How can i get the value of loss fun when use lbfgs method

[pep-pig]

i want to plot the convergence curves of different solve method ,but i can not get the value of loss function when use lbfgs,it only print value in cmd . i need save it into a list,how can i do that?

[chamoulox]

Hello,

I'm actually doing this too, Have you find something ?

I'm doing some tests so it's NOT good code:

As a beginning, i have created a global array (LOSS_TOTAL) and I have set a callback function in minimize_with_lbfgs function optimizer.minimize(sess,loss_callback=save_loss,fetches=[L_total])

and save loss is defined as follow:

def save_loss(loss):
LOSS_TOTAL.append(loss)

in order to plot, i have added some lines after the display of elapsing time in render_signe_image function:

print('Single image elapsed time: {}'.format(tock - tick))
plt.plot(LOSS_TOTAL)
plt.savefig('fig.jpg')

So now i'm able to see my loss function evolution, except it's not very useful as starting loss value is way higher than loss value after some iterations

My final goal if to stop computing if loss function reaches asymptotic value.

[pep-pig]

I add some code in lbfgs.py located in scipy\optimize , so i can plot the curve during the computing process. here is my project

[cysmith]

@vonlippmann Nice. Do you know why this happens?

[pep-pig]

i am sorry , i don't how this happen too.