aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--nerv/Makefile2
-rw-r--r--nerv/layer/duplicate.lua4
-rw-r--r--nerv/layer/identity.lua2
-rw-r--r--nerv/layer/init.lua1
-rw-r--r--nerv/main.lua36
-rw-r--r--nerv/nn/network.lua32
6 files changed, 52 insertions, 25 deletions
diff --git a/nerv/Makefile b/nerv/Makefile
index 0a2aa86..a9b4baf 100644
--- a/nerv/Makefile
+++ b/nerv/Makefile
@@ -34,7 +34,7 @@ LUA_LIBS := matrix/init.lua io/init.lua init.lua \
layer/init.lua layer/affine.lua layer/sigmoid.lua layer/tanh.lua layer/softmax_ce.lua layer/softmax.lua \
layer/window.lua layer/bias.lua layer/combiner.lua layer/mse.lua \
layer/elem_mul.lua layer/lstm.lua layer/lstm_gate.lua layer/dropout.lua layer/gru.lua \
- layer/graph.lua layer/rnn.lua layer/duplicate.lua\
+ layer/graph.lua layer/rnn.lua layer/duplicate.lua layer/identity.lua \
nn/init.lua nn/layer_repo.lua nn/param_repo.lua nn/layer_dag.lua nn/network.lua \
io/sgd_buffer.lua \
tnn/init.lua tnn/sutil.lua tnn/tnn.lua
diff --git a/nerv/layer/duplicate.lua b/nerv/layer/duplicate.lua
index fbd4a9e..1a93b26 100644
--- a/nerv/layer/duplicate.lua
+++ b/nerv/layer/duplicate.lua
@@ -30,9 +30,9 @@ function DuplicateLayer:propagate(input, output)
end
function DuplicateLayer:back_propagate(bp_err, next_bp_err)
- next_bp_err:copy_from(bp_err[1])
+ next_bp_err[1]:copy_from(bp_err[1])
for i = 2, #self.dim_out do
- next_bp_err:add(next_bp_err, bp_err[i], 1.0, 1.0)
+ next_bp_err[1]:add(next_bp_err[1], bp_err[i], 1.0, 1.0)
end
end
diff --git a/nerv/layer/identity.lua b/nerv/layer/identity.lua
index dc796fb..aeeff89 100644
--- a/nerv/layer/identity.lua
+++ b/nerv/layer/identity.lua
@@ -22,7 +22,7 @@ function IdentityLayer:propagate(input, output)
end
function IdentityLayer:back_propagate(bp_err, next_bp_err)
- next_bp_err[1]:copy_from(bp_err)
+ next_bp_err[1]:copy_from(bp_err[1])
end
function IdentityLayer:update()
diff --git a/nerv/layer/init.lua b/nerv/layer/init.lua
index 6f26d4d..39f97b1 100644
--- a/nerv/layer/init.lua
+++ b/nerv/layer/init.lua
@@ -126,6 +126,7 @@ nerv.include('dropout.lua')
nerv.include('gru.lua')
nerv.include('rnn.lua')
nerv.include('duplicate.lua')
+nerv.include('identity.lua')
-- The following lines are for backward compatibility, and will be removed in
-- the future. The use of these names are deprecated.
diff --git a/nerv/main.lua b/nerv/main.lua
index 5cb7d07..865aba0 100644
--- a/nerv/main.lua
+++ b/nerv/main.lua
@@ -1,8 +1,10 @@
-print 'Hello'
-
local global_conf = {
cumat_type = nerv.CuMatrixFloat,
param_random = function() return 0 end,
+ lrate = 0.1,
+ wcost = 0,
+ momentum = 0.9,
+ batch_size = 2,
}
local layer_repo = nerv.LayerRepo(
@@ -11,13 +13,13 @@ local layer_repo = nerv.LayerRepo(
rnn = {dim_in = {23}, dim_out = {26}},
},
['nerv.AffineLayer'] = {
- input = {dim_in = {20}, dim_out = {23}},
+ input = {dim_in = {62}, dim_out = {23}},
output = {dim_in = {26, 79}, dim_out = {79}},
},
['nerv.SigmoidLayer'] = {
sigmoid = {dim_in = {23}, dim_out = {23}},
},
- ['nerv.SoftmaxLayer'] = {
+ ['nerv.IdentityLayer'] = {
softmax = {dim_in = {79}, dim_out = {79}},
},
['nerv.DuplicateLayer'] = {
@@ -36,8 +38,30 @@ local connections = {
{'softmax[1]', '<output>[1]', 0},
}
-local graph = nerv.GraphLayer('graph', global_conf, {dim_in = {20}, dim_out = {79}, layer_repo = layer_repo, connections = connections})
+local graph = nerv.GraphLayer('graph', global_conf, {dim_in = {62}, dim_out = {79}, layer_repo = layer_repo, connections = connections})
local network = nerv.Network('network', global_conf, {network = graph})
-network:init(2,5)
+local batch = global_conf.batch_size
+local chunk = 5
+network:init(batch, chunk)
+
+local input = {}
+local output = {}
+local err_input = {}
+local err_output = {}
+local input_size = 62
+local output_size = 79
+for i = 1, chunk do
+ input[i] = {global_conf.cumat_type(batch, input_size)}
+ output[i] = {global_conf.cumat_type(batch, output_size)}
+ err_input[i] = {global_conf.cumat_type(batch, output_size)}
+ err_output[i] = {global_conf.cumat_type(batch, input_size)}
+end
+
+for i = 1, 100 do
+ network:mini_batch_init({seq_length = {5, 3}, new_seq = {2}})
+ network:propagate(input, output)
+ network:back_propagate(err_input, err_output, input, output)
+ network:update(err_input, input, output)
+end
diff --git a/nerv/nn/network.lua b/nerv/nn/network.lua
index 3cf052b..0bbcc59 100644
--- a/nerv/nn/network.lua
+++ b/nerv/nn/network.lua
@@ -320,7 +320,7 @@ function network:make_initial_store()
end
function network:set_input(input)
- for t = 1, #self.chunk_size do
+ for t = 1, self.chunk_size do
for i = 1, #self.dim_in do
local edge = self.socket.inputs[i]
local id, port, time = edge[1], edge[2], edge[3]
@@ -332,7 +332,7 @@ function network:set_input(input)
end
function network:set_output(output)
- for t = 1, #self.chunk_size do
+ for t = 1, self.chunk_size do
for i = 1, #self.dim_out do
local edge = self.socket.outputs[i]
local id, port, time = edge[1], edge[2], edge[3]
@@ -344,7 +344,7 @@ function network:set_output(output)
end
function network:set_err_input(err_input)
- for t = 1, #self.chunk_size do
+ for t = 1, self.chunk_size do
for i = 1, #self.dim_out do
local edge = self.socket.outputs[i]
local id, port, time = edge[1], edge[2], edge[3]
@@ -391,7 +391,9 @@ function network:mini_batch_init(information)
for i = 1, #self.layers do
local _, dim_out = self.layers[i]:get_dim()
for j = 1, #dim_out do
- self.legacy[t][i][j]:copy_from(self.output[t + self.chunk_size][i][j])
+ if t + self.chunk_size >= 1 and self.output_conn[i][j][1] ~= 0 then
+ self.legacy[t][i][j]:copy_from(self.output[t + self.chunk_size][i][j])
+ end
for k = 1, #self.info.new_seq do
local batch = self.info.new_seq[k]
self.legacy[t][i][j][batch - 1]:fill(self.nn_act_default)
@@ -414,8 +416,8 @@ function network:mini_batch_init(information)
end
function network:propagate(input, output)
- network:set_input(input)
- network:set_output(output)
+ self:set_input(input)
+ self:set_output(output)
for i = 1, #self.queue do
local t, id = self.queue[i].chunk, self.queue[i].id
if t <= self.max_length then
@@ -433,18 +435,18 @@ function network:propagate(input, output)
end
function network:back_propagate(bp_err, next_bp_err, input, output)
- network:set_input(input)
- network:set_output(output)
- network:set_err_input(bp_err)
- network:set_err_output(next_bp_err)
+ self:set_input(input)
+ self:set_output(output)
+ self:set_err_input(bp_err)
+ self:set_err_output(next_bp_err)
for i = #self.queue, 1, -1 do
local t, id = self.queue[i].chunk, self.queue[i].id
if t <= self.max_length then
-- flush border gradient
for j = 1, #self.border[t] do
local batch = self.border[t][j]
- local dim_in, _ = self.layers[id]:get_dim()
- for k = 1, #dim_in do
+ local _, dim_out = self.layers[id]:get_dim()
+ for k = 1, #dim_out do
self.err_input[t][id][k][batch - 1]:fill(0)
end
end
@@ -460,9 +462,9 @@ function network:back_propagate(bp_err, next_bp_err, input, output)
end
function network:update(bp_err, input, output)
- network:set_input(input)
- network:set_output(output)
- network:set_err_input(bp_err)
+ self:set_input(input)
+ self:set_output(output)
+ self:set_err_input(bp_err)
for i = 1, #self.queue do
local t, id = self.queue[i].chunk, self.queue[i].id
if t <= self.max_length then