Jupyter plots displaying error?

[xiaohk]

When I run the Jupyter cell using IJulia, the plot is displayed correctly. However, when I reopen the same notebook, the output of that cell becomes some texts rather than the plots.

I have pushed the Jupyter notebook to Github, and there seems an error displaying my plots as well. The displaying error is not corresponding to the strange text output. When I open the Jupyter locally, all plots have become texts, but there are only some plots can't be displayed in Github.

I tried to re-run all the cells in Jupyter notebook. When I reopen the notebook, all the plots became texts. On Github, the displaying error occurs to the same plots.

I am wondering how to fix it?

[tlnagy]

That's weird. What is your Julia version and could you run Pkg.status()?

I have some notebooks for a recent project of mine that have SVGs and they seem to work fine: https://github.com/tlnagy/Crispulator.jl/blob/master/meta/growth_scan_diff_amounts_increasing.ipynb

[xiaohk]

@tlnagy Yeah some SVGS can be displayed correctly, but some can't. My Julia version is 0.5.

julia> Pkg.status()
11 required packages:
 - Cairo                         0.2.35
 - Colors                        0.6.9
 - Fontconfig                    0.1.1
 - Gadfly                        0.5.2
 - IJulia                        1.3.3
 - ImageMagick                   0.1.8
 - ImageView                     0.2.0
 - Images                        0.5.14
 - MAT                           0.3.1
 - Optim                         0.7.4
 - QuartzImageIO                 0.1.4
60 additional packages:
 - AxisAlgorithms                0.1.5
 - BinDeps                       0.4.5
 - Blosc                         0.1.7
 - BufferedStreams               0.2.0
 - Calculus                      0.1.15
 - ColorTypes                    0.2.12
 - ColorVectorSpace              0.1.12
 - Compat                        0.10.0
 - Compose                       0.4.4
 - Conda                         0.4.0
 - Contour                       0.2.0
 - DataArrays                    0.3.11
 - DataFrames                    0.8.5
 - DataStructures                0.5.1
 - Dates                         0.4.4
 - DiffBase                      0.0.2
 - Distances                     0.3.2
 - Distributions                 0.11.1
 - FileIO                        0.2.1
 - FixedPointNumbers             0.2.1
 - FixedSizeArrays               0.2.5
 - ForwardDiff                   0.3.3
 - GZip                          0.2.20
 - Graphics                      0.1.3
 - HDF5                          0.7.2
 - Hexagons                      0.0.4
 - Hiccup                        0.1.1
 - Homebrew                      0.4.2
 - IniFile                       0.2.5
 - Interpolations                0.3.6
 - Iterators                     0.2.0
 - JSON                          0.8.0
 - Juno                          0.2.5
 - KernelDensity                 0.3.0
 - Libz                          0.2.1
 - LineSearches                  0.1.3
 - Loess                         0.1.0
 - MacroTools                    0.3.4
 - Measures                      0.0.3
 - Media                         0.2.4
 - NaNMath                       0.2.2
 - Nettle                        0.2.4
 - PDMats                        0.5.2
 - PositiveFactorizations        0.0.3
 - Ratios                        0.0.4
 - Reexport                      0.0.3
 - Rmath                         0.1.6
 - SHA                           0.3.0
 - SIUnits                       0.1.0
 - Showoff                       0.0.7
 - SortingAlgorithms             0.1.0
 - StatsBase                     0.12.0
 - StatsFuns                     0.3.1
 - TexExtensions                 0.0.3
 - Tk                            0.4.0
 - URIParser                     0.1.6
 - Winston                       0.12.1
 - WoodburyMatrices              0.2.1
 - ZMQ                           0.4.0
 - Zlib                          0.1.12

[tlnagy]

Do you happen to have examples that display correctly versus those that don't? Or does it seem to be random?

[xiaohk]

@tlnagy I think it happens kinda randomly. I have the following function, which displays multiple plots in a loop:

# Iterate within different λ
function multiple_λ(λ_candidates)
    for my_λ in λ_candidates
        cost1, g1! = cost_and_gradient(zeros(28, 1), my_λ)

        # Find the parameters
        res = optimize(cost1, g1!, repeat([0.5], inner = size(feature, 2)))
        mini_Θ = Optim.minimizer(res)

        # Plot the decision boundaries
        l2 = layer(z = (x1,x2) -> decision(x1, x2, mini_Θ), 
                   x = linspace(-1.0, 1.5, 1000),
                   y = linspace(-1.0, 1.5, 1000),
                   Geom.contour(levels = [0.0]),
                   Theme(line_width = 1pt))

        p5 = plot(l1, l2, coord, 
                  Scale.color_discrete_manual(colorant"deep sky blue",
                                              colorant"light pink"),
                  Guide.title("Regularization with λ = $(my_λ)"))

        display(p5)
    end
end

If I run multiple_λ([0, 1, 5, 10, 100]), it should have displayed five plots, but the output displayed on Github looks like:

[tlnagy]

What if you run multiple_λ([1, 1, 1, 10, 10])?

[xiaohk]

@tlnagy It follows the old pattern. I thought the error happens alternately, but in the whole running process 7 plots are displayed correct, fail, fail, correct, fail, correct, fail, correct.

[xiaohk]

@tlnagy It might be helpful to see the notebook. https://github.com/xiaohk1997/test/blob/master/exercise2.ipynb

[tlnagy]

That's super weird. All the 1 plots should be identical, no?

[xiaohk]

@tlnagy Yes, they should all be the same.

[tlnagy]

No idea what's going on, maybe cross-post this with IJulia.jl? Maybe @stevengj might know?

[xiaohk]

@tlnagy Ok, thanks! BTW, the plots are displayed as texts when I reopen the notebook locally.

Is this normal?

[tlnagy]

That isn't normal either haha. I haven't seen that before. Can you open the .ipynb file and copy the code corresponding to that cell? It looks like the SVG code is getting stripped.

[xiaohk]

@tlnagy All the plots are displayed like that locally. That code is:

# Visualizing the data
p3 = plot(df, x = :test1, y = :test2, color = :result,  
          Scale.color_discrete_manual(colorant"deep sky blue",
                                      colorant"light pink"))

You can visit the .ipynb file here: https://github.com/xiaohk1997/test/blob/master/exercise2.ipynb

[tlnagy]

I meant the raw JSON code of the ipynb:

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1. Logistic Regression\n",
    "In this part, we are going to implement the logistic regression to classify admission given two feature.\n",
    "## 1.1 Visualize the data\n",
    "In this exercise, I want to use the `DataFrame.jl` to implement. We now can import the data as below."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [

....

The SVGs look fine there. Not sure what is up.