rbbt-dm 1.1.55 → 1.1.58

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: da7b980af0b37e0ef665bc87eded5c966af3f9df1a58f8aa2ccec8fc2e83b980
-  data.tar.gz: 7a35ffea3a0179fd915f3bbdd4620ead3915fb18a965df54fe555df6305bf30a
+  metadata.gz: 286385d90b276d30cd5e1b21ae38c5e6a203e2ce3ac10673c434c19a2f45cfb1
+  data.tar.gz: 7879d74a364886ea8cb507be51c4979cfb598bdb273f948c3c3930a5dce199e6
 SHA512:
-  metadata.gz: ce14071303cb8e80419d1d29bdb4ab3b0e7740e579bd5051a7d70392bb833cc47c96d4b1ae76d95dcb5142055486c552df5d63b25c381bc9168a71c788c06ecc
-  data.tar.gz: eeae14bbaad12c7131ad6ffa1b44fa479a920235642fcb29b965e879f19b524b2a7446db35fe1f95cd60d23254d8d5cc89e921924ae4184233d380e0c0f443f4
+  metadata.gz: b82c77bd736c8422e49c4dc83b63d6a91da6e76857af4b5cf5aff0a9a58b4147bc50b49b1b1534e8b07ca1bce5f6a5a673c5d688fb8cd7856623370d19fd1bda
+  data.tar.gz: 1b267a85ab600b878e99f414f725255cf086165a27f8cdec42ed83349b4f36bdb9e29615e0aaada9b30f098df8382e4778cebaf2b8649e17b8985e79d9b8bd23

data/LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2010-2013 Miguel Vázquez García
+Copyright (c) 2010-2022 Miguel Vázquez García
 
 Permission is hereby granted, free of charge, to any person obtaining
 a copy of this software and associated documentation files (the

data/lib/rbbt/vector/model/random_forest.rb

@@ -16,11 +16,21 @@ model = randomForest(as.factor(label) ~ ., data = features);
 rbbt.require("randomForest");
 pred = names(model$forest$xlevels)
 for (p in pred) { 
-  if (class(features[[p]]) == "factor") { 
+  if (is.factor(features[[p]])) { 
       features[[p]] = factor(features[[p]], levels=model$forest$xlevels[[p]])
     } 
 }
 label = predict(model, features);
     EOF
   end
+
+  def importance
+    TmpFile.with_file do |tmp|
+      tsv = R.run <<-EOF
+load(file="#{model_file}");
+rbbt.tsv.write('#{tmp}', model$importance)
+      EOF
+      TSV.open(tmp)
+    end
+  end
 end

data/lib/rbbt/vector/model/spaCy.rb

@@ -4,13 +4,13 @@ require 'rbbt/nlp/spaCy'
 class SpaCyModel < VectorModel
   attr_accessor :config
 
-  def spacy(&block)
+  def self.spacy(&block)
     RbbtPython.run "spacy" do 
       RbbtPython.module_eval(&block)
     end
   end
   
-  def initialize(dir, config, lang = 'en_core_web_md')
+  def initialize(dir, config, categories = %w(positive negative), lang = 'en_core_web_md')
     @config = case
               when Path === config
                 config.read
@@ -30,20 +30,19 @@ class SpaCyModel < VectorModel
     @train_model = Proc.new do |file, features, labels|
       texts = features
       docs = []
+      unique_labels = labels.uniq
       tmpconfig = File.join(file, 'config')
       tmptrain = File.join(file, 'train.spacy')
       SpaCy.config(@config, tmpconfig)
-      spacy do
+      SpaCyModel.spacy do
         nlp = SpaCy.nlp(lang)
         docs = []
         RbbtPython.iterate nlp.pipe(texts.zip(labels), as_tuples: true), :bar => "Training documents into spacy format" do |doc,label|
-          if %w(1 true pos).include?(label.to_s.downcase) 
-            doc.cats["positive"] = 1
-            doc.cats["negative"] = 0
-          else
-            doc.cats["positive"] = 0
-            doc.cats["negative"] = 1
+          unique_labels.each do |other_label|
+            next if other_label == label
+            doc.cats[other_label] = false
           end
+          doc.cats[label] = true
           docs << doc
         end
 
@@ -51,22 +50,23 @@ class SpaCyModel < VectorModel
         doc_bin.to_disk(tmptrain)
       end
 
-      gpu = Rbbt::Config.get('gpu_id', :spacy, :spacy_train)
+      gpu = Rbbt::Config.get('gpu_id', :spacy, :spacy_train, :default => 0)
       CMD.cmd_log(:spacy, "train #{tmpconfig} --output #{file} --paths.train #{tmptrain} --paths.dev #{tmptrain}",  "--gpu-id" => gpu)
     end
  
-    @eval_model = Proc.new do |file, features|
+    @eval_model = Proc.new do |file, features,list|
       texts = features
+      texts = [texts] unless list
 
       docs = []
-      spacy do
+      SpaCyModel.spacy do
         nlp = spacy.load("#{file}/model-best")
 
         Log::ProgressBar.with_bar texts.length, :desc => "Evaluating documents" do |bar|
           texts.collect do |text|
             cats = nlp.(text).cats
             bar.tick
-            cats['positive'] > cats['negative']  ? 1 : 0
+            cats.sort_by{|l,v| v.to_f }.last.first
           end
         end
       end

data/lib/rbbt/vector/model/svm.rb

@@ -3,16 +3,16 @@ class SVMModel < VectorModel
   def initialize(dir)
     super(dir)
 
-    @extract_features = Proc.new{|element|
+    @extract_features ||= Proc.new{|element|
       element
     }
 
-    @train_model =<<-EOF
+    @train_model ||=<<-EOF
 rbbt.require('e1071');
 model = svm(as.factor(label) ~ ., data = features);
     EOF
  
-    @eval_model =<<-EOF
+    @eval_model ||=<<-EOF
 rbbt.require('e1071');
 label = predict(model, features);
     EOF

data/lib/rbbt/vector/model.rb

@@ -53,6 +53,13 @@ features = cbind(features, label = labels);
     "features[['#{name}']] = factor(features[['#{name}']], levels=#{R.ruby2R levels})"
   end * "\n" if factor_levels }
 #{code}
+# Save used factor levels
+factor_levels = c()
+for (c in names(features)){
+  if (is.factor(features[[c]]))
+    factor_levels[c] = paste(levels(features[[c]]), collapse="\t")
+}
+rbbt.tsv.write("#{model_file}.factor_levels", factor_levels, names=c('Levels'), type='flat')
 save(model, file='#{model_file}')
       EOF
     end
@@ -150,6 +157,9 @@ cat(paste(label, sep="\\n", collapse="\\n"));
       if File.exists?(@levels_file)
         @factor_levels = YAML.load(Open.read(@levels_file))
       end
+      if File.exists?(@model_file + '.factor_levels')
+        @factor_levels = TSV.open(@model_file + '.factor_levels')
+      end
     else
       @factor_levels = factor_levels 
     end
@@ -294,15 +304,16 @@ cat(paste(label, sep="\\n", collapse="\\n"));
       good_label = labels.uniq.select{|l| l.to_s == "true"}.first if good_label.nil?
       good_label = labels.uniq.select{|l| l.to_s == "1"}.first if good_label.nil?
       good_label = labels.uniq.sort.first if good_label.nil?
+      good_label = good_label.to_s
 
       test.zip(predicted).each do |gs,pred|
         gs = gs.to_s
         pred = pred.to_s
 
-        tp += 1 if gs == good_label && gs == good_label
-        tn += 1 if gs != good_label && gs != good_label 
-        fp += 1 if gs != good_label && pred == good_label
-        fn += 1 if gs == good_label && pred != good_label
+        tp += 1 if pred == good_label && gs == good_label
+        fp += 1 if pred == good_label && gs != good_label
+        tn += 1 if pred != good_label && gs != good_label 
+        fn += 1 if pred != good_label && gs == good_label
       end
 
       p = tp + fn
@@ -319,6 +330,8 @@ cat(paste(label, sep="\\n", collapse="\\n"));
       acc = []
       labels.each do |good_label|
         values = VectorModel.f1_metrics(test, predicted, good_label)
+        tp, tn, fp, fn, pr, re, f1 = values
+        Log.debug "Partial CV #{good_label} - P:#{"%.3f" % pr} R:#{"%.3f" % re} F1:#{"%.3f" % f1} - #{[tp.to_s, tn.to_s, fp.to_s, fn.to_s] * " "}"
         acc << values
       end
       Misc.zip_fields(acc).collect{|s| Misc.mean(s)}
@@ -339,12 +352,21 @@ cat(paste(label, sep="\\n", collapse="\\n"));
     end
 
     begin
-      feature_folds = Misc.divide(@features, folds)
-      labels_folds = Misc.divide(@labels, folds)
+      if folds == 1
+        feature_folds = [@features]
+        labels_folds = [@labels]
+      else
+        feature_folds = Misc.divide(@features, folds)
+        labels_folds = Misc.divide(@labels, folds)
+      end
 
       folds.times do |fix|
 
-        rest = (0..(folds-1)).to_a - [fix]
+        if folds == 1
+          rest = [fix]
+        else
+          rest = (0..(folds-1)).to_a - [fix]
+        end
 
         test_set = feature_folds[fix]
         train_set = feature_folds.values_at(*rest).inject([]){|acc,e| acc += e; acc}
@@ -354,6 +376,7 @@ cat(paste(label, sep="\\n", collapse="\\n"));
 
         @features = train_set
         @labels = train_labels
+
         self.train
         predictions = self.eval_list test_set, false
 
@@ -361,6 +384,8 @@ cat(paste(label, sep="\\n", collapse="\\n"));
 
         different_labels = test_labels.uniq
 
+        Log.debug do "Accuracy Fold #{fix}: #{(100 * test_labels.zip(predictions).select{|t,p| t == p }.length.to_f / test_labels.length).round(2)}%"  end
+
         tp, tn, fp, fn, pr, re, f1 = VectorModel.f1_metrics(test_labels, predictions, good_label)
 
         if multiclass 
@@ -376,7 +401,7 @@ cat(paste(label, sep="\\n", collapse="\\n"));
       @features = orig_features
       @labels = orig_labels
     end
-    self.train
+    self.train unless folds == 1
     res
   end
 end

data/share/spaCy/cpu/textcat_multilabel_accuracy.conf

@@ -0,0 +1,86 @@
+# This is an auto-generated partial config. To use it with 'spacy train'
+# you can run spacy init fill-config to auto-fill all default settings:
+# python -m spacy init fill-config ./base_config.cfg ./config.cfg
+[paths]
+train = null
+dev = null
+
+[system]
+gpu_allocator = null
+
+[nlp]
+lang = "en"
+pipeline = ["tok2vec","textcat_multilabel"]
+batch_size = 1000
+
+[components]
+
+[components.tok2vec]
+factory = "tok2vec"
+
+[components.tok2vec.model]
+@architectures = "spacy.Tok2Vec.v2"
+
+[components.tok2vec.model.embed]
+@architectures = "spacy.MultiHashEmbed.v2"
+width = ${components.tok2vec.model.encode.width}
+attrs = ["ORTH", "SHAPE"]
+rows = [5000, 2500]
+include_static_vectors = true
+
+[components.tok2vec.model.encode]
+@architectures = "spacy.MaxoutWindowEncoder.v2"
+width = 256
+depth = 8
+window_size = 1
+maxout_pieces = 3
+
+[components.textcat_multilabel]
+factory = "textcat_multilabel"
+
+[components.textcat_multilabel.model]
+@architectures = "spacy.TextCatEnsemble.v2"
+nO = null
+
+[components.textcat_multilabel.model.tok2vec]
+@architectures = "spacy.Tok2VecListener.v1"
+width = ${components.tok2vec.model.encode.width}
+
+[components.textcat_multilabel.model.linear_model]
+@architectures = "spacy.TextCatBOW.v1"
+exclusive_classes = true
+ngram_size = 1
+no_output_layer = false
+
+[corpora]
+
+[corpora.train]
+@readers = "spacy.Corpus.v1"
+path = ${paths.train}
+max_length = 2000
+
+[corpora.dev]
+@readers = "spacy.Corpus.v1"
+path = ${paths.dev}
+max_length = 0
+
+[training]
+dev_corpus = "corpora.dev"
+train_corpus = "corpora.train"
+
+[training.optimizer]
+@optimizers = "Adam.v1"
+
+[training.batcher]
+@batchers = "spacy.batch_by_words.v1"
+discard_oversize = false
+tolerance = 0.2
+
+[training.batcher.size]
+@schedules = "compounding.v1"
+start = 100
+stop = 1000
+compound = 1.001
+
+[initialize]
+vectors = "en_core_web_lg"

data/share/spaCy/cpu/textcat_multilabel_efficiency.conf

@@ -0,0 +1,78 @@
+# This is an auto-generated partial config. To use it with 'spacy train'
+# you can run spacy init fill-config to auto-fill all default settings:
+# python -m spacy init fill-config ./base_config.cfg ./config.cfg
+[paths]
+train = null
+dev = null
+
+[system]
+gpu_allocator = null
+
+[nlp]
+lang = "en"
+pipeline = ["tok2vec","textcat_multilabel"]
+batch_size = 1000
+
+[components]
+
+[components.tok2vec]
+factory = "tok2vec"
+
+[components.tok2vec.model]
+@architectures = "spacy.Tok2Vec.v2"
+
+[components.tok2vec.model.embed]
+@architectures = "spacy.MultiHashEmbed.v2"
+width = ${components.tok2vec.model.encode.width}
+attrs = ["ORTH", "SHAPE"]
+rows = [5000, 2500]
+include_static_vectors = false
+
+[components.tok2vec.model.encode]
+@architectures = "spacy.MaxoutWindowEncoder.v2"
+width = 96
+depth = 4
+window_size = 1
+maxout_pieces = 3
+
+[components.textcat_multilabel]
+factory = "textcat_multilabel"
+
+[components.textcat_multilabel.model]
+@architectures = "spacy.TextCatBOW.v1"
+exclusive_classes = true
+ngram_size = 1
+no_output_layer = false
+
+[corpora]
+
+[corpora.train]
+@readers = "spacy.Corpus.v1"
+path = ${paths.train}
+max_length = 2000
+
+[corpora.dev]
+@readers = "spacy.Corpus.v1"
+path = ${paths.dev}
+max_length = 0
+
+[training]
+dev_corpus = "corpora.dev"
+train_corpus = "corpora.train"
+
+[training.optimizer]
+@optimizers = "Adam.v1"
+
+[training.batcher]
+@batchers = "spacy.batch_by_words.v1"
+discard_oversize = false
+tolerance = 0.2
+
+[training.batcher.size]
+@schedules = "compounding.v1"
+start = 100
+stop = 1000
+compound = 1.001
+
+[initialize]
+vectors = null

data/share/spaCy/gpu/textcat_multilabel_accuracy.conf

@@ -0,0 +1,84 @@
+# This is an auto-generated partial config. To use it with 'spacy train'
+# you can run spacy init fill-config to auto-fill all default settings:
+# python -m spacy init fill-config ./base_config.cfg ./config.cfg
+[paths]
+train = null
+dev = null
+
+[system]
+gpu_allocator = "pytorch"
+
+[nlp]
+lang = "en"
+pipeline = ["transformer","textcat_multilabel"]
+batch_size = 128
+
+[components]
+
+[components.transformer]
+factory = "transformer"
+
+[components.transformer.model]
+@architectures = "spacy-transformers.TransformerModel.v1"
+name = "emilyalsentzer/Bio_ClinicalBERT"
+tokenizer_config = {"use_fast": true}
+
+[components.transformer.model.get_spans]
+@span_getters = "spacy-transformers.strided_spans.v1"
+window = 128
+stride = 96
+
+[components.textcat_multilabel]
+factory = "textcat_multilabel"
+
+[components.textcat_multilabel.model]
+@architectures = "spacy.TextCatEnsemble.v2"
+nO = null
+
+[components.textcat_multilabel.model.tok2vec]
+@architectures = "spacy-transformers.TransformerListener.v1"
+grad_factor = 1.0
+
+[components.textcat_multilabel.model.tok2vec.pooling]
+@layers = "reduce_mean.v1"
+
+[components.textcat_multilabel.model.linear_model]
+@architectures = "spacy.TextCatBOW.v1"
+exclusive_classes = true
+ngram_size = 1
+no_output_layer = false
+
+[corpora]
+
+[corpora.train]
+@readers = "spacy.Corpus.v1"
+path = ${paths.train}
+max_length = 500
+
+[corpora.dev]
+@readers = "spacy.Corpus.v1"
+path = ${paths.dev}
+max_length = 0
+
+[training]
+accumulate_gradient = 3
+dev_corpus = "corpora.dev"
+train_corpus = "corpora.train"
+
+[training.optimizer]
+@optimizers = "Adam.v1"
+
+[training.optimizer.learn_rate]
+@schedules = "warmup_linear.v1"
+warmup_steps = 250
+total_steps = 20000
+initial_rate = 5e-5
+
+[training.batcher]
+@batchers = "spacy.batch_by_padded.v1"
+discard_oversize = true
+size = 2000
+buffer = 256
+
+[initialize]
+vectors = null

data/share/spaCy/gpu/textcat_multilabel_efficiency.conf

@@ -0,0 +1,73 @@
+# This is an auto-generated partial config. To use it with 'spacy train'
+# you can run spacy init fill-config to auto-fill all default settings:
+# python -m spacy init fill-config ./base_config.cfg ./config.cfg
+[paths]
+train = null
+dev = null
+
+[system]
+gpu_allocator = "pytorch"
+
+[nlp]
+lang = "en"
+pipeline = ["transformer","textcat_multilabel"]
+batch_size = 128
+
+[components]
+
+[components.transformer]
+factory = "transformer"
+
+[components.transformer.model]
+@architectures = "spacy-transformers.TransformerModel.v1"
+name = "roberta-base"
+tokenizer_config = {"use_fast": true}
+
+[components.transformer.model.get_spans]
+@span_getters = "spacy-transformers.strided_spans.v1"
+window = 128
+stride = 96
+
+[components.textcat_multilabel]
+factory = "textcat_multilabel"
+
+[components.textcat_multilabel.model]
+@architectures = "spacy.TextCatBOW.v1"
+exclusive_classes = true
+ngram_size = 1
+no_output_layer = false
+
+[corpora]
+
+[corpora.train]
+@readers = "spacy.Corpus.v1"
+path = ${paths.train}
+max_length = 500
+
+[corpora.dev]
+@readers = "spacy.Corpus.v1"
+path = ${paths.dev}
+max_length = 0
+
+[training]
+accumulate_gradient = 3
+dev_corpus = "corpora.dev"
+train_corpus = "corpora.train"
+
+[training.optimizer]
+@optimizers = "Adam.v1"
+
+[training.optimizer.learn_rate]
+@schedules = "warmup_linear.v1"
+warmup_steps = 250
+total_steps = 20000
+initial_rate = 5e-5
+
+[training.batcher]
+@batchers = "spacy.batch_by_padded.v1"
+discard_oversize = true
+size = 2000
+buffer = 256
+
+[initialize]
+vectors = null

data/test/rbbt/vector/model/test_spaCy.rb

@@ -23,18 +23,23 @@ class TestSpaCyModel < Test::Unit::TestCase
       good = tsv.select("Recommended IND" => '1')
       bad = tsv.select("Recommended IND" => '0')
 
-      gsize = 2000
-      bsize = 500
+      gsize = 200
+      bsize = 50
       good.keys[0..gsize-1].each do |text|
         next if text.nil? || text.empty?
-        model.add text, '1'
+        model.add text, 'good'
       end
 
       bad.keys[0..bsize-1].each do |text|
-        model.add text, '0'
+        model.add text, 'bad'
       end
 
-      model.cross_validation
+      model.cross_validation 1
+
+      model = VectorModel.new dir
+
+      assert Misc.counts(model.eval_list(good.keys[0..50]))['good'] > 40
+      assert Misc.counts(model.eval_list(bad.keys[0..50]))['bad'] > 40
     end
 
     def test_svm_spacy

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: rbbt-dm
 version: !ruby/object:Gem::Version
-  version: 1.1.55
+  version: 1.1.58
 platform: ruby
 authors:
 - Miguel Vazquez
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2021-12-20 00:00:00.000000000 Z
+date: 2022-07-19 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: rbbt-util
@@ -116,8 +116,12 @@ files:
 - share/R/heatmap.3.R
 - share/spaCy/cpu/textcat_accuracy.conf
 - share/spaCy/cpu/textcat_efficiency.conf
+- share/spaCy/cpu/textcat_multilabel_accuracy.conf
+- share/spaCy/cpu/textcat_multilabel_efficiency.conf
 - share/spaCy/gpu/textcat_accuracy.conf
 - share/spaCy/gpu/textcat_efficiency.conf
+- share/spaCy/gpu/textcat_multilabel_accuracy.conf
+- share/spaCy/gpu/textcat_multilabel_efficiency.conf
 - test/rbbt/matrix/test_barcode.rb
 - test/rbbt/network/test_paths.rb
 - test/rbbt/statistics/test_fdr.rb