LaTeX tables

The research_helpers.latex module assembles tables, while research_helpers.display reads them back. Neither needs anything outside the standard library.

Assembling

table() is keyword-only:

from research_helpers.latex import bare, dash_range, half_up, header, table

table(
    spec='lrr',
    header_rows=[[header('System'), header('Accuracy'), header('F1')]],
    body_rows=[['Baseline', '71.23', '68.90'], ['Ours', '84.56', '82.01']],
    caption='Performance on the held-out set, in percent.',
    label='tab:scores',
)

spec: The column specification, one character per column, as tabular takes it.
wide=True: Emits table* instead of table, e.g., for a two-column document where the table spans both.
environment: tabulary by default, which sizes columns to their content within a given width. Anything in SIZED_ENVIRONMENTS takes a width argument, tabular does not, and the width is left off.
prologue / setup: LaTeX inserted before the environment and inside it. For things like a \sisetup, or a \renewcommand{\arraystretch}.

The helpers

half_up(): Python’s round() rounds half to even: round(0.685, 2) is 0.68 and round(0.675, 2) is 0.68 as well. half_up rounds half to up, half_up(0.685, 2) yields 0.69.
header(): Wraps a heading in the size and weight the table style uses, and can expand it across columns.
bare(): Escapes LaTeX’s special characters, for text arriving from data.
dash_range(): Turns 1990-2000 into 1990--2000 (i.e. using an en-dash).
EM_DASH: For use on empty cells.

Maths in the table’s typeface

A table is usually set in a sans-serif face, while its maths comes from the document’s math font, which is drawn to sit beside the serif text. Beside the table’s own figures it looks too light and too small, and scaling it up makes it too large. research_helpers.textmath sets the maths as text in the table’s own font instead.

Switch it on for every table:

[tool.research-helpers.latex]
text-math = true

or for one, with table(..., text_math=True). Write the maths as usual, $Q_{ov}$ or $9.8\times 10^{-4}$ , in headers, cells, and captions, and table rewrites it:

Maths	Set as
`Q_{ov}`	`\textit{Q}\tmsubx{\textit{ov}}`
`9.8\times 10^{-4}`	`9.8\tmmed{}×\tmmed{}10\tmsup{−4}`
`\mu \le 0.5`	`\textit{μ}\tmthick{}≤\tmthick{}0.5`

The commands come from the rhtextmath package. Copy it beside the manuscript, and load it after the fonts:

$ research-helpers sty
written  tex/rhtextmath.sty

\usepackage{rhtextmath}

research-helpers sty --check reports a copy that no longer matches the installed research-helpers, and research-helpers arxiv uploads the copy with the manuscript.

What the package does:

Variables are italic, and numbers, operator names, and upper-case Greek upright, as TeX sets them.
A relation takes TeX’s thick space on either side, a binary operator its medium one, and a sign none.
Digits and signs in a script come from the font’s own superior and inferior glyphs (OpenType sups and subs), which are drawn at the right weight. A script holding anything else is scaled to \tmscriptscale of the surrounding size, 0.78 by default. \tmsubdrop and \tmsuprise place it. Change any of the three with \renewcommand after loading the package.
With a font that has no superior glyphs, every script is scaled.

The converter accepts the most common subset of maths used by tables: letters, numbers, Greek, the common relations and binary operators, operator names, \text, one level of scripts, \bar, bars, and brackets. Anything else, a fraction for instance, raises UnsupportedMathError instead of being set wrongly. For a caption written by hand, print the text form and paste it in:

$ research-helpers textmath '\mu = 0.5'
\textit{μ}\tmthick{}=\tmthick{}0.5

It needs XeLaTeX or LuaLaTeX, since the output is Unicode and the designed glyphs come from OpenType features.

Reading a table back

show() parses the .tex file of a generated table and renders it as a table readable in a notebook:

from pathlib import Path
from research_helpers.display import show

show(Path('tex/tables/scores.tex').read_text())

parse() returns the Table without displaying it (useful for tests), since it includes cells, spans, alignment, rules, caption, and label as data.

parsed = parse(latex)
assert parsed.columns() == 3
assert parsed.body[1].cells[0][0] == 'Ours'

symbols substitutes LaTeX commands for characters when rendering (\textdagger → †), and strip removes commands whose output does not matter for reading.