Update branch_README.md

branch_README.md

@@ -1,15 +1,30 @@
 Adds scripts to train models to predict titles, note-seg, whether two sequences should be joined, and where a sequence should be split. 
 
+Scripts added:
+1. src/train_title_prediction_model.py
+2. src/classify_titles_context.py
+3. src/classify_note_seg.py
+4. src/train_join_model.py
+5. src/classify_joins.py
+6. src/train_split_model.py
+
+Data added:
+1. datasets/titles/
+2. datasets/join_segments/
+3. datasets/split_segments/
+
 # Title prediction
 src/train_title_prediction_model.py trains a BERT model to predict whether a sequence is a title or not. 
 Training, validation and test data are stored in datasets/titles/. 
 A model trained on less data is available through huggingface fberi/BertModel-lc.
 src/classify_titles_context.py uses this model to add title tags to the corpus. 
 
-A new branch (update_context_functions) of pyriksdagen adjusts the logic to include more text from the previous sequence in some cases. This may be a small improvement compared to the existing title prediction model.
+A [branch](https:/swerik-project/pyriksdagen/tree/update_context_functions) (update_context_functions) of pyriksdagen adjusts the logic to include more text from the previous sequence in some cases. This may be a small improvement compared to the existing title prediction model.
 
 # Note-seg prediction
-src/classify_note_seg.py reclassifies notes and segs in the corpus. A pre-trained model to do this can be downloaded from https:/welfare-state-analytics/bert-riksdagen-classifier/releases/tag/v0.4.0
+src/classify_note_seg.py reclassifies notes and segs in the corpus. 
+
+A pre-trained model to do this can be downloaded from https:/welfare-state-analytics/bert-riksdagen-classifier/releases/tag/v0.4.0
 
 # Join prediction
 src/train_join_model.py trains a BERT model to predict whether two adjacent sequences should be joined or not. Training, validation, and test data are stored in datasets/join_segments/.
@@ -18,4 +33,4 @@ src/classify_joins.py uses this model to join segments in the corpus.
 
 # split prediction
 src/train_split_model.py trains a BERT model to predict which tokens to split a sequence on. This can predict multiple splits per sequence. Training, validation, and test data is stored in datasets/split_segments/. Currently the model tries to perform all types of splits (paragraphs, margins, titles, etc.) but performance is quite poor. Limiting the scope to the easier to classify splits might be a way to make the model useable. 
-Since performance is unsatisfactory, there is no script to split segments in the corpus. 
+Since performance is unsatisfactory, there is no script to split segments in the corpus.