I have been trouble with trying to find a way to display a 3 element list in the form of a table. What I actually care about is drawing the table. I would like to draw a 1by3 table for each ylabel in a plot.
Below is what I have so far. If I can get each Table instance to show up, I will have what I want. Right now a reference to a table appears and I'm not sure why. If you actually look in the center left where the reference locations appear, you can see one 1by3 table.
Is it possible using matplotlib to generate a new table for each ylabel? The table info is directly related to each row in the bar graph, so it's important that I have a way that they line up.
The number of rows in the bar graph is dynamic, so creating 1 table for the whole figure and trying to dynamically line up the rows with the corresponding bar graph is a difficult problem.
# initialize figure
fig = plt.figure()
gs = gridspec.GridSpec(1, 2, width_ratios=[2, 1])
fig.set_size_inches(18.5, 10.5)
ax = fig.add_subplot(gs[0])
#excluded bar graph code
# create ylabels
for row in range(1,len(data)):
ylabel = [str(data[row][0]),str(data[row][1]),str(data[row][2])]
ylabels.append(ylabel)
#attempting to put each ylabel in a 1by3 table to display
pos = np.arange(0.5,10.5,0.5)
axTables = [None] * len(ylabels)
for x in range(0,len(ylabels)):
axTables[x] = fig.add_subplot(gs[0])
ylabels[x] = axTables[x].table(cellText=[ylabels[x]], loc='left')
locsy, labelsy = plt.yticks(pos,ylabels)
First, yticks will expect text as input, it cannot handle other objects. Second, a table needs to sit within an axes.
So in order to get a table at the position of a tick(label) the idea can be to create an axes at the position of a y ticklabel. An option is the use of mpl_toolkits.axes_grid1.inset_locator.inset_axes. Now the difficulty is that this axes needs to be positionned in data coordinates along the y axis, and in figure (or pixel-) coorinates in the horizontal direction. For this one might use a blended transform. The inset_axes allows to give the width and height as absolute measures (in inches) or in relative, which is handy because we can set the width of the axes to 100% of the bounding box, while the height is still some absolute value (we don't want the axes height to depend on the data coordinates!).
In the following a function ax_at_posy creates such axes.
The table would then sit tight inside the axes, such that all columns are the same width. One would still need to make sure the same fontsize is used throughout.
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1.inset_locator import inset_axes
import matplotlib.transforms as mtrans
# General helper function to create an axes at the position of a yticklabel
def ax_at_posy(y, ax=None, width=0.3, leftspace=0.08, height=0.2):
ax = ax or plt.gca()
trans = mtrans.blended_transform_factory(ax.figure.transFigure, ax.transData)
axins = inset_axes(ax, "100%", height,
bbox_to_anchor=(leftspace, y, width-leftspace, 0.05),
bbox_transform=trans, loc="center right", borderpad=0.8)
axins.tick_params(left=False, bottom=False, labelleft=False, labelbottom=False)
axins.axis("off")
return axins
fig, ax = plt.subplots()
fig.subplots_adjust(left=0.4)
ax.scatter(np.random.rand(30), np.random.randint(7, size=30), c=np.random.rand(30))
get_data = lambda i: "".join(np.random.choice(list("abcdefgxyzt0"), size=i+2))
data = np.vectorize(get_data)(np.random.randint(2,6,size=(7,3)))
for i, row in enumerate(data):
axi = ax_at_posy(i, ax=ax, width=0.4)
tab = axi.table(cellText=[list(row)], loc='center', bbox=(0,0,1,1))
tab.auto_set_font_size(False)
tab.set_fontsize(9)
plt.setp(tab.get_celld().values(), linewidth=0.72)
plt.show()
Related
I have tried around 15 different methods for setting the y-label for this simple confusion matrix visualization code. Currently, I have resorted to just directly labeling the rows as 'Predicted Positive' and 'Predicted Negative' but I would prefer to have 'Predicted' outside the table like I do with 'Actual'. Very confused what's going wrong. I'm assuming it has something to do with the fact that I'm plotting a table. Removing the row labels does not fix the issue. Thanks in advance!
def plot_conf_mat(data, model_name):
'''
Plot a confusion matrix based on the array data.
Expected: 2x2 matrix of form
[[TP, FP],
[FN, TN]].
Outputs a simple colored confusion matrix table
'''
#set fontsizes
SMALL_SIZE = 20
MEDIUM_SIZE = 25
BIGGER_SIZE = 30
plt.rc('font', size=MEDIUM_SIZE) # controls default text sizes
plt.rc('axes', titlesize=MEDIUM_SIZE) # fontsize of the axes title
plt.rc('axes', labelsize=SMALL_SIZE) # fontsize of the x and y labels
plt.rc('xtick', labelsize=SMALL_SIZE) # fontsize of the tick labels
plt.rc('ytick', labelsize=SMALL_SIZE) # fontsize of the tick labels
plt.rc('legend', fontsize=SMALL_SIZE) # legend fontsize
plt.rc('figure', titlesize=BIGGER_SIZE) # fontsize of the figure title
# Prepare table
columns = ('Positive', 'Negative')
rows = ('Predicted\nPositive', 'Predicted\nNegative')
cell_text = data
# Add a table at the bottom of the axes
colors = [["tab:green","tab:red"],[ "tab:red","tab:grey"]]
fig, ax = plt.subplots(figsize = (6,5))
ax.axis('tight')
ax.axis('off')
the_table = ax.table(cellText=cell_text,cellColours=colors,
colLabels=columns, rowLabels = rows, loc='center')
the_table.scale(2,5)
the_table.set_fontsize(20) #apparently it doesnt adhere to plt.rc??
ax.set_title(f'{model_name} Confusion Matrix: \n\nActual')
ax.set_ylabel('Predicted') #doesn't work!!
fig.savefig(f"{model_name}_conf_mat.pdf", bbox_inches = 'tight')
plt.show()
Out (model name redacted):
Firstly, did you know that there is a sklearn.metrics visualization option called ConfusionMatrixDisplay which might do what you are looking for. Do see if that helps.
For the table itself, matplotlib table is used to add a table to an axis. It usually contains a plot along with the table. As you only need a table, you are hiding the plot. If you comment out the line ax.axis('off'), you will see the borders of the plot. The ax.set_ylabel() will not work for this reason, as it is the label for the plot, which is hidden.
A simple workaround is to add text at the right place. Adding this instead of the set_ylabel() did the trick. You may need to fine tune the x and y coordinates.
plt.text(-0.155, -0.0275,'Predicted', fontsize= SMALL_SIZE, rotation=90)
My plot function creates horizontal bars per year for data with different size. I have to change the figure size for each set of subplots.
I need to place my two legends on lower center of each figure below the x axis label. The positions need to vary depending on the figure size and remain consistent. So for all produced figures, the legends would look like this figure.
Find a snippet of my dataframe here. I have tried to simplify the code as much as I could and I know the plot is missing some element, but I just want to get to my question's answer, not to create a perfect plot here. I understand probably I need to create a variable for my anchor bounding box but I don't know how. Here is my code:
def plot_bars(data,ax):
""" Plots a single chart of work plan for a specific routeid
data: dataframe with section length and year
Returns: None"""
ax.barh(df['year'], df['sec_len'] , left = df['sec_begin'])
ax.set_yticklabels('')
def plot_fig(df):
# Draw the plots
ax_set = df[['routeid','num_bars']].drop_duplicates('routeid')
route_set = ax_set['routeid'].values
h_ratios = ax_set['num_bars'].values
len_ratio = h_ratios.sum()/BARS_PER_PAGE # Global constant set to 40 based on experiencing
fig, axes = plt.subplots(len(route_set), 1, squeeze=False, sharex=True
, gridspec_kw={'height_ratios':h_ratios}
, figsize=(10.25,7.5*len_ratio))
for i, r in enumerate(route_set):
plot_bars(df[df['routeid']==r], axes[i,0])
plt.xlabel('Section length')
## legends
fig.legend(labels=['Legend2'], loc=8, bbox_to_anchor=(0.5, -0.45))
fig.legend( labels=['Legend1'], loc = 8, bbox_to_anchor=(0.5, -0.3))
## Title
fig.suptitle('title', fontsize=16, y=1)
fig.subplots_adjust(hspace=0, top = 1-0.03/len_ratio)
for df in df_list:
plot_fig(df)
The problem is when the figure size changes, the legends move as in these pictures:
here
here
I think the problem boils down to having the correct relative position with respect to the xlabel, so are right that you need to calculate the bbox_to_anchor using the position of the xlabel and the height/width of the axes. Something like this:
fig, (ax, ax1) = plt.subplots(nrows=2, figsize=(5, 4), gridspec_kw={'height_ratios':[4, 1]})
ax.plot(range(10), range(10), label="myLabel")
ax.set_xlabel("xlabel")
x, y = ax.xaxis.get_label().get_position() # position of xlabel
h, w = ax.bbox.height, ax.bbox.width # height and width of the Axes
leg_pos = [x + 0 / w, y - 55 / h] # this needs to be adjusted according to your needs
fig.legend(loc="lower center", bbox_to_anchor=leg_pos, bbox_transform=ax.transAxes)
plt.show()
I'd like to find a way to make an annotation that automatically aligns with the label text of a colorbar. Take this example:
import matplotlib.pyplot as plt
import numpy as np
fig, ax = plt.subplots(figsize=(5,10))
data = np.arange(1000, 0, -10).reshape(10, 10)
im = ax.imshow(data, cmap='Blues')
clb = plt.colorbar(im, shrink=0.4)
clb.ax.annotate('text', xy=(1, -0.075), xycoords='axes fraction')
I want to have the last t of "text" to be on the same x coordinate as the last 0 of 1000 in the colorbar label. I can do so manually by adjusting the xy parameter in annotate, but I have to do this for many graphs and would like to find a way to get the parameter from somewhere automatically.
How can I get the maximum x coordinate of the text labes and annotate in a way where the annotation ends on that coordinate? Could someone point me in the right direction? Thanks a lot!
Since the labels are left-aligned, but you want to align your additional text according to the end of that label, I fear there is no other choice than to find out the coordinates from the drawn figure and place the label accordingly.
import matplotlib.pyplot as plt
from matplotlib import transforms
import numpy as np
fig, ax = plt.subplots(figsize=(5,4))
data = np.arange(1000, 0, -10).reshape(10, 10)
im = ax.imshow(data, cmap='Blues')
cbar = plt.colorbar(im)
# draw figure first to be able to retrieve coordinates
fig.canvas.draw()
# get the bounding box of the last label
bbox = cbar.ax.get_yticklabels()[-1].get_window_extent()
# calculate pixels back to axes coords
labx,_ = cbar.ax.transAxes.inverted().transform([bbox.x1,0])
ax.annotate('text', xy=(labx, -0.075), xycoords=cbar.ax.transAxes,
ha = "right")
plt.show()
Note that this approach will fail once you change the figure size afterwards or change the layout in any other way. It should hence always come last in your code.
I am trying to create a nice plot which joins a 4x4 grid of subplots (placed with gridspec, each subplot is 8x8 pixels ). I constantly struggle getting the spacing between the plots to match what I am trying to tell it to do. I imagine the problem is arising from plotting a color bar on the right side of the figure, and adjusting the location of the plots in the figure to accommodate. However, it appears that this issue crops up even without the color bar included, which has further confused me. It may also have to do with the margin spacing. The images shown below are produced by the associated code. As you can see, I am trying to get the space between the plots to go to zero, but it doesn't seem to be working. Can anyone advise?
fig = plt.figure('W Heat Map', (18., 15.))
gs = gridspec.GridSpec(4,4)
gs.update(wspace=0., hspace=0.)
for index in indices:
loc = (i,j) #determined by the code
ax = plt.subplot(gs[loc])
c = ax.pcolor(physHeatArr[index,:,:], vmin=0, vmax=1500)
# take off axes
ax.axis('off')
ax.set_aspect('equal')
fig.subplots_adjust(right=0.8,top=0.9,bottom=0.1)
cbar_ax = heatFig.add_axes([0.85, 0.15, 0.05, 0.7])
cbar = heatFig.colorbar(c, cax=cbar_ax)
cbar_ax.tick_params(labelsize=16)
fig.savefig("heatMap.jpg")
Similarly, in making a square figure without the color bar:
fig = plt.figure('W Heat Map', (15., 15.))
gs = gridspec.GridSpec(4,4)
gs.update(wspace=0., hspace=0.)
for index in indices:
loc = (i,j) #determined by the code
ax = plt.subplot(gs[loc])
c = ax.pcolor(physHeatArr[index,:,:], vmin=0, vmax=400, cmap=plt.get_cmap("Reds_r"))
# take off axes
ax.axis('off')
ax.set_aspect('equal')
fig.savefig("heatMap.jpg")
When the axes aspect ratio is set to not automatically adjust (e.g. using set_aspect("equal") or a numeric aspect, or in general using imshow), there might be some white space between the subplots, even if wspace and hspaceare set to 0. In order to eliminate white space between figures, you may have a look at the following questions
How to remove gaps between *images* in matplotlib?
How to combine gridspec with plt.subplots() to eliminate space between rows of subplots
How to remove the space between subplots in matplotlib.pyplot?
You may first consider this answer to the first question, where the solution is to build a single array out of the individual arrays and then plot this single array using pcolor, pcolormesh or imshow. This makes it especially comfortable to add a colorbar later on.
Otherwise consider setting the figuresize and subplot parameters such that no whitespae will remain. Formulas for that calculation are found in this answer to the second question.
An adapted version with colorbar would look like this:
import matplotlib.pyplot as plt
import matplotlib.colors
import matplotlib.cm
import numpy as np
image = np.random.rand(16,8,8)
aspect = 1.
n = 4 # number of rows
m = 4 # numberof columns
bottom = 0.1; left=0.05
top=1.-bottom; right = 1.-0.18
fisasp = (1-bottom-(1-top))/float( 1-left-(1-right) )
#widthspace, relative to subplot size
wspace=0 # set to zero for no spacing
hspace=wspace/float(aspect)
#fix the figure height
figheight= 4 # inch
figwidth = (m + (m-1)*wspace)/float((n+(n-1)*hspace)*aspect)*figheight*fisasp
fig, axes = plt.subplots(nrows=n, ncols=m, figsize=(figwidth, figheight))
plt.subplots_adjust(top=top, bottom=bottom, left=left, right=right,
wspace=wspace, hspace=hspace)
#use a normalization to make sure the colormapping is the same for all subplots
norm=matplotlib.colors.Normalize(vmin=0, vmax=1 )
for i, ax in enumerate(axes.flatten()):
ax.imshow(image[i, :,:], cmap = "RdBu", norm=norm)
ax.axis('off')
# use a scalarmappable derived from the norm instance to create colorbar
sm = matplotlib.cm.ScalarMappable(cmap="RdBu", norm=norm)
sm.set_array([])
cax = fig.add_axes([right+0.035, bottom, 0.035, top-bottom])
fig.colorbar(sm, cax=cax)
plt.show()
I saw this example on how to create a parallel coordinate plot: Parallel Coordinates:
This creates a nice Parallel Coordinates figure, but I would like to add this plot to an already existing figure in a subplot (there should be another plot next to it in the same plot).
For the already existing figure, the figure and axes are defined as:
fig = plt.figure(figsize=plt.figaspect(2.))
ax = fig.add_subplot(1,2,1)
For the Parallel Coordinates, they suggest:
fig, axes = plt.subplots(1, dims-1, sharey=False)
How can I reconcile both initializations of the figure and the ax(es)?
One option is to create all the axes using subplots then just shift the location of the one that you don't want to have wspace=0 as is done for the Parallel Coordinate plots:
import matplotlib.pylab as plt
dims = 4
fig, axes = plt.subplots(1, dims-1 + 1, sharey=False)
plt.subplots_adjust(wspace=0)
ax1 = axes[0]
pos = ax1.get_position()
ax1.set_position(pos.translated(tx = -0.1,ty=0))
I have added 1 to the number of columns creates (leaving it explicitly -1+1) and set wspace=0 which draws all the plots adjacent to one another with no space inbetween. Take the left most axes and get the position which is a Bbox. This is nice as it gives you the ability to translate it by tx=-0.1 separating your existing figure.