NEWS (original) (raw)

handwriter 3.2.4

Fixed get_writer_profiles() andplot_writer_profiles() so that they are exported functions.

Created get_cluster_fill_rates() to calculate the proportion of graphs assigned to each cluster in a cluster template.
Created get_writer_profiles() to estimate writer profiles for all handwritten documents in the input directory. Writer profiles are the cluster fill counts or the cluster fill rates. This function runs process_batch_dir() to split the handwriting into component shapes called graphs. Then, it runsget_clusters_batch() to assign each graph to a cluster in a cluster template. Lastly, it runs get_cluster_fill_counts()or get_cluster_fill_rates() depending on whether the user chooses counts or rates.

Changed the writer_indices anddoc_indices arguments in get_clusters_batch()to be optional. The main reason for this change is so that users do not need to use a naming convention for their handwriting files in order to use handwriterRF::calculate_slr() andhandwriterRF::compare_documents() which callget_clusters_batch().
Fixed bug in get_clusters_batch() so that the warning “closing unused connection” no longer occurs if the user runs the function in parallel.

The new function plotGraphs() added in version 3.2.0 mistakenly was not exported. The function has been renamed asplot_graphs() and is now available to users.
Fixed a bug in get_clusters_batch() so it no longer returns the error message, “Unable to get cluster assignments for one or more documents,” if the file all_clusters.rds exists in the output directory.
Fixed a bug in get_clusters_batch() so it no longer mistakenly returns the error message, “Unable to get cluster assignments for one or more documents,” even if cluster assignments were successfully processed for all input documents.
If the graphs from the model training documents do not populate all available clusters in the cluster template, one or more of the graphs from the questioned document(s) could potentially be assigned to a cluster in the template that doesn’t contain any model training graphs. Previously, analyze_questioned_documents() threw an error if this occurs. Now, it only throws an error if more than 10% of the graphs from the questioned documents are assigned to clusters that are not populated by model training graphs. The mismatch between clusters used by the model document graphs and clusters used by the questioned document(s) graphs only becomes an issue if it occurs for a large number of graphs.

The new function plot_cluster_centers() creates a plot of the clusters centers from a cluster template. The cluster centers are displayed as orange shapes. The function also plots all graphs in each cluster as grey shapes with 5% transparency to depict the variability of graph shapes within each cluster.
The new function plotGraphs() plots every graph in a document processed with processDocument().

Fixed bug in processDocument() when the writing in the document is a single connected component, such as a single word written in cursive. Previously, the output ofprocessDocument() for this kind of document was formatted incorrectly.
Fixed bug in get_credible_intervals() andplot_credible_intervals() where the model writers were numbered sequentially. Now these functions use the writer IDs.
Fixed bug in format_template_data() where the function coerced writer IDs to integers even if the writer IDs contained characters.
Fixed bug in get_clusters_batch() where the function would stall but not return an error message if a document had a graph with a large number of edges (paths). Now the function ignores graphs with more than 30 edges.
Fixed bug in fit_model() where the function saves the same data in two separate files: “model_clusters.rds” and “all_clusters.rds”. The argument ‘save_master_file’ was added toget_clusters_batch(). If TRUE, a data frame of all cluster assignments, “all_clusters.rds”, will be saved. The default is FALSE.

The functions fit_model() andanalyze_questioned_documents() now allow writer IDs that contain numbers and letters.

Increased the speed of processHandwriting() by changing the function to process a handwritten document in sections instead of all at once. Nodes created byprocessHandwriting() in version 3.1.0 might differ slightly in placement from previous versions.
Fixed bugs in fit_model() andanalyze_questioned_documents() introduced by the changes toprocess_batch_list() in version 3.0.0.

Major reductions made to the memory required byprocess_batch_list() so it can now process paragraph length documents from the CSAFE Handwriting Database on machines with 8 GB of RAM.
process_batch_list() now skips to the next document if an error is returned while trying to process a document. A log file records the document name(s) and error message(s) of any problem documents. If the user reruns process_batch_list() they now have the option to either try a second time to process problem documents or skip them entirely.
New function get_clusters_batch() calculates cluster assignments of all files in a directory.

Fix initializer warning with arma::vec. Apparently different versions of gcc were raising warnings with the assignment of values in neighbors. The recommended fix was to use an initializer list.

fit_model() allows the user to fit a statistical model to known handwriting samples collected from a closed-set of persons.
analyze_questioned_documents() in conjunction with a model created by fit_model() allows a user to calculate the posterior probability that each known writer in the model is the writer of the questioned document(s).analyze_questioned_documents() only works when the questioned document(s) had to have been written by one of the model writers. This function must NOT be used if someone other than one of the model writers could written the questioned document(s) as it could yield misleading results.

processDocument() is a new wrapper function that runs readPNGBinary(), thinImage(), andprocessHandwriting() so the user doesn’t need to run these functions individually.
plotImage(), plotImageThinned(), andplotNodes now only need one input, a document processed with processDocument()
processHandwriting() no longer crashes when the input writing is a single letter or word
read_and_process() superseded in favor ofprocessDocument()
extractGraphs() superseded in favor ofprocess_batch_dir()