still working on TNN

4 files changed, 118 insertions, 90 deletions

diff --git a/nerv/examples/lmptb/lmptb/lmseqreader.lua b/nerv/examples/lmptb/lmptb/lmseqreader.lua
index 307c5a3..006b5cb 100644
--- a/nerv/examples/lmptb/lmptb/lmseqreader.lua
+++ b/nerv/examples/lmptb/lmptb/lmseqreader.lua
@@ -1,5 +1,5 @@
 require 'lmptb.lmvocab'
-require 'rnn.layer_tdag'
+require 'rnn.tnn'
 
 local LMReader = nerv.class("nerv.LMSeqReader")
 
@@ -66,7 +66,7 @@ function LMReader:refresh_stream(id)
 end
 
 --feeds: a table that will be filled by the reader
---Returns: inputs_m, labels_m
+--Returns: bool
 function LMReader:get_batch(feeds)
     if (feeds == nil or type(feeds) ~= "table") then
         nerv.error("feeds is not a table")
@@ -74,36 +74,49 @@ function LMReader:get_batch(feeds)
 
     feeds["inputs_s"] = {}
     feeds["labels_s"] = {}
-    inputs_s = feeds.inputs_s
-    labels_s = feeds.labels_s
+    local inputs_s = feeds.inputs_s
+    local labels_s = feeds.labels_s
     for i = 1, self.chunk_size, 1 do
         inputs_s[i] = {}
         labels_s[i] = {}
     end
 
+    local inputs_m = feeds.inputs_m --port 1 : word_id, port 2 : label
+    local flags = feeds.flags_now
+
     local got_new = false
     for i = 1, self.batch_size, 1 do
         local st = self.streams[i]
         for j = 1, self.chunk_size, 1 do
+            flags[j][i] = 0
             self:refresh_stream(i)
             if (st.store[st.head] ~= nil) then
                 inputs_s[j][i] = st.store[st.head]
+                inputs_m[j][1][i - 1][0] = self.vocab:get_word_str(st.store[st.head]).id - 1
             else
                 inputs_s[j][i] = self.vocab.null_token
+                inputs_m[j][1][i - 1][0] = 0
             end
+            inputs_m[j][2][i - 1]:fill(0)
             if (st.store[st.head + 1] ~= nil) then
                 labels_s[j][i] = st.store[st.head + 1]
+                inputs_m[j][2][i - 1][self.vocab:get_word_str(st.store[st.head + 1]).id - 1] = 1
             else
                 labels_s[j][i] = self.vocab.null_token
             end
             if (inputs_s[j][i] ~= self.vocab.null_token) then
+                flags[j][i] = bit.bor(flags[j][i], nerv.TNN.FC.SEQ_NORM)
                 got_new = true
                 st.store[st.head] = nil
                 st.head = st.head + 1
                 if (labels_s[j][i] == self.vocab.sen_end_token) then
+                    flags[j][i] = bit.bor(flags[j][i], nerv.TNN.FC.SEQ_END)
                     st.store[st.head] = nil --sentence end is passed
                     st.head = st.head + 1
                 end
+                if (inputs_s[j][i] == self.vocab.send_end_token) then
+                    flags[j][i] = bit.bor(flags[j][i], nerv.TNN.FC.SEQ_START)
+                end
             end 
         end
     end
diff --git a/nerv/examples/lmptb/m-tests/dagl_test.lua b/nerv/examples/lmptb/m-tests/dagl_test.lua
index 5e90551..a50107d 100644
--- a/nerv/examples/lmptb/m-tests/dagl_test.lua
+++ b/nerv/examples/lmptb/m-tests/dagl_test.lua
@@ -98,14 +98,11 @@ end
 
 --global_conf: table
 --layerRepo: nerv.LayerRepo
---Returns: a nerv.TDAGLayer
+--Returns: a nerv.TNN
 function prepare_dagLayer(global_conf, layerRepo)
-    printf("%s Initing daglayer ...\n", global_conf.sche_log_pre)
+    printf("%s Initing TNN ...\n", global_conf.sche_log_pre)
 
     --input: input_w, input_w, ... input_w_now, last_activation
-    local dim_in_t = {}
-    dim_in_t[1] = 1 --input to select_linear layer
-    dim_in_t[2] = global_conf.vocab:size() --input to softmax label
     local connections_t = {
         {"<input>[1]", "selectL1[1]", 0},
         {"selectL1[1]", "recurrentL1[1]", 0},  
@@ -124,11 +121,11 @@ function prepare_dagLayer(global_conf, layerRepo)
     end
     ]]--
 
-    local dagL = nerv.TDAGLayer("dagL", global_conf, {["dim_in"] = dim_in_t, ["dim_out"] = {1}, ["sub_layers"] = layerRepo,
+    local tnn = nerv.TNN("TNN", global_conf, {["dim_in"] = {1, global_conf.vocab:size()}, ["dim_out"] = {1}, ["sub_layers"] = layerRepo,
             ["connections"] = connections_t, 
         })
-    printf("%s Initing DAGLayer end.\n", global_conf.sche_log_pre)
-    return dagL
+    printf("%s Initing TNN end.\n", global_conf.sche_log_pre)
+    return tnn
 end
 
 train_fn = '/home/slhome/txh18/workspace/nerv/nerv/nerv/examples/lmptb/m-tests/some-text'
@@ -160,7 +157,6 @@ global_conf["vocab"] = vocab
 global_conf.vocab:build_file(global_conf.train_fn, false)
 local paramRepo = prepare_parameters(global_conf, true)
 local layerRepo = prepare_layers(global_conf, paramRepo)
-local dagL = prepare_dagLayer(global_conf, layerRepo)
-dagL:init(global_conf.batch_size, global_conf.chunk_size)
-
+local tnn = prepare_dagLayer(global_conf, layerRepo)
+tnn:init(global_conf.batch_size, global_conf.chunk_size)
 
diff --git a/nerv/examples/lmptb/m-tests/lmseqreader_test.lua b/nerv/examples/lmptb/m-tests/lmseqreader_test.lua
index 504698f..cbcdcbe 100644
--- a/nerv/examples/lmptb/m-tests/lmseqreader_test.lua
+++ b/nerv/examples/lmptb/m-tests/lmseqreader_test.lua
@@ -1,4 +1,5 @@
 require 'lmptb.lmseqreader'
+require 'lmptb.lmutil'
 
 local printf = nerv.printf
 
@@ -8,15 +9,36 @@ local vocab = nerv.LMVocab()
 vocab:build_file(test_fn)
 local chunk_size = 5
 local batch_size = 3
-local reader = nerv.LMSeqReader({}, batch_size, chunk_size, vocab)
+local global_conf = {
+    lrate = 1, wcost = 1e-6, momentum = 0,
+    cumat_type = nerv.CuMatrixFloat,
+    mmat_type = nerv.CuMatrixFloat,
+
+    hidden_size = 20,
+    chunk_size = chunk_size,
+    batch_size = batch_size, 
+    max_iter = 18,
+    param_random = function() return (math.random() / 5 - 0.1) end,
+    independent = true,
+
+    train_fn = train_fn,
+    test_fn = test_fn,
+    sche_log_pre = "[SCHEDULER]:",
+    log_w_num = 10, --give a message when log_w_num words have been processed
+    timer = nerv.Timer(),
+    
+    vocab = vocab
+}
+
+local reader = nerv.LMSeqReader(global_conf, batch_size, chunk_size, vocab)
 reader:open_file(test_fn)
-local input = {}
-local label = {}
-for i = 1, batch_size, 1 do
-    input[i] = {}
-    label[i] = {}
-end
 local feeds = {}
+feeds.flags_now = {}
+feeds.inputs_m = {}
+for j = 1, chunk_size do
+    feeds.inputs_m[j] = {global_conf.cumat_type(batch_size, 1), global_conf.cumat_type(batch_size, global_conf.vocab:size())}
+    feeds.flags_now[j] = {}
+end
 while (1) do
     local r = reader:get_batch(feeds)
     if (r == false) then break end
diff --git a/nerv/examples/lmptb/rnn/layer_tdag.lua b/nerv/examples/lmptb/rnn/tnn.lua
index 6e5d774..3f192b5 100644
--- a/nerv/examples/lmptb/rnn/layer_tdag.lua
+++ b/nerv/examples/lmptb/rnn/tnn.lua
@@ -1,4 +1,5 @@
-local DAGLayer = nerv.class("nerv.TDAGLayer", "nerv.Layer")
+local TNN = nerv.class("nerv.TNN", "nerv.Layer")
+local DAGLayer = TNN
 
 local function parse_id(str)
     --used to parse layerid[portid],time
@@ -45,13 +46,31 @@ local function discover(id, layers, layer_repo)
     return ref
 end
 
-function DAGLayer.makeInitialStore(dim, batch_size, chunk_size, global_conf)
+nerv.TNN.FC = {} --flag const
+nerv.TNN.FC.SEQ_START = 4
+nerv.TNN.FC.SEQ_END = 8 
+nerv.TNN.FC.HAS_INPUT = 1
+nerv.TNN.FC.HAS_LABEL = 2
+nerv.TNN.FC.SEQ_NORM = bit.bor(nerv.TNN.FC.HAS_INPUT, nerv.TNN.FC.HAS_LABEL) --This instance have both input and label
+
+function DAGLayer.makeInitialStore(st, p, dim, batch_size, chunk_size, global_conf, st_c, p_c)
     --Return a table of matrix storage from time (1-chunk_size)..(2*chunk_size)
-    st = {}
+    if (type(st) ~= "table") then
+        nerv.error("st should be a table")
+    end
     for i = 1 - chunk_size, chunk_size * 2 do
-        st[i] = global_conf.cumat_type(batch_size, dim)
+        if (st[i] == nil) then
+            st[i] = {}
+        end
+        st[i][p] = global_conf.cumat_type(batch_size, dim)
+        st[i][p]:fill(0)
+        if (st_c ~= nil) then
+            if (st_c[i] == nil) then
+                st_c[i] = {}
+            end
+            st_c[i][p_c] = st[i][p]
+        end
     end
-    return st
 end
 
 function DAGLayer:__init(id, global_conf, layer_conf)
@@ -111,7 +130,7 @@ function DAGLayer:__init(id, global_conf, layer_conf)
 end
 
 function DAGLayer:init(batch_size, chunk_size)
-    for i, conn in ipairs(self.parsed_conns) do 
+    for i, conn in ipairs(self.parsed_conns) do --init storage for connections inside the NN
         local _, output_dim
         local ref_from, port_from, ref_to, port_to
         ref_from, port_from = conn.src.ref, conn.src.port
@@ -121,41 +140,53 @@ function DAGLayer:init(batch_size, chunk_size)
             nerv.error("layer %s has a zero dim port", ref_from.layer.id)
         end
 
-        local mid = DAGLayer.makeInitialStore(dim, batch_size, chunk_size, global_conf)
-        local err_mid = DAGLayer.makeInitialStore(dim, batch_size, chunk_size, global_conf)
+        print("TNN initing storage", ref_from.layer.id, "->", ref_to.layer.id)
+        self.makeInitialStore(ref_from.outputs_m, port_from, dim, batch_size, chunk_size, global_conf, ref_to.inputs_m, port_to)
+        self.makeInitialStore(ref_from.err_inputs_m, port_from, dim, batch_size, chunk_size, global_conf, ref_to.err_outputs_m, port_to)
         
-        print(ref_from.layer.id, "->", ref_to.layer.id)
+    end
 
-        ref_from.outputs_m[port_from] = mid
-        ref_to.inputs_m[port_to] = mid
+    self.outputs_m = {}
+    self.err_inputs_m = {}
+    for i = 1, #self.dim_out do --Init storage for output ports 
+        local ref = self.outputs_p[i].ref
+        local p = self.outputs_p[i].port
+        self.makeInitialStore(ref.outputs_m, p, self.dim_out[i], batch_size, chunk_size, self.gconf, self.outputs_m, i)
+        self.makeInitialStore(ref.err_inputs_m, p, self.dim_out[i], batch_size, chunk_size, self.gconf, self.err_inputs_m, i)
+    end 
 
-        ref_from.err_inputs_m[port_from] = err_mid
-        ref_to.err_outputs_m[port_to] = err_mid
-    end
-    for id, ref in pairs(self.layers) do
+    self.inputs_m = {}
+    self.err_outputs_m = {}
+    for i = 1, #self.dim_in do --Init storage for input ports 
+        local ref = self.inputs_p[i].ref
+        local p = self.inputs_p[i].port
+        self.makeInitialStore(ref.inputs_m, p, self.dim_in[i], batch_size, chunk_size, self.gconf, self.inputs_m, i)
+        self.makeInitialStore(ref.err_outputs_m, p, self.dim_in[i], batch_size, chunk_size, self.gconf, self.err_outputs_m, i)
+    end 
+
+    for id, ref in pairs(self.layers) do --Calling init for child layers
         for i = 1, #ref.dim_in do
-            if ref.inputs_m[i] == nil then
+            if (ref.inputs_m[i] == nil or ref.err_outputs_m[i] == nil) then
                 nerv.error("dangling input port %d of layer %s", i, id)
             end
         end
         for i = 1, #ref.dim_out do
-            if ref.outputs_m[i] == nil then
+            if (ref.outputs_m[i] == nil or ref.err_inputs_m[i] == nil) then
                 nerv.error("dangling output port %d of layer %s", i, id)
             end
         end
         -- initialize sub layers
         ref.layer:init(batch_size)
     end
-    for i = 1, #self.dim_in do
-        if self.inputs_p[i] == nil then
-            nerv.error("<input> port %d not attached", i)
-        end
-    end
-    for i = 1, #self.dim_out do
-        if self.outputs_p[i] == nil then
-            nerv.error("<output> port %d not attached", i)
-        end
+
+    local flags_now = {}
+    for i = 1, chunk_size do
+        flags_now[i] = {}
     end
+
+    self.feeds_now = {} --feeds is for the reader to fill
+    self.feeds_now.inputs_m = self.inputs_m
+    self.feeds_now.flags_now = flags_now
 end
 
 --[[
@@ -187,48 +218,13 @@ function DAGLayer:batch_resize(batch_size)
 end
 ]]--
 
-function DAGLayer:set_inputs(inputs_m)
-    for i = 1, #self.dim_in do
-        if inputs_m[i] == nil then
-            nerv.error("inputs_m[%d] is not provided", i);
-        end
-        local ref = self.inputs_p[i].ref
-        local p = self.inputs_p[i].port
-        ref.inputs_m[p] = inputs_m[i] 
-    end
-end
-
-function DAGLayer:set_outputs(outputs_m)
-    for i = 1, #self.dim_out do
-        if outputs_m[i] == nil then
-            nerv.error("outputs_m[%d] is not provided", i);
-        end
-        local ref = self.outputs_p[i].ref
-        local p = self.outputs_p[i].port
-        ref.outputs_m[p] = outputs_m[i]
-    end
-end
-
-function DAGLayer:set_err_inputs(bp_errs_m)
-    for i = 1, #self.dim_out do
-        if bp_errs_m[i] == nil then
-            nerv.error("bp_errs_m[%d] is not provided", i);
-        end
-        local ref = self.outputs_p[i].ref
-        local p = self.outputs_p[i].port
-        ref.err_inputs_m[p] = bp_errs_m[i]
-    end
-end
-
-function DAGLayer:set_err_outputs(next_bp_err)
-    for i = 1, #self.dim_in do
-        if (next_bp_err[i] == nil) then
-            nerv.error("next_bp_err[%d] is not provided", i)
-        end
-        local ref = self.inputs_p[i].ref
-        local p = self.inputs_p[i].port
-        ref.err_outputs_m[p] = next_bp_err[i]
-    end
+--reader: some reader
+--Returns: bool, whether has new feed
+--Returns: feeds, a table that will be filled with the reader's feeds
+function DAGLayer:getFeedFromReader(reader)
+    local feeds = self.feeds_now
+    local got_new = reader:get_batch(feeds)
+    return got_new, feeds
 end
 
 function DAGLayer:update(bp_err, input, output)
@@ -266,6 +262,7 @@ function DAGLayer:back_propagate(bp_err, next_bp_err, input, output)
     end
 end
 
+--Return: nerv.ParamRepo
 function DAGLayer:get_params()
     local param_repos = {}
     for id, ref in pairs(self.queue) do