SpringRank
SpringRank copied to clipboard
cdebacco
Which beta is being calculated by get_optimal_temperature on tools.py?
On tools.py which beta is calculated by get_optimal_temperature? Global or local beta? Would you please be so kind providing functions to calculate both? Thank you very much indeed!
This is the beta maximizing the conditional loglikelihood, i.e. the global one. It uses a root-finding method (brentq), to solve eq S39 of the paper (in the Supplementary Material). I will upload in the next few days a similar equivalent to calculate the local one, i.e. an implementation for eq. S36.
I will also upload an alternative method: instead of solving via root-finding method, one can use scipy.optimize.fmin_l_bfgs_b() to make python search for the optimal parameter value in order to maximize a function. This method allows to define custom cost functions, scipy.optimize does the rest.
This is the beta maximizing the conditional loglikelihood, i.e. the global one. It uses a root-finding method (brentq), to solve eq S39 of the paper (in the Supplementary Material). I will upload in the next few days a similar equivalent to calculate the local one, i.e. an implementation for eq. S36.
I will also upload an alternative method: instead of solving via root-finding method, one can use scipy.optimize.fmin_l_bfgs_b() to make python search for the optimal parameter value in order to maximize a function. This method allows to define custom cost functions, scipy.optimize does the rest.
Il giorno lun 17 dic 2018 alle ore 13:04 edsf70 [email protected] ha scritto:
On tools.py which beta is calculated by get_optimal_temperature? Global or local beta? Would you please be so kind providing functions to calculate both? Thank you very much indeed!
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/cdebacco/SpringRank/issues/8, or mute the thread https://github.com/notifications/unsubscribe-auth/ARSx93_I3sx9u52wwdXd33Ad3o_DHy7gks5u54hagaJpZM4ZWOOf .
This is the beta maximizing the conditional loglikelihood, i.e. the global one. It uses a root-finding method (brentq), to solve eq S39 of the paper (in the Supplementary Material). I will upload in the next few days a similar equivalent to calculate the local one, i.e. an implementation for eq. S36. I will also upload an alternative method: instead of solving via root-finding method, one can use scipy.optimize.fmin_l_bfgs_b() to make python search for the optimal parameter value in order to maximize a function. This method allows to define custom cost functions, scipy.optimize does the rest. Il giorno lun 17 dic 2018 alle ore 13:04 edsf70 [email protected] ha scritto: … On tools.py which beta is calculated by get_optimal_temperature? Global or local beta? Would you please be so kind providing functions to calculate both? Thank you very much indeed! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#8>, or mute the thread https://github.com/notifications/unsubscribe-auth/ARSx93_I3sx9u52wwdXd33Ad3o_DHy7gks5u54hagaJpZM4ZWOOf .
That would be great!!! Thank you very much indeed!!
Would you please be so kind telling me if there is already an implementation for eq. S36 (local Beta)?
@edsf70 I've implemented it based on the eqs39 implementation and it turned out like this:
def eqs36(beta, s, A):
N = A.shape[0]
x = 0
for i in range(N):
for j in range(N):
if A[i, j] == 0:
continue
else:
pij = 1 / (1 + np.exp(-2 * beta * (s[i] - s[j])))
signed = np.sign(A[i, j] - (A[i, j] + A[j, i])*pij)
x += signed * ((A[i, j] - A[j, i])*(s[i] - s[j])) / ( np.cosh(2*beta*(s[i] - s[j])) + 1)
return x
Only thing is that it is not nopython=True as eqs39, but it didn't seem an issue.
