diff options
Diffstat (limited to 'nerv/examples/swb_baseline.lua')
| -rw-r--r-- | nerv/examples/swb_baseline.lua | 84 |
1 files changed, 43 insertions, 41 deletions
diff --git a/nerv/examples/swb_baseline.lua b/nerv/examples/swb_baseline.lua index 51052ba..0ce8468 100644 --- a/nerv/examples/swb_baseline.lua +++ b/nerv/examples/swb_baseline.lua @@ -1,7 +1,5 @@ require 'htk_io' gconf = {lrate = 0.8, wcost = 1e-6, momentum = 0.9, - cumat_type = nerv.CuMatrixFloat, - mmat_type = nerv.MMatrixFloat, rearrange = true, -- just to make the context order consistent with old results, deprecated frm_ext = 5, frm_trim = 5, -- trim the first and last 5 frames, TNet just does this, deprecated @@ -9,8 +7,7 @@ gconf = {lrate = 0.8, wcost = 1e-6, momentum = 0.9, cv_scp = "/slfs1/users/mfy43/swb_ivec/train_cv.scp", htk_conf = "/slfs1/users/mfy43/swb_ivec/plp_0_d_a.conf", initialized_param = {"/slfs1/users/mfy43/swb_init.nerv", - "/slfs1/users/mfy43/swb_global_transf.nerv"}, - debug = false} + "/slfs1/users/mfy43/swb_global_transf.nerv"}} function make_layer_repo(param_repo) local layer_repo = nerv.LayerRepo( @@ -18,51 +15,51 @@ function make_layer_repo(param_repo) -- global transf ["nerv.BiasLayer"] = { - blayer1 = {{bias = "bias1"}, {dim_in = {429}, dim_out = {429}}}, - blayer2 = {{bias = "bias2"}, {dim_in = {429}, dim_out = {429}}} + blayer1 = {dim_in = {429}, dim_out = {429}, params = {bias = "bias1"}}, + blayer2 = {dim_in = {429}, dim_out = {429}, params = {bias = "bias2"}} }, ["nerv.WindowLayer"] = { - wlayer1 = {{window = "window1"}, {dim_in = {429}, dim_out = {429}}}, - wlayer2 = {{window = "window2"}, {dim_in = {429}, dim_out = {429}}} + wlayer1 = {dim_in = {429}, dim_out = {429}, params = {window = "window1"}}, + wlayer2 = {dim_in = {429}, dim_out = {429}, params = {window = "window2"}} }, -- biased linearity ["nerv.AffineLayer"] = { - affine0 = {{ltp = "affine0_ltp", bp = "affine0_bp"}, - {dim_in = {429}, dim_out = {2048}}}, - affine1 = {{ltp = "affine1_ltp", bp = "affine1_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine2 = {{ltp = "affine2_ltp", bp = "affine2_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine3 = {{ltp = "affine3_ltp", bp = "affine3_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine4 = {{ltp = "affine4_ltp", bp = "affine4_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine5 = {{ltp = "affine5_ltp", bp = "affine5_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine6 = {{ltp = "affine6_ltp", bp = "affine6_bp"}, - {dim_in = {2048}, dim_out = {2048}}}, - affine7 = {{ltp = "affine7_ltp", bp = "affine7_bp"}, - {dim_in = {2048}, dim_out = {3001}}} + affine0 = {dim_in = {429}, dim_out = {2048}, + params = {ltp = "affine0_ltp", bp = "affine0_bp"}}, + affine1 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine1_ltp", bp = "affine1_bp"}}, + affine2 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine2_ltp", bp = "affine2_bp"}}, + affine3 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine3_ltp", bp = "affine3_bp"}}, + affine4 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine4_ltp", bp = "affine4_bp"}}, + affine5 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine5_ltp", bp = "affine5_bp"}}, + affine6 = {dim_in = {2048}, dim_out = {2048}, + params = {ltp = "affine6_ltp", bp = "affine6_bp"}}, + affine7 = {dim_in = {2048}, dim_out = {3001}, + params = {ltp = "affine7_ltp", bp = "affine7_bp"}} }, ["nerv.SigmoidLayer"] = { - sigmoid0 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid1 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid2 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid3 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid4 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid5 = {{}, {dim_in = {2048}, dim_out = {2048}}}, - sigmoid6 = {{}, {dim_in = {2048}, dim_out = {2048}}} + sigmoid0 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid1 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid2 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid3 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid4 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid5 = {dim_in = {2048}, dim_out = {2048}}, + sigmoid6 = {dim_in = {2048}, dim_out = {2048}} }, ["nerv.SoftmaxCELayer"] = -- softmax + ce criterion layer for finetune output { - ce_crit = {{}, {dim_in = {3001, 1}, dim_out = {1}, compressed = true}} + ce_crit = {dim_in = {3001, 1}, dim_out = {1}, compressed = true} }, ["nerv.SoftmaxLayer"] = -- softmax for decode output { - softmax = {{}, {dim_in = {3001}, dim_out = {3001}}} + softmax = {dim_in = {3001}, dim_out = {3001}} } }, param_repo, gconf) @@ -70,7 +67,7 @@ function make_layer_repo(param_repo) { ["nerv.DAGLayer"] = { - global_transf = {{}, { + global_transf = { dim_in = {429}, dim_out = {429}, sub_layers = layer_repo, connections = { @@ -80,8 +77,8 @@ function make_layer_repo(param_repo) ["blayer2[1]"] = "wlayer2[1]", ["wlayer2[1]"] = "<output>[1]" } - }}, - main = {{}, { + }, + main = { dim_in = {429}, dim_out = {3001}, sub_layers = layer_repo, connections = { @@ -102,7 +99,7 @@ function make_layer_repo(param_repo) ["sigmoid6[1]"] = "affine7[1]", ["affine7[1]"] = "<output>[1]" } - }} + } } }, param_repo, gconf) @@ -110,7 +107,7 @@ function make_layer_repo(param_repo) { ["nerv.DAGLayer"] = { - ce_output = {{}, { + ce_output = { dim_in = {429, 1}, dim_out = {1}, sub_layers = layer_repo, connections = { @@ -119,8 +116,8 @@ function make_layer_repo(param_repo) ["<input>[2]"] = "ce_crit[2]", ["ce_crit[1]"] = "<output>[1]" } - }}, - softmax_output = {{}, { + }, + softmax_output = { dim_in = {429}, dim_out = {3001}, sub_layers = layer_repo, connections = { @@ -128,7 +125,7 @@ function make_layer_repo(param_repo) ["main[1]"] = "softmax[1]", ["softmax[1]"] = "<output>[1]" } - }} + } } }, param_repo, gconf) @@ -173,6 +170,7 @@ function make_buffer(readers) return nerv.SGDBuffer(gconf, { buffer_size = gconf.buffer_size, + batch_size = gconf.batch_size, randomize = gconf.randomize, readers = readers, use_gpu = true @@ -184,6 +182,10 @@ function get_input_order() {id = "phone_state"}} end +function get_decode_input_order() + return {{id = "main_scp", global_transf = true}} +end + function get_accuracy(layer_repo) local ce_crit = layer_repo:get_layer("ce_crit") return ce_crit.total_correct / ce_crit.total_frames * 100 |