PubMatrixR

Breaking changes

• plot_pubmatrix_heatmap() and pubmatrix_heatmap() now plot raw co-occurrence counts by default, and take a new values argument to choose otherwise. Earlier versions always plotted a percentage the docs called intersection / union * 100, but the totals in that formula were row and column sums of whatever matrix you passed in, not the publication counts for each term on its own. Adding an unrelated term changed the number shown in every other cell. The old calculation is still there as values = "relative", now documented with that caveat. values = "row_pct" divides each count by its row total.

Bug fixes

• PubMatrix() no longer drops columns when A repeats a term, or rows when B does. Duplicates now raise an error, and the result matrix is filled positionally instead of by name.
• Character matrices passed to the heatmap helpers keep their row names. A single-row matrix no longer comes back transposed into a single column.
• Term file parsing handles blank lines, and a # separator on the first or last line now gives an error that names the file instead of complaining about missing terms.
• A matrix where every value is the same now warns and draws a single-colour heatmap. It used to stop with an error.
• Heatmap validation errors no longer print internal call context.

Other improvements

• Requests are spaced out to stay under the NCBI rate limits: 3 per second without an API key, 10 with one. Set PubMatrixR.min_interval to override.
• Failed queries back off exponentially between retries, and the default number of attempts (PubMatrixR.n_tries) is now 3. Both options are documented.
• Queries drop usehistory=y and set retmax=0, so responses no longer carry PMIDs the package never looks at.