def minimize(fn, epochs, batch_shape):
print('\n\nInfusing CONTENT with STYLE...\n')
t0 = datetime.now()
losses = []
epoch_duration = []
epoch_duration_float = []
x = np.random.randn(np.prod(batch_shape))
for i in range(epochs):
x, l, _ = fmin_l_bfgs_b(func=fn, x0=x, maxfun=20)
x = np.clip(x, -127, 127)
dt = datetime.now() - t0
total_min = dt.total_seconds() / 60
epoch_duration.append(dt)
epoch_duration_float.append(total_min)
print("epoch=%s, loss=%s, epoch duration = %s, total min = %.3f" %
(i, l, dt, total_min))
losses.append(l)
dt = datetime.now() - t0
print("\nTotal duration: %.3f" % np.array(epoch_duration_float).sum() +
' min')
plt.plot(losses)
plt.title("Optimization losses\nElapsed time: %.3f" %
np.array(epoch_duration_float).sum() + " min")
plt.show()
plt.plot(epoch_duration_float, color='orange')
plt.title('Epoch duration (minutes)')
plt.xlabel('Epoch')
plt.ylabel('Minutes')
plt.show()
newimg = x.reshape(*batch_shape)
final_img = unpreprocess(newimg)
return final_img[0]
def minimize(fn, epochs, batch_shape):
t0 = datetime.now()
losses = []
x = np.random.randn(np.prod(batch_shape))
for i in range(epochs):
x, l, _ = fmin_l_bfgs_b(func=fn, x0=x, maxfun=20)
x = np.clip(x, -127, 127)
print("iter=%s, loss=%s" % (i, l))
losses.append(l)
print("duration:", datetime.now() - t0)
plt.plot(losses)
plt.show()
newimg = x.reshape(*batch_shape)
final_img = unpreprocess(newimg)
return final_img[0]
def minimize(fn, epochs, batch_shape):
t0 = datetime.now()
losses = []
x = np.random.randn(np.prod(batch_shape))
for i in range(epochs):
x, l, _ = fmin_l_bfgs_b(
func=fn,
x0=x,
maxfun=20
)
x = np.clip(x, -127, 127)
print("iter=%s, loss=%s" % (i, l))
losses.append(l)
print("duration:", datetime.now() - t0)
plt.plot(losses)
plt.show()
newimg = x.reshape(*batch_shape)
final_img = unpreprocess(newimg)
return final_img[0]