...

Merge remote-tracking branch 'upstream/master'

13 files changed, 84 insertions, 53 deletions

diff --git a/common.c b/common.c
index 355d7ff..b4e39e6 100644
--- a/common.c
+++ b/common.c
@@ -3,7 +3,7 @@
 int nerv_error(lua_State *L, const char *err_mesg_fmt, ...) {
     va_list ap;
     va_start(ap, err_mesg_fmt);
-    lua_pushstring(L, "Nerv internal error: ");
+    lua_pushstring(L, "[nerv] internal error: ");
     lua_pushvfstring(L, err_mesg_fmt, ap); 
     lua_concat(L, 2);
     lua_error(L);
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()
diff --git a/examples/swb_baseline.lua b/examples/swb_baseline.lua
index f536777..28cc6d5 100644
--- a/examples/swb_baseline.lua
+++ b/examples/swb_baseline.lua
@@ -6,8 +6,8 @@ gconf = {lrate = 0.8, wcost = 1e-6, momentum = 0.9,
         tr_scp = "/slfs1/users/mfy43/swb_ivec/train_bp.scp",
         cv_scp = "/slfs1/users/mfy43/swb_ivec/train_cv.scp",
         htk_conf = "/slfs1/users/mfy43/swb_ivec/plp_0_d_a.conf",
-        global_transf = "global_transf.nerv",
-        initialized_param = "converted.nerv",
+        global_transf = "/slfs1/users/mfy43/swb_global_transf.nerv",
+        initialized_param = "/slfs1/users/mfy43/swb_init.nerv",
         debug = false}
 
 function make_param_repo(param_file)
@@ -154,10 +154,10 @@ end
 
 function print_stat(crit)
     nerv.info("*** training stat begin ***")
-    nerv.utils.printf("cross entropy:\t%.8f\n", crit.total_ce)
-    nerv.utils.printf("correct:\t%d\n", crit.total_correct)
-    nerv.utils.printf("frames:\t%d\n", crit.total_frames)
-    nerv.utils.printf("err/frm:\t%.8f\n", crit.total_ce / crit.total_frames)
-    nerv.utils.printf("accuracy:\t%.3f%%\n", get_accuracy(crit))
+    nerv.utils.printf("cross entropy:\t\t%.8f\n", crit.total_ce)
+    nerv.utils.printf("correct:\t\t%d\n", crit.total_correct)
+    nerv.utils.printf("frames:\t\t\t%d\n", crit.total_frames)
+    nerv.utils.printf("err/frm:\t\t%.8f\n", crit.total_ce / crit.total_frames)
+    nerv.utils.printf("accuracy:\t\t%.3f%%\n", get_accuracy(crit))
     nerv.info("*** training stat end ***")
 end
diff --git a/matrix/cuda_helper.h b/matrix/cuda_helper.h
index d6effdb..fde6f18 100644
--- a/matrix/cuda_helper.h
+++ b/matrix/cuda_helper.h
@@ -62,17 +62,14 @@ static const char *cublasGetErrorString(cublasStatus_t err) {
 
 #define PROFILE_START \
     do { \
-        cudaEvent_t start, stop; \
-        cudaEventCreate(&start); \
-        cudaEventCreate(&stop); \
-        cudaEventRecord(start, 0);
+        cudaEventRecord(profile_start, 0);
 #define PROFILE_STOP \
-        cudaEventRecord(stop, 0); \
-        cudaEventSynchronize(stop); \
+        cudaEventRecord(profile_stop, 0); \
+        cudaEventSynchronize(profile_stop); \
         float milliseconds = 0; \
-        cudaEventElapsedTime(&milliseconds, start, stop); \
+        cudaEventElapsedTime(&milliseconds, profile_start, profile_stop); \
         accu_profile(__func__, milliseconds / 1000); \
     } while (0);
 
-#define PROFILE_END 
+#define PROFILE_END
 #endif
diff --git a/matrix/cukernel.cu b/matrix/cukernel.cu
index fbac369..a19030a 100644
--- a/matrix/cukernel.cu
+++ b/matrix/cukernel.cu
@@ -9,6 +9,7 @@
 #undef MATRIX_ELEM
 #undef MATRIX_ELEM_PTR
 #undef MATRIX_ELEM_FMT
+#undef MATRIX_ELEM_WRITE_FMT
 
 #define cudak_(NAME) cudak_double_ ## NAME
 #define MATRIX_USE_DOUBLE
diff --git a/matrix/cumatrix.c b/matrix/cumatrix.c
index 4ebc5ff..af34fb4 100644
--- a/matrix/cumatrix.c
+++ b/matrix/cumatrix.c
@@ -1,10 +1,14 @@
 #define NERV_GENERIC_CUMATRIX
 #include "../common.h"
 #include "cuda_helper.h"
+#include <string.h>
+#define PROFILE_HASHMAP_SIZE 123457
 static cublasHandle_t cublas_handle;
+static cudaEvent_t profile_start, profile_stop;
 static HashMap *profile;
 
-int print_profile(lua_State *L) {
+static int print_profile(lua_State *L) {
+    (void)L;
     size_t i;
     fprintf(stderr, "*** [nerv cumatrix profile] **\n");
     for (i = 0; i < profile->size; i++)
@@ -18,7 +22,8 @@ int print_profile(lua_State *L) {
     return 0;
 }
 
-int clear_profile(lua_State *L) {
+static int clear_profile(lua_State *L) {
+    (void)L;
     hashmap_clear(profile);
     return 0;
 }
@@ -34,6 +39,25 @@ void accu_profile(const char *name, float delta) {
     *val += delta;
 }
 
+static const luaL_Reg cumatrix_methods[] = {
+    {"print_profile", print_profile},
+    {"clear_profile", clear_profile},
+    {NULL, NULL}
+};
+
+extern void nerv_matrix_cuda_float_init(lua_State *L);
+extern void nerv_matrix_cuda_double_init(lua_State *L);
+
+void nerv_cumatrix_init(lua_State *L) {
+    luaL_register(L, NULL, cumatrix_methods);
+    cublasCreate(&cublas_handle);
+    cudaEventCreate(&profile_start);
+    cudaEventCreate(&profile_stop);
+    profile = hashmap_create(PROFILE_HASHMAP_SIZE, bkdr_hash, strcmp);
+    nerv_matrix_cuda_float_init(L);
+    nerv_matrix_cuda_double_init(L);
+}
+
 #define MATRIX_USE_FLOAT
 #define cuda_matrix_(NAME) cuda_matrix_float_##NAME
 #define nerv_matrix_(NAME) nerv_matrix_cuda_float_##NAME
@@ -50,6 +74,7 @@ const char *nerv_matrix_(tname) = "nerv.CuMatrixFloat";
 #undef MATRIX_ELEM
 #undef MATRIX_ELEM_PTR
 #undef MATRIX_ELEM_FMT
+#undef MATRIX_ELEM_WRITE_FMT
 #undef MATRIX_CUMATRIX_HOST_TNAME
 
 #define MATRIX_USE_DOUBLE
diff --git a/matrix/generic/cumatrix.c b/matrix/generic/cumatrix.c
index 956e1e6..a8e18e0 100644
--- a/matrix/generic/cumatrix.c
+++ b/matrix/generic/cumatrix.c
@@ -11,11 +11,9 @@
 #define MATRIX_BASE_TNAME nerv_matrix_cuda_tname
 #define NERV_GENERIC_MATRIX
 #define NERV_GENERIC_CUKERNEL
-#define PROFILE_HASHMAP_SIZE 123457
 #include "../../common.h"
 #include "../cukernel.h"
 #include "../cuda_helper.h"
-#include <string.h>
 
 Matrix *nerv_matrix_(new_)(lua_State *L, long nrow, long ncol);
 void nerv_matrix_(data_free)(lua_State *L, Matrix *self);
@@ -442,8 +440,6 @@ static const luaL_Reg nerv_matrix_(extra_methods)[] = {
 
 static void cuda_matrix_(init)(lua_State *L) {
     luaN_append_methods(L, nerv_matrix_(extra_methods));
-    cublasCreate(&cublas_handle);
-    profile = hashmap_create(PROFILE_HASHMAP_SIZE, bkdr_hash, strcmp);
 }
 
 static void cuda_matrix_(free)(lua_State *L, MATRIX_ELEM *ptr) {
diff --git a/matrix/generic/matrix.c b/matrix/generic/matrix.c
index d6b0aea..e17fb42 100644
--- a/matrix/generic/matrix.c
+++ b/matrix/generic/matrix.c
@@ -6,6 +6,7 @@ extern const char *nerv_matrix_(tname);
 extern const char *MATRIX_BASE_TNAME;
 
 void nerv_matrix_(data_free)(lua_State *L, Matrix *self) {
+    (void)L;
     assert(*self->data_ref > 0);
     if (--(*self->data_ref) == 0)
     {
diff --git a/matrix/generic/mmatrix.c b/matrix/generic/mmatrix.c
index 2045d65..b0f0791 100644
--- a/matrix/generic/mmatrix.c
+++ b/matrix/generic/mmatrix.c
@@ -87,15 +87,9 @@ int nerv_matrix_(save)(lua_State *L) {
         MATRIX_ELEM *row = MATRIX_ROW_PTR(self, i);
         for (j = 0; j < ncol; j++)
             if (fprintf(fp, MATRIX_ELEM_WRITE_FMT " ", row[j]) < 0)
-            {
-                free(self);
                 return 0;
-            }
         if (fprintf(fp, "\n") < 0)
-        {
-             free(self);
              return 0;
-        }
     }
     return 0;
 }
diff --git a/matrix/init.c b/matrix/init.c
index 7b7f478..c29d7e9 100644
--- a/matrix/init.c
+++ b/matrix/init.c
@@ -5,21 +5,14 @@ const char *nerv_matrix_tname = "nerv.Matrix";
 const char *nerv_matrix_cuda_tname = "nerv.CuMatrix";
 const char *nerv_matrix_host_tname = "nerv.MMatrix";
 
-void nerv_matrix_host_float_init(lua_State *L);
-void nerv_matrix_cuda_float_init(lua_State *L);
-void nerv_matrix_host_double_init(lua_State *L);
-void nerv_matrix_cuda_double_init(lua_State *L);
-void nerv_matrix_host_int_init(lua_State *L);
-int print_profile(lua_State *L);
-int clear_profile(lua_State *L);
+void nerv_cumatrix_init(lua_State *L);
+void nerv_mmatrix_init(lua_State *L);
 
 static const luaL_Reg matrix_methods[] = {
     {"__tostring__", nerv_error_method_not_implemented },
     {"__add__", nerv_error_method_not_implemented },
     {"__sub__", nerv_error_method_not_implemented },
     {"__mul__", nerv_error_method_not_implemented },
-    {"print_profile", print_profile},
-    {"clear_profile", clear_profile},
     {NULL, NULL}
 };
 
@@ -32,13 +25,11 @@ void nerv_matrix_init(lua_State *L) {
     /* CuMatrix inherits from Matrix */
     luaT_newmetatable(L, nerv_matrix_cuda_tname, nerv_matrix_tname,
                             NULL, NULL, NULL);
-    nerv_matrix_cuda_float_init(L);
-    nerv_matrix_cuda_double_init(L);
-
+    nerv_cumatrix_init(L);
+    lua_pop(L, 1);
     /* MMatrix inherits from Matrix */
     luaT_newmetatable(L, nerv_matrix_host_tname, nerv_matrix_tname,
                             NULL, NULL, NULL);
-    nerv_matrix_host_float_init(L);
-    nerv_matrix_host_double_init(L);
-    nerv_matrix_host_int_init(L);
+    nerv_mmatrix_init(L);
+    lua_pop(L, 1);
 }
diff --git a/matrix/mmatrix.c b/matrix/mmatrix.c
index 81f8dfc..ffc058d 100644
--- a/matrix/mmatrix.c
+++ b/matrix/mmatrix.c
@@ -1,4 +1,15 @@
 #define NERV_GENERIC_MMATRIX
+#include "../common.h"
+void nerv_matrix_host_float_init(lua_State *L);
+void nerv_matrix_host_double_init(lua_State *L);
+void nerv_matrix_host_int_init(lua_State *L);
+
+void nerv_mmatrix_init(lua_State *L) {
+    nerv_matrix_host_float_init(L);
+    nerv_matrix_host_double_init(L);
+    nerv_matrix_host_int_init(L);
+}
+
 #define MATRIX_USE_FLOAT
 #define host_matrix_(NAME) host_matrix_float_##NAME
 #define nerv_matrix_(NAME) nerv_matrix_host_float_##NAME
@@ -10,6 +21,7 @@ const char *nerv_matrix_(tname) = "nerv.MMatrixFloat";
 #undef MATRIX_ELEM
 #undef MATRIX_ELEM_PTR
 #undef MATRIX_ELEM_FMT
+#undef MATRIX_ELEM_WRITE_FMT
 
 #define NERV_GENERIC_MMATRIX
 #define MATRIX_USE_DOUBLE
@@ -23,6 +35,7 @@ const char *nerv_matrix_(tname) = "nerv.MMatrixDouble";
 #undef MATRIX_ELEM
 #undef MATRIX_ELEM_PTR
 #undef MATRIX_ELEM_FMT
+#undef MATRIX_ELEM_WRITE_FMT
 
 #define NERV_GENERIC_MMATRIX
 #define MATRIX_USE_INT
diff --git a/nerv.lua b/nerv.lua
index ce6bc44..467d926 100644
--- a/nerv.lua
+++ b/nerv.lua
@@ -10,7 +10,9 @@ function nerv.error_method_not_implemented()
 end
 
 function nerv.info(fmt, ...)
-    nerv.utils.printf("[nerv] info: " .. fmt .. "\n", ...)
+    nerv.utils.printf(
+        string.format("(%s)[nerv] info: %s\n",
+            os.date("%H:%M:%S %F"), fmt), ...)
 end
 
 -- Torch C API wrapper
diff --git a/speech b/speech
-Subproject be0e51156e1af9d619160bba1aa7c2eb2df3073
+Subproject aee0d372e6b06a217f24bea5c88962b97e0ca0e