Why the height of the bar3 decreases from original value as the size of the plot is varied?

I am trying to get a 3D bar plot with some titled layout, however the height of the bars tend to decrease from its original values. I have attached the code below. please suggest how to correct this.

data_K=[0.8741,0.7742,0.7410,0.7488];
data_S=[0.9438,0.8490,0.7821,0.7330];
data_G=[0.8992,0.8496,0.8082,0.7685];
data_TS_F1=[data_K;data_S;data_G]*100;
f=figure('visible', 'on','GraphicsSmoothing','on','WindowStyle','docked');
grid on;
t1 = tiledlayout(1,1,'Padding','tight');
t1.Units = 'inches';
t1.OuterPosition = [0.40 0.05 2 1.6];
nexttile;
hp=bar3(data_TS_F1);
set(gca, 'XTick', 1:width(data_TS), 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",12, 'ZTick',0:10:100);
ax=gca;
ax.FontWeight = 'bold';
set(hp(:),'facecolor',[0.745000000000000 0.745000000000000 1]);
xtickangle(0);
ytickangle(0);

[Bruno Luong]

width(data_TS) is probably incorrect in set command

[priyam kar]

Even if I remove this width the same thing happens, I have added the code below once again

data_K=[0.8741;0.7742;0.7410;0.7488];
data_S=[0.9438;0.8490;0.7821;0.7330];
data_G=[0.8992;0.8496;0.8082;0.7685];
data_TS=[data_K,data_S,data_G]*100;
% Clor=[0.745000000000000	0.745000000000000	1];
f=figure('visible', 'on','GraphicsSmoothing','on','WindowStyle','docked');
grid on;
t1 = tiledlayout(1,1,'Padding','tight');
t1 = tiledlayout(1,1,'Padding','tight');
t1.Units = 'inches';
t1.OuterPosition = [0.40 0.05 4 2];
nexttile;
hp=bar3(data_TS);
set(gca, 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",6, 'ZTick',0:20:100);
ax=gca;
ax.FontWeight = 'bold';
set(hp(:),'facecolor',[0.745000000000000	0.745000000000000	1]);
xtickangle(0);
ytickangle(0);

[priyam kar]

The plot is attached, if we look closely the height of the bar corresponding to SVM and 1 should be above 90, but in the plot it is around 80

[dpb]

That is an optical delusion owing to the particular orientation and spacing of the limits for the x-axis xlim.

Look at the adjacent bar height GNB,1 on the back axis grid and you'll observe it is above 80 and the bar for SVM,1 is clearly higher than it is.

Try

xl=xlim;
xlim(xl+0.5*[1 -1]

and see if it doesn't look more better...

Answer 1

[dpb]

data_K=[0.8741,0.7742,0.7410,0.7488];
data_S=[0.9438,0.8490,0.7821,0.7330];
data_G=[0.8992,0.8496,0.8082,0.7685];
data_TS_F1=[data_K;data_S;data_G]*100;
f=figure('visible', 'on','GraphicsSmoothing','on','WindowStyle','docked');
grid on;
t1 = tiledlayout(1,1,'Padding','tight');
t1.Units = 'inches';
t1.OuterPosition = [0.40 0.05 2 1.6];
t1.Units = 'normalized';
nexttile;
hp=bar3(data_TS_F1);
set(gca, 'XTick', 1:width(data_TS_F1), 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",12, 'ZTick',0:10:100);
ax=gca;
ax.FontWeight = 'bold';
set(hp(:),'facecolor',[0.745000000000000 0.745000000000000 1]);
xtickangle(0);
ytickangle(0);

It's not clear what you're trying to do here, but the documentation link to the tiledlayout <position property> includes the following specific note:

"This property affects other properties, such as the Position property. If you change the units, then return the units to the default value after completing your computation to prevent affecting other functions that assume the default value."

Resetting the 'Position' property back to 'normalized' does change the behavior; the above may be more in keeping with your intent...

[dpb]

ADDENDUM

I thought it was the mucking around with the overall size that was the Q?, not the data...

data_K=[0.8741,0.7742,0.7410,0.7488];
data_S=[0.9438,0.8490,0.7821,0.7330];
data_G=[0.8992,0.8496,0.8082,0.7685];
data_TS_F1=[data_K;data_S;data_G].'*100;  % Have to transpose to generate sample orientation
hp=bar3(data_TS_F1);
grid on;
set(gca, 'XTick', 1:width(data_TS_F1), 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",12, 'ZTick',0:10:100);
ax=gca;
ax.FontWeight = 'bold';
set(hp(:),'facecolor',[0.745000000000000 0.745000000000000 1]);
xtickangle(0);
ytickangle(0);
xlim

ans = 1×2

0.6000 3.4000

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

ylim

ans = 1×2

0 5

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

ylim(ylim+0.5*[1 -1])  % adjust ylim (note transposed from above note)

This removes most of the parallax issue about where it appears the bars are with respect to the grid lines...

The problem is appearances, not that the data/figure are wrong.

Changing the viewing orientation some could also help.

[priyam kar]

Thank you very much, changing the ylim values has made things fall in correct places.

[dpb]

Things are in the right place; it's just a parallax issue with the chosen/default orientation and axis limits not being at the edge of the bars...

data_K=[0.8741,0.7742,0.7410,0.7488];
data_S=[0.9438,0.8490,0.7821,0.7330];
data_G=[0.8992,0.8496,0.8082,0.7685];
data_TS_F1=[data_K;data_S;data_G].'*100;  % Have to transpose to generate sample orientation
hp=bar3(data_TS_F1);
grid on;
hAx=gca;
set(hAx,'XTick', 1:width(data_TS_F1), 'XTickLabels', {'KNN','SVM','GNB'}, ...
        'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman", ...
        "FontSize",10,'FontWeight','bold', 'ZTick',0:10:100);
set(hp(:),'facecolor',[0.745 0.745 1]);
xtickangle(0); ytickangle(0);
ylim(ylim+0.6*[1 -1])  % adjust ylim (note transposed from above note)
pos=hAx.CameraPosition

pos = 1×3

-15.1623 19.4984 576.7263

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

hAx.CameraPosition=hAx.CameraPosition.*[1.25 1.01 0.6]; % and rotate some as well...

Moving the y axis to be identically on the edge of the bars eliminates the parallax entirely and also rotating the camera location to be more from the side helps in the visualization of the differences in bar heights.

"Salt to suit.." -- Note you can manually rotate the figure and get what is to you the most pleasing angle; the camera position is set in terms of data coordinates so it isn't exactly intuitive as to how to change them.

[priyam kar]

Yes, now I understood the parplex created due the camera postion and the axis not being exactly on the edge, Thank you very much for clarifyng the doubt.

[dpb]

No problem, glad to help.

NOTA BENE: You could see the issue more clearly if you were to redraw the original and before changing ylim draw a horizontal line at the height of SVM,1 parallel to the x- axis. You would see it project back to the limits of the y-axis and intersect above the grid line at that height.

[Bruno Luong]

Franckly I see different ylimit, the diffent view point but the bar height remains identical. Not sure what is the question and answer here. My perspecctive visualisation brain does not seem to work like yours obviously.

[dpb]

If you look at the image @priyam kar posted of his original, the grid line at 80 extends almost directly to the back top edge of the SVM,1 bar, Bruno. @priyam kar was misinterpreting that as indicating the bar height was only 80; not projecting the bar in the represented 3D space. Hence, he thought the bar itself had been redrawn to be only ~80 units high, not recognizing the top of the bar itself when projected back to the axis would be/is really at the full height. As noted/agreed, it is a misreading of the plot created by the specific combination of data and location and chosen z axis ticks in confjunction with the default ylim and camera positions.

One could rotate and not change ylim and make it more clearly apparent where the bars are in space; I just shortened the y-axis first as being simpler to make the grid itelf be drawn closer to the bars so the location of the grid lines did fall closer to the actual bar edges.

I think it is a typical optical illusion trick; if one doesn't initially relate the line to the bar, then one will see the correct interpretation but if one ever does it the other way, then it can be difficult to change...

[dpb]

data_K=[0.8741,0.7742,0.7410,0.7488];
data_S=[0.9438,0.8490,0.7821,0.7330];
data_G=[0.8992,0.8496,0.8082,0.7685];
data_TS_F1=[data_K;data_S;data_G].'*100;
grid on;
hp=bar3(data_TS_F1);
set(gca, 'XTick', 1:width(data_TS_F1), 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",10, 'ZTick',0:10:100);
set(hp(:),'facecolor',[0.745000000000000 0.745000000000000 1]);
xtickangle(0);ytickangle(0);
z=data_S(1)*100;
line([2 2],[ylim],[z z],'linestyle','-','color','r')
line([xlim],[1 1],[z z],'linestyle','-','color','r')

Draws the projection of the height of Z(SVM,1) back to the axes grid that shows where the height projection is...I think in the original unaltered view the grid lines can make one easily misinterpret and it is difficult to really project where that intersection point is owing to the distance the grid lines are from the actual bar positions.

figure
hp=bar3(data_TS_F1);
set(gca, 'XTick', 1:width(data_TS_F1), 'XTickLabels', {'KNN','SVM','GNB'}, ...
    'YTick', 1:4, 'YTickLabels', [1,3,6,12],'FontName',"Times New Roman","FontSize",10, 'ZTick',0:10:100);
set(hp(:),'facecolor',[0.745000000000000 0.745000000000000 1]);
xtickangle(0);ytickangle(0);
z=data_S(1)*100;
line([2 2],[ylim],[z z],'linestyle','-','color','r')
line([xlim],[1 1],[z z],'linestyle','-','color','r')
hAx=gca;
hAx.CameraPosition=hAx.CameraPosition.*[1.35 0.5 0.4]; % and rotate some as well...

If you rotate it around to more nearly the yz projection one can see where the bar tops are in relation to the grid lines a little more clearly, perhaps.