Upload 33 files

version1/baseline_experiments/yaml_files/yaml_original/syl.transformer.yaml

@@ -0,0 +1,80 @@
+# syl.transformer.yaml
+
+## Where the samples will be written
+save_data: ./transformer_syl
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+## Where the vocab(s) will be written
+src_vocab: ./transformer_syl/transformer_syl.vocab.src
+tgt_vocab: ./transformer_syl/transformer_syl.vocab.tgt
+vocab_size_multiple: 8
+src_words_min_frequency: 1
+tgt_words_min_frequency: 1
+share_vocab: True
+n_sample: 0
+
+#### Filter
+src_seq_length: 200
+tgt_seq_length: 200
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/myPOS_transformer_syl.src.txt
+        path_tgt: ./parallel_pseudo/sT/myPOS_transformer_syl.tgt.txt
+    valid:
+        path_src: ./parallel_pseudo/original/syl_valid.src.txt
+        path_tgt: ./parallel_pseudo/original/syl_valid.tgt.txt
+
+
+# Training
+
+save_model: ./transformer_syl/transformer_syl
+
+log_file: ./transformer_syl/log_3k_transformer_syl_ST.txt
+
+# Stop training if it does not imporve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+decoder_type: transformer
+encoder_type: transformer
+word_vec_size: 512
+hidden_size: 512
+enc_layers: 6
+dec_layers: 6
+hidden_size: 512
+word_vec_size: 512
+transformer_ff: 512
+heads: 8
+
+accum_count: 4
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+batch_type: tokens
+normalization: tokens
+dropout_steps: [0]
+dropout: [0.1]
+attention_dropout: [0.1]
+position_encoding: true
+label_smoothing: 0.1
+
+max_generator_batches: 2
+
+param_init: 0.0
+param_init_glorot: 'true'
+position_encoding: 'true'
+
+world_size: 1
+## Run with GPU no. zero
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000

version1/baseline_experiments/yaml_files/yaml_original/syl_s2s.yaml

@@ -0,0 +1,62 @@
+# syl_s2s.yaml
+## Where the samples will be written
+
+# Vocab Building
+## Where the samples will be written
+save_data: ./s2s_syl
+
+## Where the vocab(s) will be written
+src_vocab: ./s2s_syl/syl_s2s.vocab.src
+tgt_vocab: ./s2s_syl/syl_s2s.vocab.tgt
+share_vocab: True
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/s2s_syl.src.txt
+        path_tgt: ./parallel_pseudo/sT/s2s_syl.tgt.txt
+    valid:
+        path_src: ./parallel_pseudo/original/syl_valid.src.txt
+        path_tgt: ./parallel_pseudo/original/syl_valid.tgt.txt
+
+# Increase sequence length
+src_seq_length: 200  
+tgt_seq_length: 200  
+
+# Training
+
+save_model: ./s2s_syl/s2s_syl
+
+log_file: ./s2s_syl/log_3k_s2s_syl_ST.txt
+
+# Stop training if it does not imporve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 10000
+
+# To save space, limit checkpoints to last n
+keep_checkpoint: 5
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+dropout: [0.3]  # LSTM models often use higher dropout rates
+
+# Train on a single GPU
+world_size: 1
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000
+
+# Model parameters for RNN LSTM
+encoder_type: brnn
+decoder_type: rnn
+rnn_type: LSTM
+rnn_size: 512  # Adjust this based on your model size and complexity
+rnn_num_layers: 2  # Adjust this based on model architecture

version1/baseline_experiments/yaml_files/yaml_original/word.transformer.yaml

@@ -0,0 +1,78 @@
+# word.transformer.yaml
+
+## Where the samples will be written
+save_data: ./word_transformer
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+## Where the vocab(s) will be written
+src_vocab: ./word_transformer/word_transformer.vocab.src
+tgt_vocab: ./word_transformer/word_transformer.vocab.tgt
+vocab_size_multiple: 8
+src_words_min_frequency: 1
+tgt_words_min_frequency: 1
+share_vocab: True
+n_sample: 0
+
+#### Filter
+src_seq_length: 200
+tgt_seq_length: 200
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/myPOS_transformer_word.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/myPOS_transformer_word.train.tgt.txt
+    valid:
+        path_src: ./parallel_pseudo/original/word_valid.src.txt
+        path_tgt: ./parallel_pseudo/original/word_valid.tgt.txt
+
+# Model Configuration
+
+save_model: ./word_transformer/word_transformer
+log_file: ./word_transformer/log_3k_word_transformer_ST.txt
+
+# Stop training if it does not imporve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+decoder_type: transformer
+encoder_type: transformer
+word_vec_size: 512
+hidden_size: 512
+enc_layers: 6
+dec_layers: 6
+hidden_size: 512
+word_vec_size: 512
+transformer_ff: 512
+heads: 8
+
+accum_count: 4
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+batch_type: tokens
+normalization: tokens
+dropout_steps: [0]
+dropout: [0.1]
+attention_dropout: [0.1]
+position_encoding: true
+label_smoothing: 0.1
+
+max_generator_batches: 2
+
+param_init: 0.0
+param_init_glorot: 'true'
+position_encoding: 'true'
+
+world_size: 1
+## Run with GPU no. zero
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000

version1/baseline_experiments/yaml_files/yaml_original/word_s2s.yaml

@@ -0,0 +1,62 @@
+# syl_s2s.yaml
+## Where the samples will be written
+
+# Vocab Building
+## Where the samples will be written
+save_data: ./s2s_word
+
+## Where the vocab(s) will be written
+src_vocab: ./s2s_word/word_s2s.vocab.src
+tgt_vocab: ./s2s_word/word_s2s.vocab.tgt
+share_vocab: True
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/s2s_word.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/s2s_word.train.tgt.txt
+    valid:
+        path_src: ./parallel_pseudo/original/word_valid.src.txt
+        path_tgt: ./parallel_pseudo/original/word_valid.tgt.txt
+
+# Increase sequence length
+src_seq_length: 200  
+tgt_seq_length: 200  
+
+# Training
+
+save_model: ./s2s_word/s2s_word
+
+log_file: ./s2s_word/log_3k_s2s_word_ST.txt
+
+# Stop training if it does not imporve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 10000
+
+# To save space, limit checkpoints to last n
+keep_checkpoint: 5
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+dropout: [0.3]  # LSTM models often use higher dropout rates
+
+# Train on a single GPU
+world_size: 1
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000
+
+# Model parameters for RNN LSTM
+encoder_type: brnn
+decoder_type: rnn
+rnn_type: LSTM
+rnn_size: 512  # Adjust this based on your model size and complexity
+rnn_num_layers: 2  # Adjust this based on model architecture

version1/baseline_experiments/yaml_files/yaml_pos/syl.transformer.yaml

@@ -0,0 +1,95 @@
+# syl.transformer.yaml
+
+## Where the samples will be written
+save_data: ./syl_transformer_final2
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+## Where the vocab(s) will be written
+src_vocab: ./syl_transformer_final2/syl_transformer.vocab.src
+tgt_vocab: ./syl_transformer_final2/syl_transformer.vocab.tgt
+vocab_size_multiple: 8
+src_words_min_frequency: 1
+tgt_words_min_frequency: 1
+share_vocab: True
+n_sample: 0
+
+#### Filter
+src_seq_length: 200
+tgt_seq_length: 200
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/myPOS_transformer_syl_pos.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/myPOS_transformer_syl_pos.train.tgt.txt
+        transforms: [inferfeats, filtertoolong]
+        weight: 1
+        
+    valid:
+        path_src: ./parallel_pseudo/w_pos/syl_valid.src.txt.TAGGED.TAGGED
+        path_tgt: ./parallel_pseudo/original/syl_valid.tgt.txt
+        transforms: [inferfeats]     
+
+# Features options
+n_src_feats: 1
+feat_merge: "mlp"
+feat_vec_size: 512
+src_feats_defaults: "x"
+
+# Transform options
+reversible_tokenization: "joiner"
+
+# Model Configuration
+
+save_model: ./syl_transformer_final2/syl_transformer_final2
+log_file: ./syl_transformer_final2/log_final.txt
+
+# Stop training if it does not improve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+decoder_type: transformer
+encoder_type: transformer
+word_vec_size: 512
+hidden_size: 512
+enc_layers: 6
+dec_layers: 6
+transformer_ff: 2048
+heads: 8
+
+accum_count: 4
+
+# Optimization
+model_dtype: "fp16"
+optim: adam
+adam_beta1: 0.9
+adam_beta2: 0.998
+decay_method: noam
+learning_rate: 0.1
+max_grad_norm: 0.0
+
+batch_size: 64
+batch_type: tokens
+normalization: tokens
+dropout_steps: [0]
+dropout: [0.1]
+attention_dropout: [0.1]
+position_encoding: true
+label_smoothing: 0.1
+
+max_generator_batches: 2
+
+param_init: 0.0
+param_init_glorot: 'true'
+position_encoding: 'true'
+
+world_size: 1
+## Run with GPU no. zero
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000

version1/baseline_experiments/yaml_files/yaml_pos/syl_s2s.yaml

@@ -0,0 +1,75 @@
+# syl_s2s.yaml
+## Where the samples will be written
+
+# Vocab Building
+## Where the samples will be written
+save_data: ./s2s_syl_final2
+
+## Where the vocab(s) will be written
+src_vocab: ./s2s_syl_final2/syl_s2s.vocab.src
+tgt_vocab: ./s2s_syl_final2/syl_s2s.vocab.tgt
+share_vocab: True
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/s2s_syl_pos.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/s2s_syl_pos.train.tgt.txt
+        transforms: [inferfeats, filtertoolong]
+        weight: 1
+        
+    valid:
+        path_src: ./parallel_pseudo/w_pos/syl_valid.src.txt.TAGGED.TAGGED
+        path_tgt: ./parallel_pseudo/original/syl_valid.tgt.txt
+        transforms: [inferfeats]
+
+# Features options
+n_src_feats: 1  
+feat_merge: "mlp"
+feat_vec_size: 512
+src_feats_defaults: "x"
+
+# Transform options
+reversible_tokenization: "joiner"
+
+# Increase sequence length
+src_seq_length: 200  
+tgt_seq_length: 200  
+
+# Training
+
+save_model: ./s2s_syl_final2/s2s_syl_final2
+
+log_file: ./s2s_syl_final2/log_final.txt
+
+# Stop training if it does not improve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+# To save space, limit checkpoints to last n
+keep_checkpoint: 5
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+dropout: [0.3]  # LSTM models often use higher dropout rates
+
+# Train on a single GPU
+world_size: 1
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000
+
+# Model parameters for RNN LSTM
+encoder_type: brnn
+decoder_type: rnn
+rnn_type: LSTM
+rnn_size: 512  # Adjust this based on your model size and complexity
+rnn_num_layers: 3  # Adjust this based on model architecture

version1/baseline_experiments/yaml_files/yaml_pos/word.transformer.yaml

@@ -0,0 +1,96 @@
+# word.transformer.yaml
+
+## Where the samples will be written
+save_data: ./word_transformer_final2
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+## Where the vocab(s) will be written
+src_vocab: ./word_transformer_final2/word_transformer.vocab.src
+tgt_vocab: ./word_transformer_final2/word_transformer.vocab.tgt
+vocab_size_multiple: 8
+src_words_min_frequency: 1
+tgt_words_min_frequency: 1
+share_vocab: True
+n_sample: 0
+
+#### Filter
+src_seq_length: 200
+tgt_seq_length: 200
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/myPOS_transformer_word_pos.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/myPOS_transformer_word_pos.train.tgt.txt
+        transforms: [inferfeats, filtertoolong]
+        weight: 1
+        
+    valid:
+        path_src: ./parallel_pseudo/w_pos/word_valid.src.txt.TAGGED.TAGGED
+        path_tgt: ./parallel_pseudo/original/word_valid.tgt.txt
+        transforms: [inferfeats]     
+
+# Features options
+n_src_feats: 1
+feat_merge: "mlp"
+feat_vec_size: 512
+src_feats_defaults: "x"
+
+# Transform options
+reversible_tokenization: "joiner"
+
+# Model Configuration
+
+save_model: ./word_transformer_final2/word_transformer_final2
+log_file: ./word_transformer_final2/log_final.txt
+
+# Stop training if it does not improve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+decoder_type: transformer
+encoder_type: transformer
+word_vec_size: 512
+hidden_size: 512
+enc_layers: 6
+dec_layers: 6
+transformer_ff: 2048
+heads: 8
+
+accum_count: 4
+
+# Optimization
+model_dtype: "fp16"
+optim: adam
+adam_beta1: 0.9
+adam_beta2: 0.998
+decay_method: noam
+learning_rate: 0.1
+max_grad_norm: 0.0
+
+batch_size: 64
+batch_type: tokens
+normalization: tokens
+dropout_steps: [0]
+dropout: [0.1]
+attention_dropout: [0.1]
+position_encoding: true
+label_smoothing: 0.1
+
+max_generator_batches: 2
+
+param_init: 0.0
+param_init_glorot: 'true'
+position_encoding: 'true'
+
+world_size: 1
+## Run with GPU no. zero
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000
+

version1/baseline_experiments/yaml_files/yaml_pos/word_s2s.yaml

@@ -0,0 +1,75 @@
+# word_s2s.yaml
+## Where the samples will be written
+
+# Vocab Building
+## Where the samples will be written
+save_data: ./s2s_word_final2
+
+## Where the vocab(s) will be written
+src_vocab: ./s2s_word_final2/word_s2s.vocab.src
+tgt_vocab: ./s2s_word_final2/word_s2s.vocab.tgt
+share_vocab: True
+
+# Prevent overwriting existing files in the folder
+overwrite: True
+
+# Corpus opts:
+data:
+    train:
+        path_src: ./parallel_pseudo/sT/s2s_word_pos.train.src.txt
+        path_tgt: ./parallel_pseudo/sT/s2s_word_pos.train.tgt.txt
+        transforms: [inferfeats, filtertoolong]
+        weight: 1
+        
+    valid:
+        path_src: ./parallel_pseudo/w_pos/word_valid.src.txt.TAGGED.TAGGED
+        path_tgt: ./parallel_pseudo/original/word_valid.tgt.txt
+        transforms: [inferfeats]
+
+# Features options
+n_src_feats: 1  
+feat_merge: "mlp"
+feat_vec_size: 512
+src_feats_defaults: "x"
+
+# Transform options
+reversible_tokenization: "joiner"
+
+# Increase sequence length
+src_seq_length: 200  
+tgt_seq_length: 200  
+
+# Training
+
+save_model: ./s2s_word_final2/s2s_word_final2
+
+log_file: ./s2s_word_final2/log_final.txt
+
+# Stop training if it does not improve after n validations
+early_stopping: 10
+
+# Default: 5000 - Save a model checkpoint for each n
+# save_checkpoint_steps: 1000
+
+# To save space, limit checkpoints to last n
+keep_checkpoint: 5
+
+# Optimization
+learning_rate: 0.1
+
+batch_size: 64
+dropout: [0.3]  # LSTM models often use higher dropout rates
+
+# Train on a single GPU
+world_size: 1
+gpu_ranks: [0]
+
+# Number of training iterations
+train_steps: 30000
+
+# Model parameters for RNN LSTM
+encoder_type: brnn
+decoder_type: rnn
+rnn_type: LSTM
+rnn_size: 512  # Adjust this based on your model size and complexity
+rnn_num_layers: 3  # Adjust this based on model architecture