chore: add print

app/util/sefaria_specific.py

@@ -58,3 +58,15 @@ def get_ref_text_with_fallback(oref: Ref, lang: str, auto_translate=False) -> st
 
 def convert_trefs_to_sources(trefs) -> list[TopicPromptSource]:
     return [_make_topic_prompt_source(Ref(tref), '', with_commentary=False) for tref in trefs]
+
+
+def remove_refs_from_same_category(refs: list[Ref], max_category_count: int) -> list[Ref]:
+    from collections import defaultdict
+    cat_counts = defaultdict(int)
+    out_refs = []
+    for ref in refs:
+        cat_counts[ref.primary_category] += 1
+        if cat_counts[ref.primary_category] > max_category_count:
+            continue
+        out_refs.append(ref)
+    return out_refs

app/util/topic.py

@@ -1,12 +1,13 @@
 import django
 django.setup()
 from sefaria.model.topic import Topic as SefariaTopic
+from sefaria.model.text import Ref
 from functools import partial
 from basic_langchain.schema import SystemMessage, HumanMessage
 from basic_langchain.chat_models import ChatOpenAI, ChatAnthropic
 from util.general import get_by_xml_tag, run_parallel, summarize_text
 from util.webpage import get_webpage_text
-from util.sefaria_specific import filter_invalid_refs, convert_trefs_to_sources
+from util.sefaria_specific import filter_invalid_refs, convert_trefs_to_sources, remove_refs_from_same_category
 from sefaria_llm_interface.common.topic import Topic
 from sefaria.helper.topic import get_topic
 
@@ -54,7 +55,8 @@ def get_topic_description_from_webpages(topic: Topic):
 
 
 def get_topic_description_from_top_sources(topic: Topic, verbose=True):
-    top_trefs = get_top_trefs_from_slug(topic.slug, top_n=5)
+    top_trefs = get_top_trefs_from_slug(topic.slug, top_n=15)
+    top_trefs = [r.normal() for r in remove_refs_from_same_category([Ref(tref) for tref in top_trefs], 2)][:6]
     top_sources = convert_trefs_to_sources(top_trefs)
     llm = ChatAnthropic(model='claude-3-haiku-20240307', temperature=0)
     summaries = run_parallel([source.text['en'] for source in top_sources],

experiments/topic_source_curation_v2/choose.py

@@ -75,6 +75,7 @@ def choose_ideal_clusters(clusters: list[Cluster], max_clusters: int) -> list[Cl
 
 def sort_clusters(clusters: list[Cluster], topic:Topic, max_clusters: int) -> list[Cluster]:
     # sorted_clusters = _sort_by_highest_avg_pairwise_distance(clusters)
+    print(f"Sorting {len(clusters)} clusters by interestingness...")
     sorted_clusters = _sort_clusters_by_instruction(clusters, topic)
     sorted_cluster_items = run_parallel(sorted_clusters, partial(_sort_within_cluster, topic=topic), max_workers=100, desc="Sorting for interestingness within cluster")
     for cluster, sorted_items in zip(sorted_clusters, sorted_cluster_items):
@@ -141,7 +142,7 @@ def gpt_compare(a, b) -> int:
 def sort_by_instruction(documents,  comparison_instruction, key_extraction_func=lambda x:x):
     from functools import cmp_to_key
     message_suffix = " The only output should be either '1' or '2' or '0'"
-    llm = ChatOpenAI(model="gpt-4o", temperature=0)
+    llm = ChatOpenAI(model='gpt-3.5-turbo-0125', temperature=0)
     system = SystemMessage(
         content=
         comparison_instruction