9

I have a huge CSV file with about 25.5k records that (structurally) looks like the sample below. In essence, I want to display the values of the columns U,V,W,X,Y,Z of each row as a colored box. The color of each box should represent its value, i.e. the value dictates what color along some gradient is to be drawn. E.g.: The larger the value, the darker the box.

Now, there are a few things that I'd like to highlight using colors:

  • Group names should be written to the left of the item names, turned 90 degrees, once per group. (items are sorted by group)
  • The background colors of groups (and their items) should alternate
  • Within a group, item's background colors should alternate
  • There is a column special, which when TRUE/1 should change the item's bg-color(map) to a special/alternative color for the current group's bg-color. In total there are two possible group bg-colors, and four possible item bg-colors.
  • Also, when special is TRUE/1, the value color should be picked from a special/alternative gradient. In total there are two gradients/colormaps to pick from.
  • The labels/columns U,V,W,X,Y,Z at the bottom should also be colored with alternating bg-colors.
  • the labels/columns at the bottom should at a 90 degree angle, and aligned to the right (or top, in absolute terms).

Since the number of rows is so large, and I want the complete drawing to fit on a single page, the rows will need to be quite thin. (The columns should fit without compressing). Never the less, I do want the group and item text in there, because the dedicated reader should be able to zoom into the digital version of the final document, while regular/paper readers can obtain sufficient information from the alternating group colors.

Sadly, I have no idea how to start solving this problem.

conceptual rendering (done in a spreadsheet):
This is roughly what I hope to achieve. (Minus the grid lines.)

enter image description here

sample csv:

group,item,special,U,V,W,X,Y,Z
a,a1,0,0.2,,0.2,,,
a,a2,0,,0.1,,,0.4,1
a,a3,0,,0.5,,,,
a,a4*,1,0.1,,0.8,,,
a,a5*,1,,,,,0.5,0.5
a,a6,0,,,,0.3,,
b,b1,0,,0.1,,,,
b,b2,0,0.6,,,0.4,,
b,b3*,1,,,0.4,,,
c,c1*,1,,,,,,0.1
c,c2*,1,,0.2,,,0.3,0.2
c,c3,0,,,0.7,,,
c,c4,0,,0.6,,0.3,,
c,c5,0,,,,,,0.7
...

columns explained:

  • group: String
  • item: String
  • special: Boolean
  • U,V,W,X,Y,Z: real number in the range 0..1
6
  • Since just want the table in a single page or image, and with no text in each cell. I imagine it kind of like: data visualization. If I were you, I would be using python and matplotlib to just produce a single image (and \includegraphics if you want in a latex). See something like that heat map example. The important matplotlib command for this task is imshow Jan 15, 2014 at 14:30
  • If you are okay with going down the python road, then I can help you more with that. But it really feels like you need some programming anyway because you have to convert that CSV into latex. Jan 15, 2014 at 14:45
  • I'm okay with python, if I can use the exact same colors + colormaps I'm using in other pgfplots already. The \includegraphics route is fine by me.
    – derabbink
    Jan 15, 2014 at 14:55
  • I also would like to add that the final result should be a PDF with vectors; nothing rasterized.
    – derabbink
    Jan 15, 2014 at 15:01
  • Then I think it's best to use python read CSV and output some pgfplot code then. (no need to use matplotlib) Jan 15, 2014 at 15:01

1 Answer 1

9
+50

EDIT: After the OP stressed again that he want to use existing colormap from pgfplots, I think it might be better for him to use \pgfplotstabletypeset with a bit of preprocessing of csv.

However, if you're like me who is more comfortable with tikz/pgf and python, then the answer below should provide more flexibility. The one thing I'm not satisfied with this answer is that the table will not auto-adjust if group label is too long.


in.csv:

group,item,special,U,V,W,X,Y,Z
a,a1,0,0.2,,0.2,,,
a,a2,0,,0.1,,,0.4,1
a,a3,0,,0.5,,,,
a,a4*,1,0.1,,0.8,,,
a,a5*,1,,,,,0.5,0.5
a,a6,0,,,,0.3,,
b,b1,0,,0.1,,,,
b,b2,0,0.6,,,0.4,,
b,b3*,1,,,0.4,,,
c,c1*,1,,,,,,0.1
c,c2*,1,,0.2,,,0.3,0.2
c,c3,0,,,0.7,,,
c,c4,0,,0.6,,0.3,,
c,c5,0,,,,,,0.7

main.py:

import csv

# read csv into matrix
with open('in.csv') as csvf:
    f = csv.reader(csvf)
    a = [r for r in f]

# separate the header
header = a[0]
a = a[1:]

# separate by group
def sep_group(a):
    cur_group = None
    groups = []
    for r in a:
        # new group?
        if cur_group != r[0]:
            groups.append([r])
            cur_group = r[0]
        else:
            groups[-1].append(r)
    return groups
a = sep_group(a)

# calculating shade based on value in csv
def shade(x):
    return 0 if x == '' else str(int(float(x)*100))

# output while traversing data
with open('out.tex','w') as f:
    f.write('\\matrix [nodes={cell}] {\n')
    for gi,g in enumerate(a):
        for ri,r in enumerate(g):
            # empty node to place group label upon later
            f.write('\\node{}{} {{}}; &\n'.format(
                ' [alias=g{}begin]'.format(gi) if ri == 0 else '',
                ' [alias=g{}end]'.format(gi) if ri == len(g)-1 else ''))
            # item label
            f.write('\\node [group{}{}/item{},minimum width=3em] {{{}}}; &\n'.format(gi%2,'/special' if r[2] == '1' else '',ri%2,r[1]))
            # cells in the row
            f.write(' &\n'.join(
                '\\node [fill={}!{}] {{}};'.format('Emerald' if r[2]=='1' else 'black',shade(c))
                for c in r[3:]))
            f.write(' \\\\\n')
    # footer
    f.write('&')
    for ci in range(len(g[0])-3):
        f.write('&\n\\node [footer{},rotate=90] {{{}}}; '.format(ci%2,chr(ci+ord('U'))))
    f.write('\\\\\n')
    # end matrix
    f.write('};\n')
    # now overlay the group labels
    for gi,g in enumerate(a):
        f.write('\\node (last) [inner sep=0,group{},fit=(g{}begin) (g{}end)] {{}};\n'.format(gi%2,gi,gi))
        f.write('\\node [rotate=90,anchor=mid] at (last) {{{}}};\n'.format(g[0][0]))

running main.py produces out.tex:

\matrix [nodes={cell}] {
\node [alias=g0begin] {}; &
\node [group0/item0,minimum width=3em] {a1}; &
\node [fill=black!20] {}; &
\node [fill=black!0] {}; &
\node [fill=black!20] {}; &
\node [fill=black!0] {}; &
\node [fill=black!0] {}; &
\node [fill=black!0] {}; \\
\node {}; &
\node [group0/item1,minimum width=3em] {a2}; &
\node [fill=black!0] {}; &
...

out.tex is used in main.tex:

\documentclass{article}
\usepackage[dvipsnames]{xcolor}
\usepackage{tikz}
\usepackage[active,tightpage]{preview}
\usetikzlibrary{matrix}
\usetikzlibrary{fit}

\PreviewEnvironment{tikzpicture}
\setlength\PreviewBorder{5pt}

\begin{document}
\tikzset{
    cell/.style={outer sep=0pt, minimum size=2em},
    group0/.style={fill=blue!30},
    group0/.cd,
        item0/.style={fill=blue!30},
        item1/.style={fill=blue!60},
        special/.cd,
            item0/.style={fill=Emerald!30},
            item1/.style={fill=Emerald!60},
    %
    /tikz/.cd,
    group1/.style={fill=Dandelion!30},
    group1/.cd,
        item0/.style={fill=Dandelion!30},
        item1/.style={fill=Dandelion!60},
        special/.cd,
            item0/.style={fill=LimeGreen!30},
            item1/.style={fill=LimeGreen!30},
    %
    /tikz/.cd,
    footer0/.style={fill=YellowGreen!60},
    footer1/.style={fill=RedOrange!60},
}

\begin{tikzpicture}
    \input{out.tex}
\end{tikzpicture}
\end{document}

result:

output v3

This shows the matrix being painted in shades according to the values in CSV file. I try to use as similar color to OP as possible.

11
  • edits to add header and row labels to this answer are welcomed, as I might be too busy to modify this. Jan 15, 2014 at 16:03
  • I also haven't alternate the color between adjacent rows as well. Jan 15, 2014 at 16:09
  • updated, only 1 thing left: group label Jan 15, 2014 at 16:40
  • Looks good so far. Does it work with pgfplots's colormaps as well? N.B.: I only need two color(map)s to pick from for the value cells. For the item column, there should be four, though. I updated the question to reflect this clarification.
    – derabbink
    Jan 15, 2014 at 17:29
  • Actually, I have not used pgfplots before, but since pgfplots's based on tikz/pgf which is what I am using here. I assumed that it would work. Can you send a link to documentation/example of colormap to me? Jan 15, 2014 at 17:40

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .