diff options
Diffstat (limited to 'examples/asr_trainer.lua')
| -rw-r--r-- | examples/asr_trainer.lua | 29 |
1 files changed, 20 insertions, 9 deletions
diff --git a/examples/asr_trainer.lua b/examples/asr_trainer.lua index b43a547..05d770f 100644 --- a/examples/asr_trainer.lua +++ b/examples/asr_trainer.lua @@ -4,7 +4,7 @@ function build_trainer(ifname) local layer_repo = make_layer_repo(sublayer_repo, param_repo) local crit = get_criterion_layer(sublayer_repo) local network = get_network(layer_repo) - local iterative_trainer = function (ofname, scp_file, bp) + local iterative_trainer = function (prefix, scp_file, bp) gconf.randomize = bp -- build buffer local buffer = make_buffer(make_reader(scp_file, layer_repo)) @@ -18,7 +18,7 @@ function build_trainer(ifname) print_stat(crit) gconf.cnt = 0 end - if gconf.cnt == 100 then break end +-- if gconf.cnt == 100 then break end input = {data.main_scp, data.phone_state} output = {} @@ -33,9 +33,12 @@ function build_trainer(ifname) collectgarbage("collect") end print_stat(crit) - if bp then + nerv.CuMatrix.print_profile() + if (not bp) and prefix ~= nil then nerv.info("writing back...") - cf = nerv.ChunkFile(ofname, "w") + local fname = string.format("%s_cv%.3f.nerv", + prefix, get_accuracy(crit)) + cf = nerv.ChunkFile(fname, "w") for i, p in ipairs(network:get_params()) do cf:write_chunk(p) end @@ -52,7 +55,7 @@ halving_factor = 0.6 end_halving_inc = 0.1 min_iter = 1 max_iter = 20 -min_halving = 6 +min_halving = 5 gconf.batch_size = 256 gconf.buffer_size = 81920 @@ -64,10 +67,18 @@ local do_halving = false nerv.info("initial cross validation: %.3f", accu_best) for i = 1, max_iter do - nerv.info("iteration %d with lrate = %.6f", i, gconf.lrate) - local accu_tr = trainer(pf0 .. "_iter" .. i .. ".nerv", gconf.tr_scp, true) + nerv.info("[NN] begin iteration %d with lrate = %.6f", i, gconf.lrate) + local accu_tr = trainer(nil, gconf.tr_scp, true) nerv.info("[TR] training set %d: %.3f", i, accu_tr) - local accu_new = trainer(nil, gconf.cv_scp, false) + local accu_new = trainer( + string.format("%s_%s_iter_%d_lr%f_tr%.3f", + string.gsub( + (string.gsub(pf0, "(.*/)(.*)", "%2")), + "(.*)%..*", "%1"), + os.date("%Y%m%d%H%M%S"), + i, gconf.lrate, + accu_tr), + gconf.cv_scp, false) nerv.info("[CV] cross validation %d: %.3f", i, accu_new) -- TODO: revert the weights local accu_diff = accu_new - accu_best @@ -83,5 +94,5 @@ for i = 1, max_iter do if accu_new > accu_best then accu_best = accu_new end +-- nerv.Matrix.print_profile() end -nerv.Matrix.print_profile() |