Kiq'`SSKrSSKrSrSrSrSrSrSrSrSS \ "S 54S jr S r g) Nc~[R"U[RS9n[R"U[RS9nURS:XaUR S5nUR SnURS:XaUR S5nUR SS:XaUS:aUR US5nUSS2S4R S5nUSS2S4R S5n[R"SX- U- S--5U[R"S[R-5-- $) Gaussian likelihood for batched θ. x: Tensor of shape (N,) or (batch_size, N) theta: Tensor of shape (batch_size, 2) -> [mu, sigma] for each θ Returns: Tensor of shape (batch_size, N) -> likelihood of each x for each θ dtyperN@ torch as_tensorfloat32ndim unsqueezeshapeexpandexpmathsqrtpi)xtheta correlation batch_sizemusigmas M/global/cfs/cdirs/lsst/www/DESC_TD_PUBLIC/users/bastienc/kestrel/Functions.py DistGaussianrs /A OOE 7E zzQ"QJ vv{ KKNwwqzQ:> HHZ $ q!t  q !B !Q$K ! !! $E 99Tafe^a// 0EDIIcDGGm [mu0, sigma, m] correlation: Same shape as x Returns: Tensor of shape (batch_size, N) rrrrNr r r r )rrrrmu0rmrs rDistGaussian_EVOLr$&s /A OOE 7E//+U]]CK zzQ"QJ vv{ KKNwwqzQ:> HHZ $1!++A. q Z!^!((R8 1+   "C !Q$K ! !! $E ad a A ; B 99Taf-!33 4 #PTPWPW-@X8X YYr c[R"U[RS9n[R"U[RS9nURS:XaUR S5nUR SnURS:XaUR S5nUR SS:XaUS:aUR US5nUSS2S4R S5nSnSU- nSU- nXg- [R"Xg- *U-5-nU$)u Exponential likelihood for batched θ. x: Tensor of shape (N,) or (batch_size, N) theta: Returns: Tensor of shape (batch_size, N) -> likelihood of each x for each θ rrrrN??r rrrrrrr rrrrtau tau_proposal lambda_targetlambda_proposalweightss rDistExponentialr/Os /A OOE 7E zzQ"QJ vv{ KKNwwqzQ:> HHZ $ 1+   "CL#IML(O.%))m>]<^ab [mu1, sigma1, mu2, sigma2, a] for each θ Returns: Tensor of shape (batch_size, N) -> likelihood of each x for each θ rrrrNr r r r ) rrrrmu1sigma1mu2sigma2aG1G2s rDistDoubleGaussianr<s /A OOE 7E zzQ"QJ vv{ KKNwwqzQ:> HHZ $ 1+   "C 1a4[ " "1 %F 1+   "C 1a4[ " "1 %F ad a A 4AGV+a// 0FTYYsTWW}=U4U VB 599Tagv-11 2 2ftyytww?W6W XB 5Lr c[R"U[RURS9n[R"U[RS9n[R"U[RURS9nURS:XaUR S5nUR SnURS:XaUR S5nUR SS:XaUS:aURUS5nURS:XaUR S5nUR SS:XaUS:aURUS5nUSS2S4R S5nUSS2S4R S5nUSS2S4R S5nUS[R"XR-S- 5-- S-nX- U- n[R"S US--5U[R"S [R-5-- n U $) z Logistic-shaped likelihood for importance sampling. x: (N,) or (B, N) theta: (B, 3) -> [L, k, sigma] correlation: (N,) or (B, N) Returns: (B, N) likelihood weights )rdevicerrrrNr r r ) r rrr>rrrrrrrr) rrrBLkrmeanzpdfs r DistLogisticrFs u||DA OOE 7E//+U]]5<<XK zzQ" AAvv{ KKNwwqzQ1q5 HHQO1!++A. q QU!((B/  !Q$K ! !! $A !Q$K ! !! $A !Q$K ! !! $E EIIao233 4q 8D UA ))D1a4K EDIIcDGGm,D$D EC Jr c8[R"U[RS9n[R"U[RS9nURS:XaUR S5nUR SnURS:XaUR S5nUR SS:XaUS:aUR US5nUSS2S4R S5nUSS2S4R S5nSU- nXS- -[R"U*U-5-nU$)u Gamma likelihood for batched θ. x: Tensor of shape (N,) or (batch_size, N) theta: Returns: Tensor of shape (batch_size, N) -> likelihood of each x for each θ rrrrNr'r()rrrrr*gammar,r.s r DistGammarIs /A OOE 7E zzQ"QJ vv{ KKNwwqzQ:> HHZ $ 1+   "C !A#J  #E#IM!)nuyy=)9A)=>>G Nr g333333?infcd[R"U[RS9n[R"U[RS9nURS:XaUR S5nUR SnURS:XaUR S5nUR SS:XaUS:aUR US5nUSS2S4R S5nUSS2S4R S5n[R"SX- U- S--5U[R"S[R-5-- nS n X6- U- n XF- U- n U "U 5U "U 5- n [R"U S S 9n X:X:*-n [R"U5nXU RU5U - X'U$) u Truncated Gaussian likelihood for batched θ. x: Tensor (N,) or (batch_size, N) theta: Tensor (batch_size, 2) -> [mu, sigma] low, high: truncation bounds (can be scalars or tensors) Returns: Tensor (batch_size, N) rrrrNr r r chSS[R"U[R"S5- 5--$)Nr&rr )r erfrr)rDs r normal_cdf)DistTruncatedGaussian..normal_cdf6s(a%))A ! $45566r g-q=)min)r rrrrrrrrrrclamp zeros_like expand_as)rrrlowhighrrrrErNr9bZmask truncated_pdfs rDistTruncatedGaussianrZs /A OOE 7E zzQ"QJ vv{ KKNwwqzQ:> HHZ $ q!t  q !B !Q$K ! !! $E ))DQVUNQ.. /5499S477];S3S TC7 UA eA1 1 %A  A5!A H #D$$S)M)akk#&6t&<r_sN V<'ZR$L&R$L*X"J15#E%L2h r