def on_epoch_end(self, iternum, GAN, logs={}):
out = io.BytesIO()
val = None
if self.title == 'genimg':
plots.save_img_grid(GAN.genrtor,
GAN.noise_vect_len,
fname=out,
Xterm=False,
scale=GAN.cscale)
elif self.title == 'pixhist':
val = plots.pix_intensity_hist(GAN.val_imgs,
GAN.genrtor,
GAN.noise_vect_len,
scaling=GAN.datascale,
fname=out,
Xterm=False)
out.seek(0)
image = _filebuf_to_tf_summary_img(out, self.title)
summary = tf.Summary(
value=[tf.Summary.Value(tag=self.title, image=image)])
writer = tf.summary.FileWriter(GAN.expDir + 'logs/imgs')
writer.add_summary(summary, iternum)
writer.close()
out.close()
return val
run = './expts/' + config + '/run' + runId + '/'
modelpath = run + 'models/g_cosmo_best.h5'
if not os.path.isfile(modelpath):
print("Error: File %s with pre-trained weights could not be found")
sys.exit()
GAN = GANbuild.DCGAN(config, run)
GAN.genrtor.load_weights(modelpath)
# Plot generated images
plots.save_img_grid(GAN.genrtor,
GAN.noise_vect_len,
GAN.invtransform,
GAN.C_axis,
Xterm=True,
scale=GAN.cscale,
multichannel=GAN.multichannel)
# Plot pixel intensity histogram and calculate chi-square score
chi = plots.pix_intensity_hist(GAN.val_imgs,
GAN.genrtor,
GAN.noise_vect_len,
GAN.invtransform,
GAN.C_axis,
multichannel=GAN.multichannel,
Xterm=True)
# Plot power spectrum and calculate chi-square score
pschi = plots.pspect(GAN.val_imgs,
#z = Input(shape=(1,noise_vect_len))
#genimg = genrtor(z)
#discrim.trainable = False
#decision = discrim(genimg)
#stacked = Model(z, decision)
#stacked.compile(loss=lossfn, optimizer=keras.optimizers.Adam(lr=0.0002, beta_1=0.5))
noise_vects1 = np.random.normal(loc=0.0, size=(10, 1, noise_vect_len))
reals = real_imgs[:10, :, :, :]
fakes = genrtor.predict(noise_vects1)
#plots.save_realimg_grid(real_imgs, Xterm=True, scale='lin')
#plots.save_img_grid(genrtor, noise_vect_len, 0, Xterm=True, scale='pwr')
plots.save_img_grid(genrtor, noise_vect_len, 0, Xterm=True, scale='pwr')
plots.save_img_grid(genrtor, noise_vect_len, 0, Xterm=True, scale='pwr')
#wdw = [-1.1, 1.1, 1e-4, 3e4]
chi = plots.pix_intensity_hist(real_imgs,
genrtor,
noise_vect_len,
'lin',
Xterm=True)
plt.show()
print('Chi=%f' % chi)
#noise_vects1 = np.random.normal(loc=0.0, size=(127000, 1, noise_vect_len))
#fakes = genrtor.predict(noise_vects1)
#np.save('./data/gen/fullcrop.npy', fakes)
'''
print(discrim.predict(reals))