls)

Set Up and Execution

Now load some data from get_test_inputs and assign variables for the properties of the data. The package includes a set of classic category structures; we'll use a 4-D family resemblance + unidimensional rule category structure for this demo.

# get some sample data
test_inputs <- get_test_inputs('unifr')

# set inputs and class labels
ins  <- test_inputs$ins
labs <- test_inputs$labels

# get number of categories and features
nfeats <- dim(ins)[2]
ncats <- length(unique(labs))

Next, setup the model parameters. The design pattern for catlearn is called a stateful list processor. This means we provide the model with a list that contains all of the model's hyperparamters and it returns a similar list that includes the model's learned parameters (weights).

# construct a state list
st <- list(learning_rate = 0.25, num_feats = nfeats, num_hids = 3, num_cats = ncats,
  beta_val = 0, phi = 1, continuous = FALSE, in_wts = NULL, out_wts = NULL, wts_range = 1,
  wts_center = 0, colskip = 4)

We then use the catlearn.suppls package to create a training matrix, tr.

# tr_init makes the empty training matrix
tr <- tr_init(nfeats, ncats)

# tr_add fills in the data and procedure (i.e., training, test, model reset)
tr <- tr_add(inputs = ins, tr = tr, labels = labs, blocks = 12, ctrl = 0, 
  shuffle = TRUE, reset = TRUE)

Finally, we run the model with our state list st and training matrix tr.

diva_model <- slpDIVA(st, tr)

Inspecting Results

To examine performance we can match the output (response probabilities) to the training matrix tr.

# name the output columns
colnames(diva_model$out) <- paste0('o', 1:dim(diva_model$out)[2])

# Combine the inputs and outputs (if desired)
trn_result <- cbind(tr, round(diva_model$out, 4))
tail(trn_result)

##       ctrl trial blk example x1 x2 x3 x4 t1 t2     o1     o2
## [91,]    0    91  12       2  1  1 -1  1  1 -1 1.0000 0.0000
## [92,]    0    92  12       4 -1  1  1  1  1 -1 0.9992 0.0008
## [93,]    0    93  12       3  1 -1  1  1  1 -1 0.9973 0.0027
## [94,]    0    94  12       7 -1  1 -1 -1 -1  1 0.0000 1.0000
## [95,]    0    95  12       8  1 -1 -1 -1 -1  1 0.0005 0.9995
## [96,]    0    96  12       6 -1 -1  1 -1 -1  1 0.0000 1.0000

Evaluating performance

Use response_probs to extract the response probabilities for the target categories (for every training step (trial) or averaged across blocks)

response_probs(tr, diva_model$out, blocks = TRUE)

##  [1] 0.7561462 0.9372188 0.9514638 0.9502348 0.9754091 0.9816299 0.9768297
##  [8] 0.9940707 0.9915841 0.9967900 0.9980584 0.9994837

Plotting

plot_training is a simple function used to plot the learning of one or more models.

# # # run a few models
model_list <- list(slpDIVA(st, tr), slpDIVA(st, tr), slpDIVA(st, tr))

# # # get response probabilities
model_list_resp_probs <- 
  lapply(model_list, function(x) {
    response_probs(tr, x$out, blocks = TRUE)})

# # # plot the leanrning curves
plot_training(model_list_resp_probs)

Advanced Topics

Examining the state of the model across training

ghonk/catlearn.suppls documentation built on May 3, 2019, 5:16 p.m.