Add heatmap layer

[MySchizoBuddy]

Bokeh already allows heatmaps. Can you add a heatmap layer to rBokeh as well. For an example of existing interactive heatmap check out the d3heatmap R package. The dendrograms are not needed

[hafen]

Sounds good. This should be a very simply layer function making use of ly_image or ly_crect.

[MySchizoBuddy]

Bokeh has a build in heatmap function. http://bokeh.pydata.org/en/latest/docs/gallery/cat_heatmap_chart.html

[kaneplusplus]

+1 This would be helpful for gene expression data.

[ziyadsaeed]

was this feature released in the current comments. The current heatmap documentation is http://bokeh.pydata.org/en/latest/docs/gallery/heatmap_chart.html

[hafen]

Thanks for the link @ziyadsaeed. That is part of the high-level python api so we will still need to implement our own that translates to the rect glyphs. Should be pretty straightforward. I'll give this issue precedence.

[ziyadsaeed]

I see histogram related commits in the log. So is this feature implemented?

[hafen]

No sorry I'll get on it - should be pretty straightforward. @jrounds, did you have an implementation we could base off of?

[jrounds]

Honestly I am not sure. I would need to figure out

• how to handle arguments properly
• how to generalize a color gradient in an rbokeh way
• how to draw a color legend properly

I do have this snippet that demos how to use rbokeh to make heatmaps "by-hand" and it has the elements in place:

ly_heatMap = function(fig, x, y, heat_color_var, data,title){
        data = as.data.frame(data)
        #color
        colorQuantileFn = pnorm; 
        colorQuantileFn = ecdf(data[[heat_color_var]])
        fill_alpha=.8; width=.99; height=.99;  NA.color ="#333333"
        colorFn = colorRamp(c("blue", "white", "red"))
        colMapFn = function(v) {
                col = colorFn(colorQuantileFn(v))
                col[is.nan(v)] = 33  
                col[is.na(v)] = 33
                rgb(col, maxColorValue=255)
            }   
        color = colMapFn(data[[heat_color_var]])
        r = quantile(data[[heat_color_var]], c(0,10)/10, na.rm=TRUE)
        ld = list()
        ld$legend_values = seq(r[1],r[2], length.out=20)
        ld$width = unique(round(diff(ld$legend_values), 5))
        ld$legend_colors = colMapFn(ld$legend_values)
        ld$y = 1
        legend_data = as.data.frame(ld)
        #bquote({legend_data = .(legend_data)})
        legend = figure(height=150, width=800,tools=NULL, xlab="", ygrid=FALSE, xgrid=FALSE,xaxes="below", yaxes=FALSE , xlim=r) %>% ly_crect(x=legend_values,y=y, data=legend_data,color=legend_colors,width=width, line_alpha=0,fill_alpha=fill_alpha)

        hover_data = data
        hover_data = subset(hover_data, select=setdiff(colnames(hover_data), c("color")))

        p = fig %>% ly_crect( x=x, y=y, .99, .99, data=data, fill_color=color, line_color=color, fill_alpha=fill_alpha,hover=hover_data, legend=TRUE)
        p = p %>% theme_axis("x", major_label_orientation = 90)

        plots = list()
        plots[[title]] = p
        plots[["Legend"]] = legend
        p =  grid_plot(plots, nrow=2, ncol=1)
        p
}

library(reshape2)
mtscaled = as.matrix(scale(mtcars))
data =  melt(mtscaled, c("Model", "Measure"))
data[1:3,"value"] = as.numeric(NA)
call = list()
call$fig = figure()
call$data = data 
call$heat_color_var = heat_color_var = "value"
call$x= "Measure"
call$y = "Model"
call$title = "Car Model Heatmap"
do.call(ly_heatMap,call)

[colinbrislawn]

Any updates on this? I'm moving from trelliscope to trelliscopejs, and exploring rbokeh as an alternative to ggplot2. An elegant way to do heatmaps would be really nice.

[hafen]

I do plan to support this. I'm working on a complete overhaul of the internals of rbokeh right now that will clear things up a lot for easier future development. This should be out in 4-6 weeks. I'd prefer to add this to the updated version and think I (or a contributor) could do it pretty soon after that is released.

[colinbrislawn]

I'm glad to hear there is an update planned! Thank you for developing this constellation of software.

EDIT: Well this works surprisingly well.

df %>%
  figure(xgrid = FALSE, ygrid = FALSE, xlim = c(.5, 5.5), ylim = c(.5, 4.5)) %>%
    ly_crect(x, y, hover = value, color = value)

[colinbrislawn]

Wow, thanks for the tip, past Colin!

Here's how I'm going it today (for 2021 Colin).

df %>%
  figure(xgrid = FALSE, ygrid = FALSE, legend_location = NULL,
    width = 96*8, height = 96*7, toolbar_location = NULL) %>%
  ly_crect(x, y, hover = value, color = value) %>%
  theme_axis("x", major_label_orientation = 30) %>%
  x_range(c("a", "b", "c"))