From ade38c607a8d0dc8dc1d701084ed88df2fa89df9 Mon Sep 17 00:00:00 2001
From: Tom Smeding <tom@tomsmeding.com>
Date: Tue, 26 Nov 2024 23:05:30 +0100
Subject: Working argument accum mode (...)
The derivative of 'neural' in full accum mode is pretty atrocious now; I
think this is because when you have code like this:
\(a :: Arr 1 R) ->
let b = a
in let c = b
in sum d
then because the argument, as well as both let bindings, bind a value of
array type, each will introduce an accumulator, hence resulting in three
(!) nested `with` clauses that each just contribute their result back to
their parent. This is pointless, and we should fix this.
---
test/Main.hs | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
(limited to 'test/Main.hs')
diff --git a/test/Main.hs b/test/Main.hs
index d18884e..b6f9f2b 100644
--- a/test/Main.hs
+++ b/test/Main.hs
@@ -62,11 +62,11 @@ gradientByCHAD' = \simplIters env term input -> second (second (toTanE env input
toTan typ primal der = case typ of
STNil -> der
STPair t1 t2 -> case der of
- Left () -> bimap (zeroTan t1) (zeroTan t2) primal
- Right (d₁, d₂) -> bimap (\p1 -> toTan t1 p1 d₁) (\p2 -> toTan t2 p2 d₂) primal
+ Nothing -> bimap (zeroTan t1) (zeroTan t2) primal
+ Just (d₁, d₂) -> bimap (\p1 -> toTan t1 p1 d₁) (\p2 -> toTan t2 p2 d₂) primal
STEither t1 t2 -> case der of
- Left () -> bimap (zeroTan t1) (zeroTan t2) primal
- Right d -> case (primal, d) of
+ Nothing -> bimap (zeroTan t1) (zeroTan t2) primal
+ Just d -> case (primal, d) of
(Left p, Left d') -> Left (toTan t1 p d')
(Right p, Right d') -> Right (toTan t2 p d')
_ -> error "Primal and cotangent disagree on Either alternative"
--
cgit v1.2.3-70-g09d2