Reuse asymmetric.setup & input validation from asymmetric & org_to_jsonl

Create asymmetric.setup method to
  - initialize model
  - generate compressed jsonl
  - compute embeddings

put input_files, input_file_filter validation in org_to_jsonl for
reuse in main.py, asymmetic.py

main.py

@@ -38,14 +38,8 @@ def search(q: str, n: Optional[int] = 5, t: Optional[str] = 'notes'):
 
 @app.get('/regenerate')
 def regenerate():
-    # Generate Compressed JSONL from Notes in Input Files
+    # Extract Entries, Generate Embeddings
-    org_to_jsonl(args.input_files, args.input_filter, args.compressed_jsonl, args.verbose)
+    extracted_entries, computed_embeddings, _, _, _ = asymmetric.setup(args.input_files, args.input_filter, args.compressed_jsonl, args.embeddings, regenerate=True, verbose=args.verbose)
-
-    # Extract Entries from Compressed JSONL
-    extracted_entries = asymmetric.extract_entries(args.compressed_jsonl, args.verbose)
-
-    # Compute Embeddings from Extracted Entries
-    computed_embeddings = asymmetric.compute_embeddings(extracted_entries, bi_encoder, args.embeddings, regenerate=True, verbose=args.verbose)
 
     # Now Update State
     # update state variables after regeneration complete
@@ -69,23 +63,7 @@ if __name__ == '__main__':
     parser.add_argument('--verbose', action='count', default=0, help="Show verbose conversion logs. Default: 0")
     args = parser.parse_args()
 
-    # Input Validation
+    entries, corpus_embeddings, bi_encoder, cross_encoder, top_k = asymmetric.setup(args.input_files, args.input_filter, args.compressed_jsonl, args.embeddings, args.regenerate, args.verbose)
-    if is_none_or_empty(args.input_files) and is_none_or_empty(args.input_filter):
-        print("At least one of org-files or org-file-filter is required to be specified")
-        exit(1)
-
-    # Initialize Model
-    bi_encoder, cross_encoder, top_k = asymmetric.initialize_model()
-
-    # Map notes in Org-Mode files to (compressed) JSONL formatted file
-    if not args.compressed_jsonl.exists() or args.regenerate:
-        org_to_jsonl(args.input_files, args.input_filter, args.compressed_jsonl, args.verbose)
-
-    # Extract Entries
-    entries = asymmetric.extract_entries(args.compressed_jsonl, args.verbose)
-
-    # Compute or Load Embeddings
-    corpus_embeddings = asymmetric.compute_embeddings(entries, bi_encoder, args.embeddings, regenerate=args.regenerate, verbose=args.verbose)
 
     # Start Application Server
     uvicorn.run(app)

processor/org_mode/org_to_jsonl.py

@@ -12,6 +12,11 @@ import gzip
 
 # Define Functions
 def org_to_jsonl(org_files, org_file_filter, output_file, verbose=0):
+    # Input Validation
+    if is_none_or_empty(org_files) and is_none_or_empty(org_file_filter):
+        print("At least one of org-files or org-file-filter is required to be specified")
+        exit(1)
+
     # Get Org Files to Process
     org_files = get_org_files(org_files, org_file_filter, verbose)
 
@@ -132,10 +137,5 @@ if __name__ == '__main__':
     parser.add_argument('--verbose', '-v', action='count', default=0, help="Show verbose conversion logs, Default: 0")
     args = parser.parse_args()
 
-    # Input Validation
-    if is_none_or_empty(args.input_files) and is_none_or_empty(args.input_filter):
-        print("At least one of org-files or org-file-filter is required to be specified")
-        exit(1)
-
     # Map notes in Org-Mode files to (compressed) JSONL formatted file
     org_to_jsonl(args.input_files, args.input_filter, args.output_file, args.verbose)

search_type/asymmetric.py

@@ -11,6 +11,8 @@ import torch
 import argparse
 import pathlib
 from utils.helpers import get_absolute_path
+from processor.org_mode.org_to_jsonl import org_to_jsonl
+
 
 def initialize_model():
     "Initialize model for assymetric semantic search. That is, where query smaller than results"
@@ -140,24 +142,37 @@ def collate_results(hits, entries, count=5):
         in hits[0:count]]
 
 
+def setup(input_files, input_filter, compressed_jsonl, embeddings, regenerate=False, verbose=False):
+    # Initialize Model
+    bi_encoder, cross_encoder, top_k = initialize_model()
+
+    # Map notes in Org-Mode files to (compressed) JSONL formatted file
+    if not compressed_jsonl.exists() or regenerate:
+        org_to_jsonl(input_files, input_filter, compressed_jsonl, verbose)
+
+    # Extract Entries
+    entries = extract_entries(compressed_jsonl, verbose)
+
+    # Compute or Load Embeddings
+    corpus_embeddings = compute_embeddings(entries, bi_encoder, embeddings, regenerate=regenerate, verbose=verbose)
+
+    return entries, corpus_embeddings, bi_encoder, cross_encoder, top_k
+
+
 if __name__ == '__main__':
     # Setup Argument Parser
     parser = argparse.ArgumentParser(description="Map Org-Mode notes into (compressed) JSONL format")
-    parser.add_argument('--compressed-jsonl', '-j', required=True, type=pathlib.Path, help="Input file for compressed JSONL formatted notes to compute embeddings from")
+    parser.add_argument('--input-files', '-i', nargs='*', help="List of org-mode files to process")
-    parser.add_argument('--embeddings', '-e', required=True, type=pathlib.Path, help="File to save/load model embeddings to/from")
+    parser.add_argument('--input-filter', type=str, default=None, help="Regex filter for org-mode files to process")
+    parser.add_argument('--compressed-jsonl', '-j', type=pathlib.Path, default=pathlib.Path(".notes.jsonl.gz"), help="Compressed JSONL formatted notes file to compute embeddings from")
+    parser.add_argument('--embeddings', '-e', type=pathlib.Path, default=pathlib.Path(".notes_embeddings.pt"), help="File to save/load model embeddings to/from")
+    parser.add_argument('--regenerate', action='store_true', default=False, help="Regenerate embeddings from org-mode files. Default: false")
     parser.add_argument('--results-count', '-n', default=5, type=int, help="Number of results to render. Default: 5")
     parser.add_argument('--interactive', action='store_true', default=False, help="Interactive mode allows user to run queries on the model. Default: true")
     parser.add_argument('--verbose', action='count', default=0, help="Show verbose conversion logs. Default: 0")
     args = parser.parse_args()
 
-    # Initialize Model
+    entries, corpus_embeddings, bi_encoder, cross_encoder, top_k = setup(args.input_files, args.input_filter, args.compressed_jsonl, args.embeddings, args.regenerate, args.verbose)
-    bi_encoder, cross_encoder, top_k = initialize_model()
-
-    # Extract Entries
-    entries = extract_entries(args.compressed_jsonl, args.verbose)
-
-    # Compute or Load Embeddings
-    corpus_embeddings = compute_embeddings(entries, bi_encoder, args.embeddings, args.verbose)
 
     # Run User Queries on Entries in Interactive Mode
     while args.interactive: