From a32195e3e2ae9ca0f0c7a82e73e6bddb64568c05 Mon Sep 17 00:00:00 2001
From: Determinant <ted.sybil@gmail.com>
Date: Thu, 10 Mar 2016 13:40:11 +0800
Subject: major change: clearer param binding semantics; permit rebinding;
 enable resuming from previous training

---
 nerv/layer/affine.lua | 43 +++++++++++++++++++++++--------------------
 1 file changed, 23 insertions(+), 20 deletions(-)

(limited to 'nerv/layer/affine.lua')

diff --git a/nerv/layer/affine.lua b/nerv/layer/affine.lua
index 4156dde..38743aa 100644
--- a/nerv/layer/affine.lua
+++ b/nerv/layer/affine.lua
@@ -8,21 +8,19 @@ local AffineLayer = nerv.class('nerv.AffineLayer', 'nerv.Layer')
 --- A parameter that consists of a single matrix
 -- @type nerv.MatrixParam
 
+function MatrixParam:check(checker)
+    -- check trans matrix type
+    checker(self.trans)
+end
+
 --- Read from a file handle.
 -- @param handle the file handle
 function MatrixParam:read(handle)
     self.trans = self.gconf.mmat_type.load(handle)
-    if not self.gconf.use_cpu then
-        self.trans = self.gconf.cumat_type.new_from_host(self.trans)
-    end
 end
 
 function MatrixParam:write(handle)
-    local trans = self.trans
-    if not self.gconf.use_cpu then
-        trans = self.trans:new_to_host()
-    end
-    trans:save(handle)
+    self.trans:save(handle)
 end
 
 function MatrixParam:train_init()
@@ -30,6 +28,12 @@ function MatrixParam:train_init()
     self.correction:fill(0)
 end
 
+function MatrixParam:copy(copier)
+    local target = nerv.MatrixParam(self.id, self.gconf)
+    target.trans = copier(self.trans)
+    return target
+end
+
 function MatrixParam:_update_by_gradient(gradient, alpha, beta)
     local gconf = self.gconf
     -- momentum gain
@@ -77,25 +81,24 @@ end
 
 --- The constructor.
 function AffineLayer:__init(id, global_conf, layer_conf)
-    self.id = id
-    self.dim_in = layer_conf.dim_in
-    self.dim_out = layer_conf.dim_out
-    if layer_conf.ltp ~= nil and layer_conf.ltp1 == nil then
-        layer_conf.ltp1 = layer_conf.ltp
-    end
+    nerv.Layer.__init(self, id, global_conf, layer_conf)
+    self:check_dim_len(-1, 1) -- exactly one output, allow multiple inputs
+    self:bind_params()
+end
+
+function AffineLayer:bind_params()
     for i = 1, #self.dim_in do
         local pid = "ltp" .. i
         local pid_list = i == 1 and {pid, "ltp"} or pid
-        self["ltp" .. i] = self:find_param(pid_list, layer_conf, global_conf,
+        self["ltp" .. i] = self:find_param(pid_list, self.lconf, self.gconf,
                                             nerv.LinearTransParam,
-                                            {self.dim_in[i], self.dim_out[1]}) 
+                                            {self.dim_in[i], self.dim_out[1]})
     end
     self.ltp = self.ltp1 -- alias of ltp1
-    self.bp = self:find_param("bp", layer_conf, global_conf,
+    self.bp = self:find_param("bp", self.lconf, self.gconf,
                                 nerv.BiasParam,
                                 {1, self.dim_out[1]})
-    self.gconf = global_conf
-    self:check_dim_len(-1, 1) -- exactly one output, allow multiple inputs
+
 end
 
 function AffineLayer:init(batch_size)
@@ -142,7 +145,7 @@ function AffineLayer:back_propagate(bp_err, next_bp_err, input, output)
 end
 
 function AffineLayer:get_params()
-    local pr = nerv.ParamRepo({self.ltp1, self.bp})
+    local pr = nerv.ParamRepo({self.ltp1, self.bp}, self.loc_type)
     for i = 2, #self.dim_in do
         pr:add(self["ltp" .. i].id, self["ltp" .. i])
     end
-- 
cgit v1.2.3