diff options
Diffstat (limited to 'layer/affine.lua')
| -rw-r--r-- | layer/affine.lua | 31 |
1 files changed, 22 insertions, 9 deletions
diff --git a/layer/affine.lua b/layer/affine.lua index 573b98d..90a1d16 100644 --- a/layer/affine.lua +++ b/layer/affine.lua @@ -12,14 +12,27 @@ function MatrixParam:write(pfhandle) self.trans:new_to_host():save(pfhandle) end -function AffineLayer:__init(id, global_conf, ltp, bp) +function AffineLayer:__init(id, global_conf, layer_conf) self.id = id - self.ltp = ltp - self.bp = bp + self.ltp = layer_conf.ltp + self.bp = layer_conf.bp + self.dim_in = layer_conf.dim_in + self.dim_out = layer_conf.dim_out self.gconf = global_conf + self:check_dim_len(1, 1) -- exactly one input and one output end function AffineLayer:init() + if self.ltp.trans:ncol() ~= self.bp.trans:ncol() then + nerv.error("mismatching dimensions of linear transform and bias paramter") + end + if self.dim_in[1] ~= self.ltp.trans:nrow() then + nerv.error("mismatching dimensions of linear transform parameter and input") + end + if self.dim_out[1] ~= self.ltp.trans:ncol() then + nerv.error("mismatching dimensions of linear transform parameter and output") + end + -- linear transform correction self.ltc = self.ltp.trans:create() self.ltc:fill(0) @@ -36,10 +49,10 @@ function nerv.AffineLayer:update(bp_err, input, output) local gconf = self.gconf -- momentum gain local mmt_gain = 1.0 / (1.0 - gconf.momentum); - local n = input[0]:nrow() * mmt_gain + local n = input[1]:nrow() * mmt_gain -- update corrections (accumulated errors) - ltc:mul(input[0], bp_err[0], 1.0, gconf.momentum, 'T', 'N') - bc:add(bc, bp_err[0]:colsum(), gconf.momentum, 1.0) + ltc:mul(input[1], bp_err[1], 1.0, gconf.momentum, 'T', 'N') + bc:add(bc, bp_err[1]:colsum(), gconf.momentum, 1.0) -- perform update ltp:add(ltp, ltc, 1.0, -gconf.lrate / n) bp:add(bp, bc, 1.0, -gconf.lrate / n) @@ -49,11 +62,11 @@ end function nerv.AffineLayer:propagate(input, output) -- apply linear transform - output[0]:mul(input[0], self.ltp.trans, 1.0, 0.0, 'N', 'N') + output[1]:mul(input[1], self.ltp.trans, 1.0, 0.0, 'N', 'N') -- add bias - output[0]:add_row(self.bp.trans, 1.0) + output[1]:add_row(self.bp.trans, 1.0) end function nerv.AffineLayer:back_propagate(next_bp_err, bp_err, input, output) - next_bp_err[0]:mul(bp_err[0], self.ltp.trans, 1.0, 0.0, 'N', 'T') + next_bp_err[1]:mul(bp_err[1], self.ltp.trans, 1.0, 0.0, 'N', 'T') end |