Can gen-AC be trained on conditional probabilities/density?

[Raymvp]

Dear Yuting, Thanks for the awesome paper and implementation. I've noticed that your experiments are based on probabilities and densities. I've thought of an interesting application that could utilize your method. However, I would need to use conditional probabilities/densities to train the model. This would require copula partial derivatives. Would you please kindly provide detailed guidance on how to use partial derivatives to train the gen-AC model? Thank you! Xin

[yutingng]

for a start, in main.py, forward, the mode 'cond_cdf' will compute the partial derivative.

---

From: Xin Liu @.> Sent: Friday, April 26, 2024 10:15 AM To: yutingng/gen-AC @.> Cc: Subscribed @.***> Subject: [yutingng/gen-AC] Can gen-AC be trained on conditional probabilities/density? (Issue #1)

Dear Yuting, Thanks for the awesome paper and implementation. I've noticed that your experiments are based on probabilities and densities. I've thought of an interesting application that could utilize your method. However, I would need to use conditional probabilities/densities to train the model. This would require copula partial derivatives. Would you please kindly provide detailed guidance on how to use partial derivatives to train the gen-AC model? Thank you! Xin

— Reply to this email directly, view it on GitHubhttps://urldefense.com/v3/__https://github.com/yutingng/gen-AC/issues/1__;!!OToaGQ!penecJIrutaq6MSR2aHJqM73QNYruOgKReCg140rjI0qkBbtGLWQEssrbqxel35U8J_ryulOkeRRSU60xlA-q4eM4A$, or unsubscribehttps://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/ADK3XOK35CRDFYCIM3JWYFTY7JOO5AVCNFSM6AAAAABG23W4N2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGI3DKOBYGI2TANQ__;!!OToaGQ!penecJIrutaq6MSR2aHJqM73QNYruOgKReCg140rjI0qkBbtGLWQEssrbqxel35U8J_ryulOkeRRSU60xlD63DwhFQ$. You are receiving this because you are subscribed to this thread.Message ID: @.***>

[Raymvp]

Thank you very much for reply. I have another question. In section 4.3, there is a sentence: "An alternative method may be to train the outer generator using a composite likelihood with bivariate margins since bivariate margins are Archimedean with generator given by the outer generator." Could you please elaborate on how to do this? I can train the inner generator using con_cdf, but I don't know how to train the outer generator.