aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--nerv/layer/affine.lua30
-rw-r--r--nerv/layer/affine_recurrent.lua14
2 files changed, 22 insertions, 22 deletions
diff --git a/nerv/layer/affine.lua b/nerv/layer/affine.lua
index c24af16..c5084c4 100644
--- a/nerv/layer/affine.lua
+++ b/nerv/layer/affine.lua
@@ -19,29 +19,29 @@ end
function MatrixParam:update(gradient)
local gconf = self.gconf
- if (gconf.momentum > 0) then
+ if gconf.momentum > 0 then
self.correction:add(self.correction, gradient, gconf.momentum, 1.0)
-- momentum gain
- local mmt_gain = 1.0 / (1.0 - gconf.momentum);
+ local mmt_gain = 1.0 / (1.0 - gconf.momentum)
local n = self.gconf.batch_size * mmt_gain
-- perform update
- self.trans:add(self.trans, self.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/n)
+ self.trans:add(self.trans, self.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
else
- self.trans:add(self.trans, gradient, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/gconf.batch_size)
+ self.trans:add(self.trans, gradient, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / gconf.batch_size)
end
end
function MatrixParam:updateEI(err, input)
local gconf = self.gconf
- if (gconf.momentum > 0) then
+ if gconf.momentum > 0 then
self.correction:mul(input, err, 1.0, gconf.momentum, 'T', 'N')
-- momentum gain
- local mmt_gain = 1.0 / (1.0 - gconf.momentum);
+ local mmt_gain = 1.0 / (1.0 - gconf.momentum)
local n = self.gconf.batch_size * mmt_gain
-- perform update
- self.trans:add(self.trans, self.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/n)
+ self.trans:add(self.trans, self.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
else
- self.trans:mul(input, err, -gconf.lrate/gconf.batch_size, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, 'T', 'N')
+ self.trans:mul(input, err, - gconf.lrate / gconf.batch_size, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, 'T', 'N')
end
end
@@ -92,20 +92,20 @@ function AffineLayer:batch_resize(batch_size)
end
function AffineLayer:update(bp_err, input, output)
- if (self.direct_update == true) then
+ if self.direct_update == true then
local gconf = self.gconf
- if (gconf.momentum > 0) then
+ if gconf.momentum > 0 then
self.ltp.correction:mul(input[1], bp_err[1], 1.0, gconf.momentum, 'T', 'N')
self.bp.correction:add(self.bp.correction, bp_err[1]:colsum(), gconf.momentum, 1)
-- momentum gain
- local mmt_gain = 1.0 / (1.0 - gconf.momentum);
+ local mmt_gain = 1.0 / (1.0 - gconf.momentum)
local n = self.gconf.batch_size * mmt_gain
-- perform update
- self.ltp.trans:add(self.ltp.trans, self.ltp.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate / n)
- self.bp.trans:add(self.bp.trans, self.bp.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate / n)
+ self.ltp.trans:add(self.ltp.trans, self.ltp.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
+ self.bp.trans:add(self.bp.trans, self.bp.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
else
- self.ltp.trans:mul(input[1], bp_err[1], -gconf.lrate / gconf.batch_size, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, 'T', 'N')
- self.bp.trans:add(self.bp.trans, bp_err[1]:colsum(), 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate / gconf.batch_size)
+ self.ltp.trans:mul(input[1], bp_err[1], - gconf.lrate / gconf.batch_size, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, 'T', 'N')
+ self.bp.trans:add(self.bp.trans, bp_err[1]:colsum(), 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / gconf.batch_size)
end
else
self.ltp:updateEI(bp_err[1], input[1])
diff --git a/nerv/layer/affine_recurrent.lua b/nerv/layer/affine_recurrent.lua
index b465e95..7f9c3f5 100644
--- a/nerv/layer/affine_recurrent.lua
+++ b/nerv/layer/affine_recurrent.lua
@@ -42,23 +42,23 @@ function Recurrent:batch_resize(batch_size)
end
function Recurrent:update(bp_err, input, output)
- if (self.direct_update == true) then
+ if self.direct_update == true then
local ltp_hh = self.ltp_hh.trans
local bp = self.bp.trans
local gconf = self.gconf
if (gconf.momentum > 0) then
-- momentum gain
- local mmt_gain = 1.0 / (1.0 - gconf.momentum);
+ local mmt_gain = 1.0 / (1.0 - gconf.momentum)
local n = input[1]:nrow() * mmt_gain
-- update corrections (accumulated errors)
self.ltp_hh.correction:mul(input[2], bp_err[1], 1.0, gconf.momentum, 'T', 'N')
self.bp.correction:add(self.bp.correction, bp_err[1]:colsum(), gconf.momentum, 1.0)
-- perform update and weight decay
- ltp_hh:add(ltp_hh, self.ltp_hh.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/n)
- bp:add(bp, self.bp.correction, 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/n)
+ ltp_hh:add(ltp_hh, self.ltp_hh.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
+ bp:add(bp, self.bp.correction, 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / n)
else
- ltp_hh:mul(input[2], bp_err[1], -gconf.lrate/gconf.batch_size, 1.0-gconf.wcost*gconf.lrate/gconf.batch_size, 'T', 'N')
- bp:add(bp, bp_err[1]:colsum(), 1.0-gconf.lrate*gconf.wcost/gconf.batch_size, -gconf.lrate/gconf.batch_size)
+ ltp_hh:mul(input[2], bp_err[1], - gconf.lrate / gconf.batch_size, 1.0 - gconf.wcost * gconf.lrate / gconf.batch_size, 'T', 'N')
+ bp:add(bp, bp_err[1]:colsum(), 1.0 - gconf.lrate * gconf.wcost / gconf.batch_size, - gconf.lrate / gconf.batch_size)
end
else
--self.ltp_hh_grad:mul(input[2], bp_err[1], 1.0, 0.0, 'T', 'N')
@@ -85,7 +85,7 @@ function Recurrent:back_propagate(bp_err, next_bp_err, input, output)
end
]]--
if (self.clip ~= nil) then
- next_bp_err[2]:clip(-self.clip, self.clip)
+ next_bp_err[2]:clip(- self.clip, self.clip)
end
end