Plot confusion matrix with User/Producer accuracy

Plot confusion matrix from Cross-Validation, with accuracy (user/prod) as subplot.

Import librairies

from museotoolbox.ai import SuperLearner
from museotoolbox.cross_validation import RandomStratifiedKFold
from museotoolbox.charts import PlotConfusionMatrix
from museotoolbox import datasets
from sklearn.ensemble import RandomForestClassifier

Load HistoricalMap dataset

X,y = datasets.load_historical_data(low_res=True,return_X_y=True)

Create CV

RSKF = RandomStratifiedKFold(n_splits=2,
                random_state=12,verbose=False)

Initialize Random-Forest

classifier = RandomForestClassifier()

Start learning

SL = SuperLearner(classifier=classifier,param_grid=dict(n_estimators=[10,100]))
SL.fit(X,y,cv=RSKF)

Get kappa from each fold

for stats in SL.get_stats_from_cv(confusion_matrix=False,kappa=True):
    print(stats['kappa'])

Out:

0.9283198828158116
0.9208440420358427

Get each confusion matrix from folds

cms = []
for stats in SL.get_stats_from_cv(confusion_matrix=True):
    cms.append(stats['confusion_matrix'])
    print(stats['confusion_matrix'])

Out:

[[932  13   0   2   0]
 [ 31 255   0   3   0]
 [  0   0 283   1   0]
 [  0  12   3  51   0]
 [  0   1   0   0   0]]
[[930  16   0   1   0]
 [ 32 241   0  16   0]
 [  0   0 284   0   0]
 [  0   9   1  56   0]
 [  1   0   0   0   0]]

Plot confusion matrix

import numpy as np

# a bug in Sphinx doesn't show the whole plot, sorry.

labels = ['Forest','Agriculture','Bare soil','Water','Building']
from matplotlib.pyplot import cm as colorMap
meanCM = np.mean(cms,axis=0).astype(np.int16)
pltCM = PlotConfusionMatrix(meanCM.T) # Translate for Y = prediction and X = truth
pltCM.add_text()
pltCM.add_x_labels(labels,rotation=90)
pltCM.add_y_labels(labels)
pltCM.color_diagonal(diag_color=colorMap.Purples,matrix_color=colorMap.Reds)
pltCM.add_accuracy()
pltCM.add_f1()

Out:

/home/docs/checkouts/readthedocs.org/user_builds/museotoolbox/checkouts/latest/museotoolbox/charts/__init__.py:192: UserWarning: FixedFormatter should only be used together with FixedLocator
  rotation=rotation, fontsize=self.font_size)