[enhancement] Simplify config parameters (#40)

* [refactor] Simplify keys for scheduler, optimizer and training

* [refactor] Simplify keys with _attributes -> _configs

Author: apsdehal

configs/captioning/coco/butd.yml

@@ -1,6 +1,6 @@
 includes:
 - common/defaults/configs/datasets/captioning/coco.yml
-model_attributes:
+model_config:
   butd: &butd
     model_data_dir: ../data/
     metrics:
@@ -35,13 +35,13 @@ model_attributes:
         weights_file: detectron/fc6/fc7_w.pkl
     inference:
       type: greedy
-optimizer_attributes:
+optimizer:
   type: Adamax
   params:
     eps: 1.0e-08
     lr: 0.01
     weight_decay: 0
-training_parameters:
+training:
   clip_norm_mode: all
   clip_gradients: true
   lr_ratio: 0.1

configs/captioning/coco/butd_beam_search.yml

@@ -1,10 +1,10 @@
 includes:
 - ../configs/captioning/coco/butd.yml
-model_attributes:
+model_config:
   butd: &butd
     inference:
       type: beam_search
       params:
         beam_length: 5
-training_parameters:
+training:
   batch_size: 1

configs/captioning/coco/butd_nucleus_sampling.yml

@@ -1,10 +1,10 @@
 includes:
 - ../configs/captioning/coco/butd.yml
-model_attributes:
+model_config:
   butd: &butd
     inference:
       type: nucleus_sampling
       params:
         sum_threshold: 0.8
-training_parameters:
+training:
   batch_size: 1

configs/captioning/m4c_textcaps/butd.yml

@@ -1,6 +1,6 @@
 includes:
 - common/defaults/configs/datasets/captioning/coco.yml
-dataset_attributes:
+dataset_config:
   coco:
     image_features:
       train:
@@ -43,7 +43,7 @@ dataset_attributes:
     use_ocr: false
     # Return spatial information of OCR tokens if present
     use_ocr_info: false
-model_attributes:
+model_config:
   butd: &butd
     model_data_dir: ../data/
     metrics:
@@ -78,13 +78,13 @@ model_attributes:
         weights_file: detectron/fc6/fc7_w.pkl
     inference:
       type: greedy
-optimizer_attributes:
+optimizer:
   type: Adamax
   params:
     eps: 1.0e-08
     lr: 0.01
     weight_decay: 0
-training_parameters:
+training:
   clip_norm_mode: all
   clip_gradients: true
   lr_ratio: 0.1

configs/captioning/m4c_textcaps/butd_beam_search.yml

@@ -1,15 +1,15 @@
 includes:
 - ../configs/captioning/m4c_textcaps/butd.yml
-dataset_attributes:
+dataset_config:
   coco:
     imdb_files:
       val:
       - imdb/m4c_textcaps/imdb_val_filtered_by_image_id.npy
-model_attributes:
+model_config:
   butd: &butd
     inference:
       type: beam_search
       params:
         beam_length: 5
-training_parameters:
+training:
   batch_size: 1

configs/captioning/m4c_textcaps/butd_eval_pretrained_coco_model.yml

@@ -1,6 +1,6 @@
 includes:
 - ../configs/captioning/m4c_textcaps/butd_beam_search.yml
-dataset_attributes:
+dataset_config:
   coco:
     processors:
       text_processor:

configs/captioning/m4c_textcaps/m4c_captioner.yml

@@ -1,7 +1,7 @@
 includes:
 - common/defaults/configs/datasets/captioning/m4c_textcaps.yml
 # Use soft copy
-dataset_attributes:
+dataset_config:
   m4c_textcaps:
     image_features:
       train:
@@ -43,7 +43,7 @@ dataset_attributes:
         type: phoc
         params:
           max_length: 50
-model_attributes:
+model_config:
   m4c_captioner:
     lr_scale_frcn: 0.1
     lr_scale_text_bert: 0.1
@@ -73,13 +73,13 @@ model_attributes:
     losses:
     - type: m4c_decoding_bce_with_mask
     remove_unk_in_pred: true
-optimizer_attributes:
+optimizer:
   params:
     eps: 1.0e-08
     lr: 1e-4
     weight_decay: 0
   type: Adam
-training_parameters:
+training:
     clip_norm_mode: all
     clip_gradients: true
     max_grad_l2_norm: 0.25

configs/captioning/m4c_textcaps/m4c_captioner_coco.yml

@@ -1,7 +1,7 @@
 includes:
 - common/defaults/configs/datasets/captioning/m4c_textcaps.yml
 # Use soft copy
-dataset_attributes:
+dataset_config:
   m4c_textcaps:
     image_features:
       train:
@@ -43,7 +43,7 @@ dataset_attributes:
         type: phoc
         params:
           max_length: 50
-model_attributes:
+model_config:
   m4c_captioner:
     lr_scale_frcn: 0.1
     lr_scale_text_bert: 0.1
@@ -73,13 +73,13 @@ model_attributes:
     losses:
     - type: m4c_decoding_bce_with_mask
     remove_unk_in_pred: true
-optimizer_attributes:
+optimizer:
   params:
     eps: 1.0e-08
     lr: 1e-4
     weight_decay: 0
   type: Adam
-training_parameters:
+training:
     clip_norm_mode: all
     clip_gradients: true
     max_grad_l2_norm: 0.25

configs/captioning/m4c_textcaps/m4c_captioner_coco_eval_on_textcaps.yml

@@ -1,6 +1,6 @@
 includes:
 - ../configs/captioning/m4c_textcaps/m4c_captioner_coco.yml
-dataset_attributes:
+dataset_config:
   m4c_textcaps:
     image_features:
       val:

configs/captioning/m4c_textcaps/m4c_captioner_coco_textcaps_joint.yml

@@ -1,7 +1,7 @@
 includes:
 - common/defaults/configs/datasets/captioning/m4c_textcaps.yml
 # Use soft copy
-dataset_attributes:
+dataset_config:
   m4c_textcaps:
     image_features:
       train:
@@ -111,7 +111,7 @@ dataset_attributes:
         type: phoc
         params:
           max_length: 50
-model_attributes:
+model_config:
   m4c_captioner:
     lr_scale_frcn: 0.1
     lr_scale_text_bert: 0.1
@@ -141,13 +141,13 @@ model_attributes:
     losses:
     - type: m4c_decoding_bce_with_mask
     remove_unk_in_pred: true
-optimizer_attributes:
+optimizer:
   params:
     eps: 1.0e-08
     lr: 1e-4
     weight_decay: 0
   type: Adam
-training_parameters:
+training:
     clip_norm_mode: all
     clip_gradients: true
     max_grad_l2_norm: 0.25

configs/captioning/m4c_textcaps/m4c_captioner_without_ocr.yml

@@ -1,7 +1,7 @@
 includes:
 - common/defaults/configs/datasets/captioning/m4c_textcaps.yml
 # Use soft copy
-dataset_attributes:
+dataset_config:
   m4c_textcaps:
     image_features:
       train:
@@ -44,7 +44,7 @@ dataset_attributes:
         params:
           max_length: 50
     use_ocr: False  # remove all the OCRs from each image
-model_attributes:
+model_config:
   m4c_captioner:
     lr_scale_frcn: 0.1
     lr_scale_text_bert: 0.1
@@ -74,13 +74,13 @@ model_attributes:
     losses:
     - type: m4c_decoding_bce_with_mask
     remove_unk_in_pred: true
-optimizer_attributes:
+optimizer:
   params:
     eps: 1.0e-08
     lr: 1e-4
     weight_decay: 0
   type: Adam
-training_parameters:
+training:
     clip_norm_mode: all
     clip_gradients: true
     max_grad_l2_norm: 0.25

configs/pythia_bert/cc_pretrain.yml

@@ -2,7 +2,7 @@ includes:
 - ../configs/pythia_bert/coco_pretrain.yml
 - common/defaults/configs/datasets/captioning/masked_cc.yml
 
-model_attributes:
+model_config:
   pythia_bert:
     combine_embeddings: true
     model_data_dir: ../data/
@@ -56,6 +56,6 @@ model_attributes:
       num_heads: 8
       dropout: 0.1
 
-dataset_attributes:
+dataset_config:
   masked_cc:
     return_info: true

configs/pythia_bert/coco_pretrain.yml

@@ -1,7 +1,7 @@
 includes:
 - common/defaults/configs/datasets/captioning/masked_coco.yml
 
-model_attributes:
+model_config:
   pythia_bert:
     combine_embeddings: true
     model_data_dir: ../data/
@@ -55,24 +55,24 @@ model_attributes:
       num_heads: 8
       dropout: 0.1
 
-dataset_attributes:
+dataset_config:
   masked_coco:
     return_info: true
 
-optimizer_attributes:
+optimizer:
   type: adam_w
   params:
     lr: 5e-5
     eps: 1e-8
 
-scheduler_attributes:
+scheduler:
   type: warmup_linear
   params:
     num_warmup_steps: 1000
     num_training_steps: 11000
-    
-training_parameters:
+
+training:
   batch_size: 480
   lr_scheduler: true
   # Don't forget to update schedule_attributes if you update this
-  max_updates: 11000
+  max_updates: 11000

configs/pythia_bert/combined_coco_vqa2.yml

@@ -2,17 +2,17 @@ includes:
 - ../configs/visual_bert/coco_pretrain.yml
 - ../configs/visual_bert/vqa2.yml
 
-model_attributes:
+model_config:
   visual_bert:
     training_head_type: pretraining,vqa
 
-scheduler_attributes:
+scheduler:
   type: warmup_linear
   params:
     num_warmup_steps: 3000
     num_training_steps: 33000
-  
-training_parameters:
+
+training:
   max_updates: 34000
   monitored_metric: vqa2/vqa_accuracy
-  metric_minimize: false
+  metric_minimize: false

configs/pythia_bert/masked_vqa2_pretrain.yml

@@ -2,7 +2,7 @@ includes:
 - common/defaults/configs/datasets/vqa/masked_vqa2.yml
 - ../configs/pythia_bert/coco_pretrain.yml
 
-dataset_attributes:
+dataset_config:
   masked_vqa2:
     imdb_files:
       train:

configs/pythia_bert/visual_entailment.yml

@@ -2,15 +2,15 @@ includes:
 - common/defaults/configs/datasets/reasoning/visual_entailment.yml
 - ../configs/pythia_bert/vqa2.yml
 
-model_attributes:
+model_config:
   pythia_bert:
     training_head_type: visual_entailment
     losses:
     - type: cross_entropy
     metrics:
     - accuracy
 
-dataset_attributes:
+dataset_config:
   visual_entailment:
     return_info: true
     processors:
@@ -23,14 +23,13 @@ dataset_attributes:
               do_lower_case: true
           mask_probability: 0
           max_seq_length: 128
-        
-scheduler_attributes:
+
+scheduler:
   type: warmup_linear
   params:
     num_warmup_steps: 6000
     num_training_steps: 60000
 
-training_parameters:
+training:
   monitored_metric: visual_entailment/accuracy
   metric_minimize: false
-