# Visualizing averaging of multiple Confusion Matrices

import h5py

import matplotlib.pyplot as plt

fname = 'conf_mat_smush_inception_4e-3x3.h5'

dname = 'dataset'

layer = 1

prefix = 'conf_mat_smush_full_googlenet_inception_'

fnames = ['3a-output.h5', '3b-output.h5',

]

data = None

for fname in fnames:

print(prefix fname)

h5f = h5py.File(prefix fname, 'r')

if data is None:

data = h5f[dname][:]

else:

data = h5f[dname][:]

h5f.close()

prefix = 'conf_mat_smush_full_overfeat_'

fnames = ['10.h5', '11.h5',

]

for fname in fnames:

print(prefix fname)

h5f = h5py.File(prefix fname, 'r')

if data is None:

data = h5f[dname][:]

else:

data = h5f[dname][:]

h5f.close()

prefix = 'conf_mat_smush_full_caffenet_'

fnames = ['conv3.h5', 'conv4.h5',

]

for fname in fnames:

print(prefix fname)

h5f = h5py.File(prefix fname, 'r')

if data is None:

data = h5f[dname][:]

else:

data = h5f[dname][:]

h5f.close()

prefix = 'conf_mat_smush_full_vgg19_'

fnames = ['conv4_4.h5', 'conv5_4.h5',

]

for fname in fnames:

print(prefix fname)

h5f = h5py.File(prefix fname, 'r')

if data is None:

data = h5f[dname][:]

else:

data = h5f[dname][:]

h5f.close()

print(data)

h5f = h5py.File('conf_mat_sum.h5', 'w')

h5f.create_dataset('dataset', data=data)

h5f.close()

h5f = h5py.File('conf_mat_avg.h5', 'w')

h5f.create_dataset('dataset', data=data / 21.0)

h5f.close()

# If multiple confusion matrices are saved, only display a specific layer

if len(data.shape) == 3:

plt.imshow(data[layer, ...])

else:

plt.imshow(data)

plt.show()