jesus branch: script and (minor) code fixes; add some more example scripts.

diff --git a/egs/swbd/s5c/local/chain/run_tdnn_3e.sh b/egs/swbd/s5c/local/chain/run_tdnn_3e.sh
new file mode 100755 (executable)
index 0000000..7f80fd1
--- /dev/null
+++ b/egs/swbd/s5c/local/chain/run_tdnn_3e.sh
@@ -0,0 +1,274 @@
+#!/bin/bash
+
+# _3e is as _3d, but increasing the --num-jesus-blocks from 100 (the default)
+# to 200 in order to reduce computation in the Jesus layer.
+
+# _3d is as _2y, and re-using the egs, but using --jesus-opts and
+# configs from make_jesus_configs.py.
+#  --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000" \
+#   --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3 -6,-3,0,3 -6,-3,0,3"
+
+# _2y is as _2o, but increasing the --frames-per-iter by a factor of 1.5, from
+# 800k to 1.2 million.  The aim is to avoid some of the per-job overhead
+# (model-averaging, etc.), since each iteration takes only a minute or so.
+#  I added the results to the table below.  It seems the same on average-
+# which is good.  We'll probably keep this configuration.
+
+# _2o is as _2m, but going back to our original 2-state topology, which it turns
+# out that I never tested to WER.
+# hm--- it's about the same, or maybe slightly better!
+# caution: accidentally overwrote most of this dir, but kept the key stuff.
+
+# note: when I compare with the rerun of 2o (not shown), this run is actually
+# better.
+# WER on          2m        2o          2y    [ now comparing 2o->2y:]
+# train_dev,tg    17.22     17.24       16.99  0.2% better
+# train_dev,fg    15.87     15.93       15.86  0.1% better
+# eval2000,tg     18.7      18.7        18.9   0.2% worse
+# eval2000,fg     17.0      16.9        17.0   0.1% worse
+
+# train-prob,final  -0.0803   -0.0835
+# valid-prob,final  -0.0116   -0.0122
+
+# _2m is as _2k, but setting --leftmost-questions-truncate=-1, i.e. disabling
+# that mechanism.
+
+# _2k is as _2i, but doing the same change as in _s -> _2e, in which we
+#  set --apply-deriv-weights false and --frames-overlap-per-eg 0.
+
+# _2i is as _2d but with a new set of code for estimating the LM, in which we compute
+# the log-like change when deciding which states to back off.  The code is not the same
+# as the one in 2{f,g,h}.  We have only the options --num-extra-lm-states=2000.  By
+# default it estimates a 4-gram, with 3-gram as the no-prune order.  So the configuration
+# is quite similar to 2d, except new/more-exact code is used.
+
+# _2d is as _2c but with different LM options:
+# --lm-opts "--ngram-order=4 --leftmost-context-questions=/dev/null --num-extra-states=2000"
+# ... this gives us a kind of pruned 4-gram language model, instead of a 3-gram.
+# the --leftmost-context-questions=/dev/null option overrides the leftmost-context-questions
+# provided from the tree-building, and effectively puts the leftmost context position as a single
+# set.
+#   This seems definitely helpful: on train_dev, with tg improvement is 18.12->17.55 and with fg
+#  from 16.73->16.14; and on eval2000, with tg from 19.8->19.5 and with fg from 17.8->17.6.
+
+# _2c is as _2a but after a code change in which we start using transition-scale
+# and self-loop-scale of 1 instead of zero in training; we change the options to
+# mkgraph used in testing, to set the scale to 1.0.  This shouldn't affect
+# results at all; it's is mainly for convenience in pushing weights in graphs,
+# and checking that graphs are stochastic.
+
+# _2a is as _z but setting --lm-opts "--num-extra-states=8000".
+
+# _z is as _x but setting  --lm-opts "--num-extra-states=2000".
+# (see also y, which has --num-extra-states=500).
+
+# _x is as _s but setting  --lm-opts "--num-extra-states=0".
+#  this is a kind of repeat of the u->v experiment, where it seemed to make things
+#  worse, but there were other factors involved in that so I want to be sure.
+
+# _s is as _q but setting pdf-boundary-penalty to 0.0
+# This is helpful: 19.8->18.0 after fg rescoring on all of eval2000,
+# and 18.07 -> 16.96 on train_dev, after fg rescoring.
+
+# _q is as _p except making the same change as from n->o, which
+# reduces the parameters to try to reduce over-training.  We reduce
+# relu-dim from 1024 to 850, and target num-states from 12k to 9k,
+# and modify the splicing setup.
+# note: I don't rerun the tree-building, I just use the '5o' treedir.
+
+# _p is as _m except with a code change in which we switch to a different, more
+# exact mechanism to deal with the edges of the egs, and correspondingly
+# different script options... we now dump weights with the egs, and apply the
+# weights to the derivative w.r.t. the output instead of using the
+# --min-deriv-time and --max-deriv-time options.  Increased the frames-overlap
+# to 30 also.  This wil.  give 10 frames on each side with zero derivs, then
+# ramping up to a weight of 1.0 over 10 frames.
+
+# _m is as _k but after a code change that makes the denominator FST more
+# compact.  I am rerunning in order to verify that the WER is not changed (since
+# it's possible in principle that due to edge effects related to weight-pushing,
+# the results could be a bit different).
+#  The results are inconsistently different but broadly the same.  On all of eval2000,
+#  the change k->m is 20.7->20.9 with tg LM and 18.9->18.6 after rescoring.
+#  On the train_dev data, the change is  19.3->18.9 with tg LM and 17.6->17.6 after rescoring.
+
+
+# _k is as _i but reverting the g->h change, removing the --scale-max-param-change
+# option and setting max-param-change to 1..  Using the same egs.
+
+# _i is as _h but longer egs: 150 frames instead of 75, and
+# 128 elements per minibatch instead of 256.
+
+# _h is as _g but different application of max-param-change (use --scale-max-param-change true)
+
+# _g is as _f but more splicing at last layer.
+
+# _f is as _e but with 30 as the number of left phone classes instead
+# of 10.
+
+# _e is as _d but making it more similar in configuration to _b.
+# (turns out b was better than a after all-- the egs' likelihoods had to
+# be corrected before comparing them).
+# the changes (vs. d) are: change num-pdfs target from 8k to 12k,
+# multiply learning rates by 5, and set final-layer-normalize-target to 0.5.
+
+# _d is as _c but with a modified topology (with 4 distinct states per phone
+# instead of 2), and a slightly larger num-states (8000) to compensate for the
+# different topology, which has more states.
+
+# _c is as _a but getting rid of the final-layer-normalize-target (making it 1.0
+# as the default) as it's not clear that it was helpful; using the old learning-rates;
+# and modifying the target-num-states to 7000.
+
+# _b is as as _a except for configuration changes: using 12k num-leaves instead of
+# 5k; using 5 times larger learning rate, and --final-layer-normalize-target=0.5,
+# which will make the final layer learn less fast compared with other layers.
+
+set -e
+
+# configs for 'chain'
+stage=12
+train_stage=-10
+get_egs_stage=-10
+speed_perturb=true
+dir=exp/chain/tdnn_3e  # Note: _sp will get added to this if $speed_perturb == true.
+
+# training options
+num_epochs=4
+initial_effective_lrate=0.001
+final_effective_lrate=0.0001
+leftmost_questions_truncate=-1
+max_param_change=1.0
+final_layer_normalize_target=0.5
+num_jobs_initial=3
+num_jobs_final=16
+minibatch_size=128
+frames_per_eg=150
+remove_egs=false
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+# The iVector-extraction and feature-dumping parts are the same as the standard
+# nnet3 setup, and you can skip them by setting "--stage 8" if you have already
+# run those things.
+
+suffix=
+if [ "$speed_perturb" == "true" ]; then
+  suffix=_sp
+fi
+
+dir=${dir}$suffix
+train_set=train_nodup$suffix
+ali_dir=exp/tri4_ali_nodup$suffix
+treedir=exp/chain/tri5_2y_tree$suffix
+lang=data/lang_chain_2y
+
+
+# if we are using the speed-perturbed data we need to generate
+# alignments for it.
+local/nnet3/run_ivector_common.sh --stage $stage \
+  --speed-perturb $speed_perturb \
+  --generate-alignments $speed_perturb || exit 1;
+
+
+if [ $stage -le 9 ]; then
+  # Get the alignments as lattices (gives the CTC training more freedom).
+  # use the same num-jobs as the alignments
+  nj=$(cat exp/tri4_ali_nodup$suffix/num_jobs) || exit 1;
+  steps/align_fmllr_lats.sh --nj $nj --cmd "$train_cmd" data/$train_set \
+    data/lang exp/tri4 exp/tri4_lats_nodup$suffix
+  rm exp/tri4_lats_nodup$suffix/fsts.*.gz # save space
+fi
+
+
+if [ $stage -le 10 ]; then
+  # Create a version of the lang/ directory that has one state per phone in the
+  # topo file. [note, it really has two states.. the first one is only repeated
+  # once, the second one has zero or more repeats.]
+  rm -rf $lang
+  cp -r data/lang $lang
+  silphonelist=$(cat $lang/phones/silence.csl) || exit 1;
+  nonsilphonelist=$(cat $lang/phones/nonsilence.csl) || exit 1;
+  # Use our special topology... note that later on may have to tune this
+  # topology.
+  steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >$lang/topo
+fi
+
+if [ $stage -le 11 ]; then
+  # Build a tree using our new topology.
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+      --leftmost-questions-truncate $leftmost_questions_truncate \
+      --cmd "$train_cmd" 9000 data/$train_set $lang $ali_dir $treedir
+fi
+
+if [ $stage -le 12 ]; then
+  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
+    utils/create_split_dir.pl \
+     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/swbd-$(date +'%m_%d_%H_%M')/s5c/$dir/egs/storage $dir/egs/storage
+  fi
+
+ touch $dir/egs/.nodelete # keep egs around when that run dies.
+
+ steps/nnet3/chain/train_tdnn.sh --stage $train_stage \
+    --egs-dir exp/chain/tdnn_2y_sp/egs \
+    --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000 --num-jesus-blocks 200" \
+    --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3 -6,-3,0,3 -6,-3,0,3" \
+    --apply-deriv-weights false \
+    --frames-per-iter 1200000 \
+    --lm-opts "--num-extra-lm-states=2000" \
+    --get-egs-stage $get_egs_stage \
+    --minibatch-size $minibatch_size \
+    --egs-opts "--frames-overlap-per-eg 0" \
+    --frames-per-eg $frames_per_eg \
+    --num-epochs $num_epochs --num-jobs-initial $num_jobs_initial --num-jobs-final $num_jobs_final \
+    --feat-type raw \
+    --online-ivector-dir exp/nnet3/ivectors_${train_set} \
+    --cmvn-opts "--norm-means=false --norm-vars=false" \
+    --initial-effective-lrate $initial_effective_lrate --final-effective-lrate $final_effective_lrate \
+    --max-param-change $max_param_change \
+    --final-layer-normalize-target $final_layer_normalize_target \
+    --relu-dim 850 \
+    --cmd "$decode_cmd" \
+    --remove-egs $remove_egs \
+    data/${train_set}_hires $treedir exp/tri4_lats_nodup$suffix $dir  || exit 1;
+fi
+
+if [ $stage -le 13 ]; then
+  # Note: it might appear that this $lang directory is mismatched, and it is as
+  # far as the 'topo' is concerned, but this script doesn't read the 'topo' from
+  # the lang directory.
+  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_sw1_tg $dir $dir/graph_sw1_tg
+fi
+
+decode_suff=sw1_tg
+graph_dir=$dir/graph_sw1_tg
+if [ $stage -le 14 ]; then
+  for decode_set in train_dev eval2000; do
+      (
+      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+          --nj 50 --cmd "$decode_cmd" \
+          --online-ivector-dir exp/nnet3/ivectors_${decode_set} \
+         $graph_dir data/${decode_set}_hires $dir/decode_${decode_set}_${decode_suff} || exit 1;
+      if $has_fisher; then
+          steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" \
+            data/lang_sw1_{tg,fsh_fg} data/${decode_set}_hires \
+            $dir/decode_${decode_set}_sw1_{tg,fsh_fg} || exit 1;
+      fi
+      ) &
+  done
+fi
+wait;
+exit 0;

diff --git a/egs/swbd/s5c/local/chain/run_tdnn_3f.sh b/egs/swbd/s5c/local/chain/run_tdnn_3f.sh
new file mode 100755 (executable)
index 0000000..c39d812
--- /dev/null
+++ b/egs/swbd/s5c/local/chain/run_tdnn_3f.sh
@@ -0,0 +1,282 @@
+#!/bin/bash
+
+
+# _3f is as _3e, but modifying the splicing setup to add (left) recurrence:
+# added the :3's in   --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3:-3 -6,-3,0,3:-3 -6,-3,0,3:-3"
+# Therefore it's
+# no longer really a tdnn, more like an RNN combined with TDNN.  BTW, I'm not re-dumping egs with extra
+# context, and this isn't really ideal - I want to see if this seems promising first.
+
+# _3e is as _3d, but increasing the --num-jesus-blocks from 100 (the default)
+# to 200 in order to reduce computation in the Jesus layer.
+
+# _3d is as _2y, and re-using the egs, but using --jesus-opts and
+# configs from make_jesus_configs.py.
+#  --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000" \
+#   --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3 -6,-3,0,3 -6,-3,0,3"
+
+# _2y is as _2o, but increasing the --frames-per-iter by a factor of 1.5, from
+# 800k to 1.2 million.  The aim is to avoid some of the per-job overhead
+# (model-averaging, etc.), since each iteration takes only a minute or so.
+#  I added the results to the table below.  It seems the same on average-
+# which is good.  We'll probably keep this configuration.
+
+# _2o is as _2m, but going back to our original 2-state topology, which it turns
+# out that I never tested to WER.
+# hm--- it's about the same, or maybe slightly better!
+# caution: accidentally overwrote most of this dir, but kept the key stuff.
+
+# note: when I compare with the rerun of 2o (not shown), this run is actually
+# better.
+# WER on          2m        2o          2y    [ now comparing 2o->2y:]
+# train_dev,tg    17.22     17.24       16.99  0.2% better
+# train_dev,fg    15.87     15.93       15.86  0.1% better
+# eval2000,tg     18.7      18.7        18.9   0.2% worse
+# eval2000,fg     17.0      16.9        17.0   0.1% worse
+
+# train-prob,final  -0.0803   -0.0835
+# valid-prob,final  -0.0116   -0.0122
+
+# _2m is as _2k, but setting --leftmost-questions-truncate=-1, i.e. disabling
+# that mechanism.
+
+# _2k is as _2i, but doing the same change as in _s -> _2e, in which we
+#  set --apply-deriv-weights false and --frames-overlap-per-eg 0.
+
+# _2i is as _2d but with a new set of code for estimating the LM, in which we compute
+# the log-like change when deciding which states to back off.  The code is not the same
+# as the one in 2{f,g,h}.  We have only the options --num-extra-lm-states=2000.  By
+# default it estimates a 4-gram, with 3-gram as the no-prune order.  So the configuration
+# is quite similar to 2d, except new/more-exact code is used.
+
+# _2d is as _2c but with different LM options:
+# --lm-opts "--ngram-order=4 --leftmost-context-questions=/dev/null --num-extra-states=2000"
+# ... this gives us a kind of pruned 4-gram language model, instead of a 3-gram.
+# the --leftmost-context-questions=/dev/null option overrides the leftmost-context-questions
+# provided from the tree-building, and effectively puts the leftmost context position as a single
+# set.
+#   This seems definitely helpful: on train_dev, with tg improvement is 18.12->17.55 and with fg
+#  from 16.73->16.14; and on eval2000, with tg from 19.8->19.5 and with fg from 17.8->17.6.
+
+# _2c is as _2a but after a code change in which we start using transition-scale
+# and self-loop-scale of 1 instead of zero in training; we change the options to
+# mkgraph used in testing, to set the scale to 1.0.  This shouldn't affect
+# results at all; it's is mainly for convenience in pushing weights in graphs,
+# and checking that graphs are stochastic.
+
+# _2a is as _z but setting --lm-opts "--num-extra-states=8000".
+
+# _z is as _x but setting  --lm-opts "--num-extra-states=2000".
+# (see also y, which has --num-extra-states=500).
+
+# _x is as _s but setting  --lm-opts "--num-extra-states=0".
+#  this is a kind of repeat of the u->v experiment, where it seemed to make things
+#  worse, but there were other factors involved in that so I want to be sure.
+
+# _s is as _q but setting pdf-boundary-penalty to 0.0
+# This is helpful: 19.8->18.0 after fg rescoring on all of eval2000,
+# and 18.07 -> 16.96 on train_dev, after fg rescoring.
+
+# _q is as _p except making the same change as from n->o, which
+# reduces the parameters to try to reduce over-training.  We reduce
+# relu-dim from 1024 to 850, and target num-states from 12k to 9k,
+# and modify the splicing setup.
+# note: I don't rerun the tree-building, I just use the '5o' treedir.
+
+# _p is as _m except with a code change in which we switch to a different, more
+# exact mechanism to deal with the edges of the egs, and correspondingly
+# different script options... we now dump weights with the egs, and apply the
+# weights to the derivative w.r.t. the output instead of using the
+# --min-deriv-time and --max-deriv-time options.  Increased the frames-overlap
+# to 30 also.  This wil.  give 10 frames on each side with zero derivs, then
+# ramping up to a weight of 1.0 over 10 frames.
+
+# _m is as _k but after a code change that makes the denominator FST more
+# compact.  I am rerunning in order to verify that the WER is not changed (since
+# it's possible in principle that due to edge effects related to weight-pushing,
+# the results could be a bit different).
+#  The results are inconsistently different but broadly the same.  On all of eval2000,
+#  the change k->m is 20.7->20.9 with tg LM and 18.9->18.6 after rescoring.
+#  On the train_dev data, the change is  19.3->18.9 with tg LM and 17.6->17.6 after rescoring.
+
+
+# _k is as _i but reverting the g->h change, removing the --scale-max-param-change
+# option and setting max-param-change to 1..  Using the same egs.
+
+# _i is as _h but longer egs: 150 frames instead of 75, and
+# 128 elements per minibatch instead of 256.
+
+# _h is as _g but different application of max-param-change (use --scale-max-param-change true)
+
+# _g is as _f but more splicing at last layer.
+
+# _f is as _e but with 30 as the number of left phone classes instead
+# of 10.
+
+# _e is as _d but making it more similar in configuration to _b.
+# (turns out b was better than a after all-- the egs' likelihoods had to
+# be corrected before comparing them).
+# the changes (vs. d) are: change num-pdfs target from 8k to 12k,
+# multiply learning rates by 5, and set final-layer-normalize-target to 0.5.
+
+# _d is as _c but with a modified topology (with 4 distinct states per phone
+# instead of 2), and a slightly larger num-states (8000) to compensate for the
+# different topology, which has more states.
+
+# _c is as _a but getting rid of the final-layer-normalize-target (making it 1.0
+# as the default) as it's not clear that it was helpful; using the old learning-rates;
+# and modifying the target-num-states to 7000.
+
+# _b is as as _a except for configuration changes: using 12k num-leaves instead of
+# 5k; using 5 times larger learning rate, and --final-layer-normalize-target=0.5,
+# which will make the final layer learn less fast compared with other layers.
+
+set -e
+
+# configs for 'chain'
+stage=12
+train_stage=-10
+get_egs_stage=-10
+speed_perturb=true
+dir=exp/chain/tdnn_3f  # Note: _sp will get added to this if $speed_perturb == true.
+
+# training options
+num_epochs=4
+initial_effective_lrate=0.001
+final_effective_lrate=0.0001
+leftmost_questions_truncate=-1
+max_param_change=1.0
+final_layer_normalize_target=0.5
+num_jobs_initial=3
+num_jobs_final=16
+minibatch_size=128
+frames_per_eg=150
+remove_egs=false
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+# The iVector-extraction and feature-dumping parts are the same as the standard
+# nnet3 setup, and you can skip them by setting "--stage 8" if you have already
+# run those things.
+
+suffix=
+if [ "$speed_perturb" == "true" ]; then
+  suffix=_sp
+fi
+
+dir=${dir}$suffix
+train_set=train_nodup$suffix
+ali_dir=exp/tri4_ali_nodup$suffix
+treedir=exp/chain/tri5_2y_tree$suffix
+lang=data/lang_chain_2y
+
+
+# if we are using the speed-perturbed data we need to generate
+# alignments for it.
+local/nnet3/run_ivector_common.sh --stage $stage \
+  --speed-perturb $speed_perturb \
+  --generate-alignments $speed_perturb || exit 1;
+
+
+if [ $stage -le 9 ]; then
+  # Get the alignments as lattices (gives the CTC training more freedom).
+  # use the same num-jobs as the alignments
+  nj=$(cat exp/tri4_ali_nodup$suffix/num_jobs) || exit 1;
+  steps/align_fmllr_lats.sh --nj $nj --cmd "$train_cmd" data/$train_set \
+    data/lang exp/tri4 exp/tri4_lats_nodup$suffix
+  rm exp/tri4_lats_nodup$suffix/fsts.*.gz # save space
+fi
+
+
+if [ $stage -le 10 ]; then
+  # Create a version of the lang/ directory that has one state per phone in the
+  # topo file. [note, it really has two states.. the first one is only repeated
+  # once, the second one has zero or more repeats.]
+  rm -rf $lang
+  cp -r data/lang $lang
+  silphonelist=$(cat $lang/phones/silence.csl) || exit 1;
+  nonsilphonelist=$(cat $lang/phones/nonsilence.csl) || exit 1;
+  # Use our special topology... note that later on may have to tune this
+  # topology.
+  steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >$lang/topo
+fi
+
+if [ $stage -le 11 ]; then
+  # Build a tree using our new topology.
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+      --leftmost-questions-truncate $leftmost_questions_truncate \
+      --cmd "$train_cmd" 9000 data/$train_set $lang $ali_dir $treedir
+fi
+
+if [ $stage -le 12 ]; then
+  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
+    utils/create_split_dir.pl \
+     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/swbd-$(date +'%m_%d_%H_%M')/s5c/$dir/egs/storage $dir/egs/storage
+  fi
+
+ touch $dir/egs/.nodelete # keep egs around when that run dies.
+
+ steps/nnet3/chain/train_tdnn.sh --stage $train_stage \
+    --egs-dir exp/chain/tdnn_2y_sp/egs \
+    --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000 --num-jesus-blocks 200" \
+    --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3:-3 -6,-3,0,3:-3 -6,-3,0,3:-3" \
+    --apply-deriv-weights false \
+    --frames-per-iter 1200000 \
+    --lm-opts "--num-extra-lm-states=2000" \
+    --get-egs-stage $get_egs_stage \
+    --minibatch-size $minibatch_size \
+    --egs-opts "--frames-overlap-per-eg 0" \
+    --frames-per-eg $frames_per_eg \
+    --num-epochs $num_epochs --num-jobs-initial $num_jobs_initial --num-jobs-final $num_jobs_final \
+    --feat-type raw \
+    --online-ivector-dir exp/nnet3/ivectors_${train_set} \
+    --cmvn-opts "--norm-means=false --norm-vars=false" \
+    --initial-effective-lrate $initial_effective_lrate --final-effective-lrate $final_effective_lrate \
+    --max-param-change $max_param_change \
+    --final-layer-normalize-target $final_layer_normalize_target \
+    --relu-dim 850 \
+    --cmd "$decode_cmd" \
+    --remove-egs $remove_egs \
+    data/${train_set}_hires $treedir exp/tri4_lats_nodup$suffix $dir  || exit 1;
+fi
+
+if [ $stage -le 13 ]; then
+  # Note: it might appear that this $lang directory is mismatched, and it is as
+  # far as the 'topo' is concerned, but this script doesn't read the 'topo' from
+  # the lang directory.
+  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_sw1_tg $dir $dir/graph_sw1_tg
+fi
+
+decode_suff=sw1_tg
+graph_dir=$dir/graph_sw1_tg
+if [ $stage -le 14 ]; then
+  for decode_set in train_dev eval2000; do
+      (
+      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+         --extra-left-context 20 \
+          --nj 50 --cmd "$decode_cmd" \
+          --online-ivector-dir exp/nnet3/ivectors_${decode_set} \
+         $graph_dir data/${decode_set}_hires $dir/decode_${decode_set}_${decode_suff} || exit 1;
+      if $has_fisher; then
+          steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" \
+            data/lang_sw1_{tg,fsh_fg} data/${decode_set}_hires \
+            $dir/decode_${decode_set}_sw1_{tg,fsh_fg} || exit 1;
+      fi
+      ) &
+  done
+fi
+wait;
+exit 0;

diff --git a/egs/swbd/s5c/local/chain/run_tdnn_3g.sh b/egs/swbd/s5c/local/chain/run_tdnn_3g.sh
new file mode 100755 (executable)
index 0000000..cf742d3
--- /dev/null
+++ b/egs/swbd/s5c/local/chain/run_tdnn_3g.sh
@@ -0,0 +1,284 @@
+#!/bin/bash
+
+# _3g is as _3f but using 100 blocks instead of 200, as in d->e 200 groups was found
+# to be worde.
+
+# _3f is as _3e, but modifying the splicing setup to add (left) recurrence:
+# added the :3's in   --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3:-3 -6,-3,0,3:-3 -6,-3,0,3:-3"
+# Therefore it's
+# no longer really a tdnn, more like an RNN combined with TDNN.  BTW, I'm not re-dumping egs with extra
+# context, and this isn't really ideal - I want to see if this seems promising first.
+
+# _3e is as _3d, but increasing the --num-jesus-blocks from 100 (the default)
+# to 200 in order to reduce computation in the Jesus layer.
+
+# _3d is as _2y, and re-using the egs, but using --jesus-opts and
+# configs from make_jesus_configs.py.
+#  --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000" \
+#   --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3 -6,-3,0,3 -6,-3,0,3"
+
+# _2y is as _2o, but increasing the --frames-per-iter by a factor of 1.5, from
+# 800k to 1.2 million.  The aim is to avoid some of the per-job overhead
+# (model-averaging, etc.), since each iteration takes only a minute or so.
+#  I added the results to the table below.  It seems the same on average-
+# which is good.  We'll probably keep this configuration.
+
+# _2o is as _2m, but going back to our original 2-state topology, which it turns
+# out that I never tested to WER.
+# hm--- it's about the same, or maybe slightly better!
+# caution: accidentally overwrote most of this dir, but kept the key stuff.
+
+# note: when I compare with the rerun of 2o (not shown), this run is actually
+# better.
+# WER on          2m        2o          2y    [ now comparing 2o->2y:]
+# train_dev,tg    17.22     17.24       16.99  0.2% better
+# train_dev,fg    15.87     15.93       15.86  0.1% better
+# eval2000,tg     18.7      18.7        18.9   0.2% worse
+# eval2000,fg     17.0      16.9        17.0   0.1% worse
+
+# train-prob,final  -0.0803   -0.0835
+# valid-prob,final  -0.0116   -0.0122
+
+# _2m is as _2k, but setting --leftmost-questions-truncate=-1, i.e. disabling
+# that mechanism.
+
+# _2k is as _2i, but doing the same change as in _s -> _2e, in which we
+#  set --apply-deriv-weights false and --frames-overlap-per-eg 0.
+
+# _2i is as _2d but with a new set of code for estimating the LM, in which we compute
+# the log-like change when deciding which states to back off.  The code is not the same
+# as the one in 2{f,g,h}.  We have only the options --num-extra-lm-states=2000.  By
+# default it estimates a 4-gram, with 3-gram as the no-prune order.  So the configuration
+# is quite similar to 2d, except new/more-exact code is used.
+
+# _2d is as _2c but with different LM options:
+# --lm-opts "--ngram-order=4 --leftmost-context-questions=/dev/null --num-extra-states=2000"
+# ... this gives us a kind of pruned 4-gram language model, instead of a 3-gram.
+# the --leftmost-context-questions=/dev/null option overrides the leftmost-context-questions
+# provided from the tree-building, and effectively puts the leftmost context position as a single
+# set.
+#   This seems definitely helpful: on train_dev, with tg improvement is 18.12->17.55 and with fg
+#  from 16.73->16.14; and on eval2000, with tg from 19.8->19.5 and with fg from 17.8->17.6.
+
+# _2c is as _2a but after a code change in which we start using transition-scale
+# and self-loop-scale of 1 instead of zero in training; we change the options to
+# mkgraph used in testing, to set the scale to 1.0.  This shouldn't affect
+# results at all; it's is mainly for convenience in pushing weights in graphs,
+# and checking that graphs are stochastic.
+
+# _2a is as _z but setting --lm-opts "--num-extra-states=8000".
+
+# _z is as _x but setting  --lm-opts "--num-extra-states=2000".
+# (see also y, which has --num-extra-states=500).
+
+# _x is as _s but setting  --lm-opts "--num-extra-states=0".
+#  this is a kind of repeat of the u->v experiment, where it seemed to make things
+#  worse, but there were other factors involved in that so I want to be sure.
+
+# _s is as _q but setting pdf-boundary-penalty to 0.0
+# This is helpful: 19.8->18.0 after fg rescoring on all of eval2000,
+# and 18.07 -> 16.96 on train_dev, after fg rescoring.
+
+# _q is as _p except making the same change as from n->o, which
+# reduces the parameters to try to reduce over-training.  We reduce
+# relu-dim from 1024 to 850, and target num-states from 12k to 9k,
+# and modify the splicing setup.
+# note: I don't rerun the tree-building, I just use the '5o' treedir.
+
+# _p is as _m except with a code change in which we switch to a different, more
+# exact mechanism to deal with the edges of the egs, and correspondingly
+# different script options... we now dump weights with the egs, and apply the
+# weights to the derivative w.r.t. the output instead of using the
+# --min-deriv-time and --max-deriv-time options.  Increased the frames-overlap
+# to 30 also.  This wil.  give 10 frames on each side with zero derivs, then
+# ramping up to a weight of 1.0 over 10 frames.
+
+# _m is as _k but after a code change that makes the denominator FST more
+# compact.  I am rerunning in order to verify that the WER is not changed (since
+# it's possible in principle that due to edge effects related to weight-pushing,
+# the results could be a bit different).
+#  The results are inconsistently different but broadly the same.  On all of eval2000,
+#  the change k->m is 20.7->20.9 with tg LM and 18.9->18.6 after rescoring.
+#  On the train_dev data, the change is  19.3->18.9 with tg LM and 17.6->17.6 after rescoring.
+
+
+# _k is as _i but reverting the g->h change, removing the --scale-max-param-change
+# option and setting max-param-change to 1..  Using the same egs.
+
+# _i is as _h but longer egs: 150 frames instead of 75, and
+# 128 elements per minibatch instead of 256.
+
+# _h is as _g but different application of max-param-change (use --scale-max-param-change true)
+
+# _g is as _f but more splicing at last layer.
+
+# _f is as _e but with 30 as the number of left phone classes instead
+# of 10.
+
+# _e is as _d but making it more similar in configuration to _b.
+# (turns out b was better than a after all-- the egs' likelihoods had to
+# be corrected before comparing them).
+# the changes (vs. d) are: change num-pdfs target from 8k to 12k,
+# multiply learning rates by 5, and set final-layer-normalize-target to 0.5.
+
+# _d is as _c but with a modified topology (with 4 distinct states per phone
+# instead of 2), and a slightly larger num-states (8000) to compensate for the
+# different topology, which has more states.
+
+# _c is as _a but getting rid of the final-layer-normalize-target (making it 1.0
+# as the default) as it's not clear that it was helpful; using the old learning-rates;
+# and modifying the target-num-states to 7000.
+
+# _b is as as _a except for configuration changes: using 12k num-leaves instead of
+# 5k; using 5 times larger learning rate, and --final-layer-normalize-target=0.5,
+# which will make the final layer learn less fast compared with other layers.
+
+set -e
+
+# configs for 'chain'
+stage=12
+train_stage=-10
+get_egs_stage=-10
+speed_perturb=true
+dir=exp/chain/tdnn_3g  # Note: _sp will get added to this if $speed_perturb == true.
+
+# training options
+num_epochs=4
+initial_effective_lrate=0.001
+final_effective_lrate=0.0001
+leftmost_questions_truncate=-1
+max_param_change=1.0
+final_layer_normalize_target=0.5
+num_jobs_initial=3
+num_jobs_final=16
+minibatch_size=128
+frames_per_eg=150
+remove_egs=false
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+# The iVector-extraction and feature-dumping parts are the same as the standard
+# nnet3 setup, and you can skip them by setting "--stage 8" if you have already
+# run those things.
+
+suffix=
+if [ "$speed_perturb" == "true" ]; then
+  suffix=_sp
+fi
+
+dir=${dir}$suffix
+train_set=train_nodup$suffix
+ali_dir=exp/tri4_ali_nodup$suffix
+treedir=exp/chain/tri5_2y_tree$suffix
+lang=data/lang_chain_2y
+
+
+# if we are using the speed-perturbed data we need to generate
+# alignments for it.
+local/nnet3/run_ivector_common.sh --stage $stage \
+  --speed-perturb $speed_perturb \
+  --generate-alignments $speed_perturb || exit 1;
+
+
+if [ $stage -le 9 ]; then
+  # Get the alignments as lattices (gives the CTC training more freedom).
+  # use the same num-jobs as the alignments
+  nj=$(cat exp/tri4_ali_nodup$suffix/num_jobs) || exit 1;
+  steps/align_fmllr_lats.sh --nj $nj --cmd "$train_cmd" data/$train_set \
+    data/lang exp/tri4 exp/tri4_lats_nodup$suffix
+  rm exp/tri4_lats_nodup$suffix/fsts.*.gz # save space
+fi
+
+
+if [ $stage -le 10 ]; then
+  # Create a version of the lang/ directory that has one state per phone in the
+  # topo file. [note, it really has two states.. the first one is only repeated
+  # once, the second one has zero or more repeats.]
+  rm -rf $lang
+  cp -r data/lang $lang
+  silphonelist=$(cat $lang/phones/silence.csl) || exit 1;
+  nonsilphonelist=$(cat $lang/phones/nonsilence.csl) || exit 1;
+  # Use our special topology... note that later on may have to tune this
+  # topology.
+  steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >$lang/topo
+fi
+
+if [ $stage -le 11 ]; then
+  # Build a tree using our new topology.
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+      --leftmost-questions-truncate $leftmost_questions_truncate \
+      --cmd "$train_cmd" 9000 data/$train_set $lang $ali_dir $treedir
+fi
+
+if [ $stage -le 12 ]; then
+  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
+    utils/create_split_dir.pl \
+     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/swbd-$(date +'%m_%d_%H_%M')/s5c/$dir/egs/storage $dir/egs/storage
+  fi
+
+ touch $dir/egs/.nodelete # keep egs around when that run dies.
+
+ steps/nnet3/chain/train_tdnn.sh --stage $train_stage \
+    --egs-dir exp/chain/tdnn_2y_sp/egs \
+    --jesus-opts "--affine-output-dim 600 --jesus-output-dim 1800 --jesus-hidden-dim 15000" \
+    --splice-indexes "-2,-1,0,1,2 -1,2 -3,0,3:-3 -6,-3,0,3:-3 -6,-3,0,3:-3" \
+    --apply-deriv-weights false \
+    --frames-per-iter 1200000 \
+    --lm-opts "--num-extra-lm-states=2000" \
+    --get-egs-stage $get_egs_stage \
+    --minibatch-size $minibatch_size \
+    --egs-opts "--frames-overlap-per-eg 0" \
+    --frames-per-eg $frames_per_eg \
+    --num-epochs $num_epochs --num-jobs-initial $num_jobs_initial --num-jobs-final $num_jobs_final \
+    --feat-type raw \
+    --online-ivector-dir exp/nnet3/ivectors_${train_set} \
+    --cmvn-opts "--norm-means=false --norm-vars=false" \
+    --initial-effective-lrate $initial_effective_lrate --final-effective-lrate $final_effective_lrate \
+    --max-param-change $max_param_change \
+    --final-layer-normalize-target $final_layer_normalize_target \
+    --relu-dim 850 \
+    --cmd "$decode_cmd" \
+    --remove-egs $remove_egs \
+    data/${train_set}_hires $treedir exp/tri4_lats_nodup$suffix $dir  || exit 1;
+fi
+
+if [ $stage -le 13 ]; then
+  # Note: it might appear that this $lang directory is mismatched, and it is as
+  # far as the 'topo' is concerned, but this script doesn't read the 'topo' from
+  # the lang directory.
+  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_sw1_tg $dir $dir/graph_sw1_tg
+fi
+
+decode_suff=sw1_tg
+graph_dir=$dir/graph_sw1_tg
+if [ $stage -le 14 ]; then
+  for decode_set in train_dev eval2000; do
+      (
+      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+         --extra-left-context 20 \
+          --nj 50 --cmd "$decode_cmd" \
+          --online-ivector-dir exp/nnet3/ivectors_${decode_set} \
+         $graph_dir data/${decode_set}_hires $dir/decode_${decode_set}_${decode_suff} || exit 1;
+      if $has_fisher; then
+          steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" \
+            data/lang_sw1_{tg,fsh_fg} data/${decode_set}_hires \
+            $dir/decode_${decode_set}_sw1_{tg,fsh_fg} || exit 1;
+      fi
+      ) &
+  done
+fi
+wait;
+exit 0;

diff --git a/egs/wsj/s5/steps/nnet3/decode.sh b/egs/wsj/s5/steps/nnet3/decode.sh
index f4de09740ae7d14937cd28b2fa5511a73d048a4b..5379484a95f64268dafe3cbd8ab49511bf7623a9 100755 (executable)
--- a/egs/wsj/s5/steps/nnet3/decode.sh
+++ b/egs/wsj/s5/steps/nnet3/decode.sh
@@ -26,6 +26,8 @@ num_threads=1 # if >1, will use gmm-latgen-faster-parallel
 parallel_opts=  # ignored now.
 scoring_opts=
 skip_scoring=false
+extra_left_context=0
+extra_right_context=0
 feat_type=
 online_ivector_dir=
 minimize=false
@@ -146,6 +148,8 @@ if [ $stage -le 1 ]; then
   $cmd --num-threads $num_threads JOB=1:$nj $dir/log/decode.JOB.log \
     nnet3-latgen-faster$thread_string $ivector_opts $frame_subsampling_opt \
      --frames-per-chunk=$frames_per_chunk \
+     --extra-left-context=$extra_left_context \
+     --extra-right-context=$extra_right_context \
      --minimize=$minimize --max-active=$max_active --min-active=$min_active --beam=$beam \
      --lattice-beam=$lattice_beam --acoustic-scale=$acwt --allow-partial=true \
      --word-symbol-table=$graphdir/words.txt "$model" \

diff --git a/egs/wsj/s5/steps/nnet3/make_jesus_configs.py b/egs/wsj/s5/steps/nnet3/make_jesus_configs.py
index 22b8827cf6f5ea0b7bc58e504c3711393470d558..b569da3b3880754287aeecd00ec3df69f51ece59 100755 (executable)
--- a/egs/wsj/s5/steps/nnet3/make_jesus_configs.py
+++ b/egs/wsj/s5/steps/nnet3/make_jesus_configs.py
@@ -181,7 +181,7 @@ for l in range(1, num_hidden_layers + 1):
         # TODO: test where it's best to have the recurrence from- maybe the output
         # of the jesus layer itself, before the affine layer?
         if len(recurrence_array[l-1]) > 0:
-            print('component name=clip-gradient{0} dim={1} clipping-threshold={2} '
+            print('component type=ClipGradientComponent name=clip-gradient{0} dim={1} clipping-threshold={2} '
                   'norm-based-clipping=true '.format(
                     l, args.affine_output_dim, args.clipping_threshold), file=f)
             print('component-node name=clip-gradient{0} component=clip-gradient{0} '

diff --git a/src/nnet3/nnet-simple-component.cc b/src/nnet3/nnet-simple-component.cc
index bf3e523633feb7635e75bf1e07304e53c02e1693..b3900901f6d161fb13e10b206a31789593379c98 100644 (file)
--- a/src/nnet3/nnet-simple-component.cc
+++ b/src/nnet3/nnet-simple-component.cc
@@ -1084,8 +1084,8 @@ void RepeatedAffineComponent::InitFromConfig(ConfigLine *cfl) {
                "num-repeats must divide input-dim");
   KALDI_ASSERT(output_dim % num_repeats == 0 &&
                "num-repeats must divide output-dim");
-  BaseFloat param_stddev = 1.0 / std::sqrt(input_dim),
-      bias_stddev = 1.0;
+  BaseFloat param_stddev = 1.0 / std::sqrt(input_dim / num_repeats),
+      bias_stddev = 0.0;
   cfl->GetValue("param-stddev", &param_stddev);
   cfl->GetValue("bias-stddev", &bias_stddev);
   Init(learning_rate, input_dim, output_dim, num_repeats,
@@ -4246,6 +4246,15 @@ void CompositeComponent::Read(std::istream &is, bool binary) {
   ExpectToken(is, binary, ostr_end.str());
 }
 
+// virtual
+void CompositeComponent::ZeroStats() {
+  // we call ZeroStats() on all components without checking their flags; this
+  // will do nothing if the component doesn't store stats.  (components like
+  // ReLU and sigmoid and tanh store stats on activations).
+  for (size_t i = 0; i < components_.size(); i++)
+   components_[i]->ZeroStats();
+}
+
 // virtual
 void CompositeComponent::Write(std::ostream &os, bool binary) const {
   // we want this Write function to work for derived types.

diff --git a/src/nnet3/nnet-simple-component.h b/src/nnet3/nnet-simple-component.h
index 9fc090bbcfe6ea52b0c841e062d03dbb2dbcefc4..19117cdf489e11d071aef226185c545f45321d0f 100644 (file)
--- a/src/nnet3/nnet-simple-component.h
+++ b/src/nnet3/nnet-simple-component.h
@@ -1581,6 +1581,11 @@ class CompositeComponent: public UpdatableComponent {
                         Component *to_update,
                         CuMatrixBase<BaseFloat> *in_deriv) const;
 
+  // note, we don't implement StoreStats() as it would be inefficient.  Instead,
+  // by default we call StoreStats() on all members that have the flag set,
+  // inside the Backprop.
+  virtual void ZeroStats();
+
   virtual void Read(std::istream &is, bool binary);
   virtual void Write(std::ostream &os, bool binary) const;