split the utils and reorg (#19)

* split the utils

* fix isort

* move prompting into agents

* move prompting into agents

research_town/agent_base.py → research_town/agents/agent_base.py

@@ -4,10 +4,10 @@
 
 import requests
 
-from .utils import (
-    bfs,
+from ..utils.author_relation import bfs
+from ..utils.paper_collection import get_bert_embedding
+from .agent_prompting import (
     generate_ideas,
-    get_bert_embedding,
     summarize_research_direction,
     summarize_research_field,
 )

research_town/agents/agent_prompting.py

@@ -0,0 +1,133 @@
+import time
+from typing import Any, Dict, List, Tuple
+
+import openai
+
+from ..utils.paper_collection import (
+    get_bert_embedding,
+    neiborhood_search,
+)
+
+KEY = "7a1821d4e4a3e41e3d523e97e0fd8950dedac2824aef99cb19d550500cb21a42"
+openai.api_base = "https://api.together.xyz"
+llm_model = "mistralai/Mixtral-8x7B-Instruct-v0.1"
+
+
+def summarize_research_field(
+    profile: Dict[str, str],
+    keywords: List[str],
+    dataset: Dict[str, Any],
+    data_embedding: Dict[str, Any],
+) -> Tuple[List[str], List[str]]:
+    openai.api_key = KEY
+    query_format = (
+        "Given the profile of me, keywords, some recent paper titles and abstracts. Could you summarize the keywords of high level research backgrounds and trends in this field (related to my profile if possible)."
+        "Here is my profile: {profile}"
+        "Here are the keywords: {keywords}"
+    )
+
+    input = {"profile": profile, "keywords": keywords}
+
+    query = query_format.format_map(input)
+
+    query_embedding = get_bert_embedding([query])
+    text_chunk_l = []
+    title_chunk = []
+    data_embedding_l = []
+    for k in dataset.keys():
+        title_chunk.extend(dataset[k]["info"])
+        text_chunk_l.extend(dataset[k]["abstract"])
+        data_embedding_l.extend(data_embedding[k])
+
+    chunks_embedding_text_all = data_embedding_l
+    num_chunk = 10
+
+    neib_all = neiborhood_search(chunks_embedding_text_all, query_embedding, num_chunk)
+    neib_all = neib_all.reshape(-1)
+
+    context = []
+    retrieve_paper = []
+    for i in neib_all:
+        context.append(text_chunk_l[i])
+        retrieve_paper.append(title_chunk[i])
+
+    input["papers"] = "; ".join(context)
+    prompt = query_format.format_map(input)
+
+    try:
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            max_tokens=512,
+        )
+    except Exception:
+        time.sleep(20)
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            max_tokens=512,
+        )
+
+    content = completion.choices[0].message["content"]
+    content_l = [content]
+    return content_l, retrieve_paper
+
+
+def generate_ideas(trend: str) -> List[str]:
+    prompt_qa = (
+        "Here is a high-level summarized trend of a research field {trend}. "
+        "How do you view this field? Do you have any novel ideas or insights? "
+        "Please give me 3 to 5 novel ideas and insights in bullet points. Each bullet point should be concise, containing 2 or 3 sentences."
+    )
+    openai.api_key = KEY
+    input = {"trend": trend}
+    prompt = prompt_qa.format_map(input)
+    try:
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0,
+            seed=42,
+            top_p=1,
+        )
+    except Exception:
+        time.sleep(20)
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0,
+            seed=42,
+            top_p=1,
+        )
+    content = completion.choices[0].message["content"]
+    return [content]
+
+
+def summarize_research_direction(personal_info: str) -> List[str]:
+    prompt_qa = (
+        "Based on the list of the researcher's first person persona from different times, please write a comprehensive first person persona. "
+        "Focus more on more recent personas. Be concise and clear (around 300 words). "
+        "Here are the personas from different times: {personalinfo}"
+    )
+    openai.api_key = KEY
+    input = {"personalinfo": personal_info}
+    prompt = prompt_qa.format_map(input)
+    try:
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0,
+            seed=42,
+            top_p=0,
+        )
+    except Exception:
+        time.sleep(20)
+        completion = openai.ChatCompletion.create(
+            model=llm_model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0,
+            seed=42,
+            top_p=0,
+        )
+    content = completion.choices[0].message["content"]
+    return [content]

research_town/env_base.py → research_town/envs/env_base.py

@@ -1,7 +1,7 @@
 from typing import Dict
 
-from .agent_base import BaseResearchAgent
-from .kb_base import BaseKnowledgeBase
+from ..agents.agent_base import BaseResearchAgent
+from ..kbs.kb_base import BaseKnowledgeBase
 
 
 class BaseMultiAgentEnv(object):

research_town/kb_base.py → research_town/kbs/kb_base.py

@@ -1,6 +1,6 @@
 from typing import Dict, List
 
-from .utils import get_daily_papers
+from ..utils.paper_collection import get_daily_papers
 
 
 class BaseKnowledgeBase(object):