Restore src_features for v3.0 (#2308)

* Restored src_features for v3

Author: anderleich

.github/workflows/push.yml

@@ -47,7 +47,6 @@ jobs:
           -save_data /tmp/onmt_feat \
           -src_vocab /tmp/onmt_feat.vocab.src \
           -tgt_vocab /tmp/onmt_feat.vocab.tgt \
-          -src_feats_vocab '{"feat0": "/tmp/onmt_feat.vocab.feat0"}' \
           -n_sample -1 \
           && rm -rf /tmp/sample
     - name: Test field/transform dump
@@ -259,21 +258,34 @@ jobs:
           -config data/features_data.yaml \
           -src_vocab /tmp/onmt_feat.vocab.src \
           -tgt_vocab /tmp/onmt_feat.vocab.tgt \
-          -src_feats_vocab '{"feat0": "/tmp/onmt_feat.vocab.feat0"}' \
           -src_vocab_size 1000 -tgt_vocab_size 1000 \
           -hidden_size 2 -batch_size 10 \
           -num_workers 0 -bucket_size 1024 \
           -word_vec_size 5 -hidden_size 10 \
           -report_every 5 -train_steps 10 \
           -save_model /tmp/onmt.model \
           -save_checkpoint_steps 10
+    - name: Testing training with features and dynamic scoring
+      run: |
+        python onmt/bin/train.py \
+          -config data/features_data.yaml \
+          -src_vocab /tmp/onmt_feat.vocab.src \
+          -tgt_vocab /tmp/onmt_feat.vocab.tgt \
+          -src_vocab_size 1000 -tgt_vocab_size 1000 \
+          -hidden_size 2 -batch_size 10 \
+          -word_vec_size 5 -hidden_size 10 \
+          -num_workers 0 -bucket_size 1024 \
+          -report_every 5 -train_steps 10 \
+          -train_metrics "BLEU" "TER" \
+          -valid_metrics "BLEU" "TER" \
+          -save_model /tmp/onmt.model \
+          -save_checkpoint_steps 10
     - name: Testing translation with features
       run: |
         python translate.py \
             -model /tmp/onmt.model_step_10.pt \
-            -src data/data_features/src-test.txt \
-            -src_feats "{'feat0': 'data/data_features/src-test.feat0'}" \
-            -verbose
+            -src data/data_features/src-test-with-feats.txt \
+            -n_src_feats 1 -verbose
     - name: Test RNN translation
       run: |
         head data/src-test.txt > /tmp/src-test.txt

data/data_features/src-test-with-feats.txt

@@ -0,0 +1 @@
+she￨C is￨B a￨A hard-working.￨B

data/data_features/src-test.feat0

@@ -0,0 +1,3 @@
+however,￨A according￨A to￨A the￨A logs,￨B she￨A is￨A a￨A hard-working.￨C
+however,￨A according￨B to￨C the￨D logs,￨E
+she￨C is￨B a￨A hard-working.￨B

data/data_features/src-train-with-feats.txt

@@ -0,0 +1 @@
+she￨C is￨B a￨A hard-working.￨B

data/data_features/src-train.feat0

@@ -1,11 +1,19 @@
+
 # Corpus opts:
 data:
     corpus_1:
+        path_src: data/data_features/src-train-with-feats.txt
+        path_tgt: data/data_features/tgt-train.txt
+        transforms: [inferfeats]
+    corpus_2:
         path_src: data/data_features/src-train.txt
         path_tgt: data/data_features/tgt-train.txt
-        src_feats:
-            feat0: data/data_features/src-train.feat0
-        transforms: [filterfeats, inferfeats]
+        transforms: [inferfeats]
     valid:
-        path_src: data/data_features/src-val.txt
+        path_src: data/data_features/src-val-with-feats.txt
         path_tgt: data/data_features/tgt-val.txt
+        transforms: [inferfeats]
+
+# # Feats options
+n_src_feats: 1
+src_feats_defaults: "0"

data/data_features/src-val-with-feats.txt

@@ -620,39 +620,34 @@ Training options to perform vocabulary update are:
 
 ## How can I use source word features?
 
-Extra information can be added to the words in the source sentences by defining word features. 
+Additional word-level information can be incorporated into the model by defining word features in the source sentence. 
 
-Features should be defined in a separate file using blank spaces as a separator and with each row corresponding to a source sentence. An example of the input files:
+Word features must be appended to the actual textual data by using the special character ￨ as a feature separator. For instance:
 
-data.src
 ```
-however, according to the logs, she is hard-working.
+however￨C ￭,￨N according￨L to￨L the￨L logs￨L ￭,￨N she￨L is￨L hard-working￨L ￭.￨N
 ```
 
-feat.txt
+Prior tokenization is not necessary, features will be inferred by using the `FeatInferTransform` transform if tokenization has been applied. For instace:
+
 ```
-A C C C C A A B
+SRC: however,￨C according￨L to￨L the￨L logs,￨L she￨L is￨L hard-working.￨L
+TOKENIZED SRC: however ￭, according to the logs ￭, she is hard-working ￭.
+RESULT: however￨C ￭,￨C according￨L to￨L the￨L logs￨L ￭,￨L she￨L is￨L hard￨L ￭-￭￨L working￨L ￭.￨L
 ```
 
-Prior tokenization is not necessary, features will be inferred by using the `FeatInferTransform` transform if tokenization has been applied.
+**Options**
+- `-n_src_feats`: the expected number of source features per token.
+- `-src_feats_defaults` (optional): provides default values for features. This can be really useful when mixing task specific data (with features) with general data which has not been annotated.
 
-No previous tokenization:
-```
-SRC: this is a test.
-FEATS: A A A B
-TOKENIZED SRC: this is a test ￭.
-RESULT: A A A B <null>
-```
+For the Transformer architecture make sure the following options are appropriately set:
 
-Previously tokenized:
-```
-SRC: this is a test ￭.
-FEATS: A A A B A
-RESULT: A A A B A
-```
+- `src_word_vec_size` and `tgt_word_vec_size` or `word_vec_size`
+- `feat_merge`: how to handle features vecs
+- `feat_vec_size` or maybe `feat_vec_exponent`
 
 **Notes**
-- `FilterFeatsTransform` and `FeatInferTransform` are required in order to ensure the functionality.
+- `FeatInferTransform` transform is required in order to ensure the functionality.
 - Not possible to do shared embeddings (at least with `feat_merge: concat` method)
 
 Sample config file:
@@ -662,50 +657,36 @@ data:
     dummy:
         path_src: data/train/data.src
         path_tgt: data/train/data.tgt
-        src_feats:
-            feat_0: data/train/data.src.feat_0
-            feat_1: data/train/data.src.feat_1
-        transforms: [filterfeats, onmt_tokenize, inferfeats, filtertoolong]
+        transforms: [onmt_tokenize, inferfeats, filtertoolong]
         weight: 1
     valid:
         path_src: data/valid/data.src
         path_tgt: data/valid/data.tgt
-        src_feats:
-            feat_0: data/valid/data.src.feat_0
-            feat_1: data/valid/data.src.feat_1
-        transforms: [filterfeats, onmt_tokenize, inferfeats]
+        transforms: [onmt_tokenize, inferfeats]
 
 # Transform options
 reversible_tokenization: "joiner"
-prior_tokenization: true
 
 # Vocab opts
 src_vocab: exp/data.vocab.src
 tgt_vocab: exp/data.vocab.tgt
-src_feats_vocab: 
-    feat_0: exp/data.vocab.feat_0
-    feat_1: exp/data.vocab.feat_1
+
+# Features options
+n_src_feats: 2
+src_feats_defaults: "0￨1"
 feat_merge: "sum"
 ```
 
-During inference you can pass features by using the `--src_feats` argument. `src_feats` is expected to be a Python like dict, mapping feature names with their data file.
+To allow source features in the server add the following parameters in the server's config file:
 
 ```
-{'feat_0': '../data.txt.feats0', 'feat_1': '../data.txt.feats1'}
-```
-
-**Important note!** During inference, input sentence is expected to be tokenized. Therefore feature inferring should be handled prior to running the translate command. Example:
-
-```bash
-python translate.py -model model_step_10.pt -src ../data.txt.tok -output ../data.out --src_feats "{'feat_0': '../data.txt.feats0', 'feat_1': '../data.txt.feats1'}"
+"features": {
+    "n_src_feats": 2,
+    "src_feats_defaults": "0￨1",
+    "reversible_tokenization": "joiner"
+}
 ```
 
-When using the Transformer architecture make sure the following options are appropriately set:
-
-- `src_word_vec_size` and `tgt_word_vec_size` or `word_vec_size`
-- `feat_merge`: how to handle features vecs
-- `feat_vec_size` and maybe `feat_vec_exponent`
-
 ## How can I set up a translation server ?
 A REST server was implemented to serve OpenNMT-py models. A discussion is opened on the OpenNMT forum: [discussion link](https://forum.opennmt.net/t/simple-opennmt-py-rest-server/1392).
 

data/data_features/src-val.feat0

@@ -6,10 +6,10 @@
 from onmt.utils.parse import ArgumentParser
 from onmt.opts import dynamic_prepare_opts
 from onmt.inputters.text_corpus import build_corpora_iters, get_corpora
-from onmt.inputters.text_utils import process
+from onmt.inputters.text_utils import process, append_features_to_text
 from onmt.transforms import make_transforms, get_transforms_cls
 from onmt.constants import CorpusName, CorpusTask
-from collections import Counter, defaultdict
+from collections import Counter
 import multiprocessing as mp
 
 
@@ -40,21 +40,11 @@ def write_files_from_queues(sample_path, queues):
                     break
 
 
-# Just for debugging purposes
-# It appends features to subwords when dumping to file
-def append_features_to_example(example, features):
-    ex_toks = example.split(' ')
-    feat_toks = features.split(' ')
-    toks = [f"{subword}￨{feat}" for subword, feat in
-            zip(ex_toks, feat_toks)]
-    return " ".join(toks)
-
-
 def build_sub_vocab(corpora, transforms, opts, n_sample, stride, offset):
     """Build vocab on (strided) subpart of the data."""
     sub_counter_src = Counter()
     sub_counter_tgt = Counter()
-    sub_counter_src_feats = defaultdict(Counter)
+    sub_counter_src_feats = [Counter() for _ in range(opts.n_src_feats)]
     datasets_iterables = build_corpora_iters(
         corpora, transforms, opts.data,
         skip_empty_level=opts.skip_empty_level,
@@ -70,19 +60,22 @@ def build_sub_vocab(corpora, transforms, opts, n_sample, stride, offset):
                 continue
             src_line, tgt_line = (maybe_example['src']['src'],
                                   maybe_example['tgt']['tgt'])
-            src_line_pretty = src_line
-            for feat_name, feat_line in maybe_example["src"].items():
-                if feat_name not in ["src", "src_original"]:
-                    sub_counter_src_feats[feat_name].update(
-                        feat_line.split(' '))
-                    if opts.dump_samples:
-                        src_line_pretty = append_features_to_example(
-                            src_line_pretty, feat_line)
             sub_counter_src.update(src_line.split(' '))
             sub_counter_tgt.update(tgt_line.split(' '))
+
+            if 'feats' in maybe_example['src']:
+                src_feats_lines = maybe_example['src']['feats']
+                for i in range(opts.n_src_feats):
+                    sub_counter_src_feats[i].update(
+                        src_feats_lines[i].split(' '))
+            else:
+                src_feats_lines = []
+
             if opts.dump_samples:
+                src_pretty_line = append_features_to_text(
+                    src_line, src_feats_lines)
                 build_sub_vocab.queues[c_name][offset].put(
-                    (i, src_line_pretty, tgt_line))
+                    (i, src_pretty_line, tgt_line))
             if n_sample > 0 and ((i+1) * stride + offset) >= n_sample:
                 if opts.dump_samples:
                     build_sub_vocab.queues[c_name][offset].put("break")
@@ -113,7 +106,7 @@ def build_vocab(opts, transforms, n_sample=3):
     corpora = get_corpora(opts, task=CorpusTask.TRAIN)
     counter_src = Counter()
     counter_tgt = Counter()
-    counter_src_feats = defaultdict(Counter)
+    counter_src_feats = [Counter() for _ in range(opts.n_src_feats)]
     from functools import partial
     queues = {c_name: [mp.Queue(opts.vocab_sample_queue_size)
                        for i in range(opts.num_threads)]
@@ -134,7 +127,8 @@ def build_vocab(opts, transforms, n_sample=3):
                 func, range(0, opts.num_threads)):
             counter_src.update(sub_counter_src)
             counter_tgt.update(sub_counter_tgt)
-            counter_src_feats.update(sub_counter_src_feats)
+            for i in range(opts.n_src_feats):
+                counter_src_feats[i].update(sub_counter_src_feats[i])
     if opts.dump_samples:
         write_process.join()
     return counter_src, counter_tgt, counter_src_feats
@@ -166,10 +160,10 @@ def build_vocab_main(opts):
     src_counter, tgt_counter, src_feats_counter = build_vocab(
         opts, transforms, n_sample=opts.n_sample)
 
-    logger.info(f"Counters src:{len(src_counter)}")
-    logger.info(f"Counters tgt:{len(tgt_counter)}")
-    for feat_name, feat_counter in src_feats_counter.items():
-        logger.info(f"Counters {feat_name}:{len(feat_counter)}")
+    logger.info(f"Counters src: {len(src_counter)}")
+    logger.info(f"Counters tgt: {len(tgt_counter)}")
+    for i, feat_counter in enumerate(src_feats_counter):
+        logger.info(f"Counters src feat_{i}: {len(feat_counter)}")
 
     def save_counter(counter, save_path):
         check_path(save_path, exist_ok=opts.overwrite, log=logger.warning)
@@ -186,8 +180,8 @@ def save_counter(counter, save_path):
         save_counter(src_counter, opts.src_vocab)
         save_counter(tgt_counter, opts.tgt_vocab)
 
-    for k, v in src_feats_counter.items():
-        save_counter(v, opts.src_feats_vocab[k])
+    for i, c in enumerate(src_feats_counter):
+        save_counter(c, f"{opts.src_vocab}_feat{i}")
 
 
 def _get_parser():

data/features_data.yaml

@@ -34,11 +34,10 @@ def build_vocab(opt, specials):
     """ Build vocabs dict to be stored in the checkpoint
         based on vocab files having each line [token, count]
     Args:
-        opt: src_vocab, tgt_vocab, src_feats_vocab
+        opt: src_vocab, tgt_vocab, n_src_feats
     Return:
         vocabs: {'src': pyonmttok.Vocab, 'tgt': pyonmttok.Vocab,
-                 'src_feats' : {'feat0': pyonmttok.Vocab,
-                                'feat1': pyonmttok.Vocab, ...},
+                 'src_feats' : [pyonmttok.Vocab, ...]},
                  'data_task': seq2seq or lm
                 }
     """
@@ -85,10 +84,10 @@ def _pad_vocab_to_multiple(vocab, multiple):
                                                opt.vocab_size_multiple)
         vocabs['tgt'] = tgt_vocab
 
-    if opt.src_feats_vocab:
-        src_feats = {}
-        for feat_name, filepath in opt.src_feats_vocab.items():
-            src_f_vocab = _read_vocab_file(filepath, 1)
+    if opt.n_src_feats > 0:
+        src_feats_vocabs = []
+        for i in range(opt.n_src_feats):
+            src_f_vocab = _read_vocab_file(f"{opt.src_vocab}_feat{i}", 1)
             src_f_vocab = pyonmttok.build_vocab_from_tokens(
                 src_f_vocab,
                 maximum_size=0,
@@ -101,8 +100,8 @@ def _pad_vocab_to_multiple(vocab, multiple):
             if opt.vocab_size_multiple > 1:
                 src_f_vocab = _pad_vocab_to_multiple(src_f_vocab,
                                                      opt.vocab_size_multiple)
-            src_feats[feat_name] = src_f_vocab
-        vocabs['src_feats'] = src_feats
+            src_feats_vocabs.append(src_f_vocab)
+        vocabs["src_feats"] = src_feats_vocabs
 
     vocabs['data_task'] = opt.data_task
 
@@ -146,10 +145,8 @@ def vocabs_to_dict(vocabs):
     vocabs_dict['src'] = vocabs['src'].ids_to_tokens
     vocabs_dict['tgt'] = vocabs['tgt'].ids_to_tokens
     if 'src_feats' in vocabs.keys():
-        vocabs_dict['src_feats'] = {}
-        for feat in vocabs['src_feats'].keys():
-            vocabs_dict['src_feats'][feat] = \
-                vocabs['src_feats'][feat].ids_to_tokens
+        vocabs_dict['src_feats'] = [feat_vocab.ids_to_tokens
+                                    for feat_vocab in vocabs['src_feats']]
     vocabs_dict['data_task'] = vocabs['data_task']
     return vocabs_dict
 
@@ -167,9 +164,8 @@ def dict_to_vocabs(vocabs_dict):
     else:
         vocabs['tgt'] = pyonmttok.build_vocab_from_tokens(vocabs_dict['tgt'])
     if 'src_feats' in vocabs_dict.keys():
-        vocabs['src_feats'] = {}
-        for feat in vocabs_dict['src_feats'].keys():
-            vocabs['src_feats'][feat] = \
-                pyonmttok.build_vocab_from_tokens(
-                    vocabs_dict['src_feats'][feat])
+        vocabs['src_feats'] = []
+        for feat_vocab in vocabs_dict['src_feats']:
+            vocabs['src_feats'].append(
+                pyonmttok.build_vocab_from_tokens(feat_vocab))
     return vocabs