support scenario and social goal generation based on new inspirationa…

…l prompts

.gitignore

@@ -199,3 +199,5 @@ cython_debug/
 ./llm_rl/preprocess/GPT4-4_Redis_Easy_No_Slide
 
 llm_rl/*cache/
+data_generate/**/*.jsonl
+data_generate/**/*.tsv

data_generate/generate.py

@@ -424,11 +424,12 @@ async def agenerate_env_profile(
     """
     return await agenerate(
         model_name=model_name,
-        template="""Please generate scenarios and goals following the examples below. 
+        template="""Please generate scenarios and goals following those examples below:
         Examples:
         {examples}
-        Additionally, generate creative scenarios based on one or more inspirational prompt. The scenario and social goal is motivated by them but not very related to those prompts, when creating the goals, try to find one point that both sides may not agree upon initially and need to collaboratively resolve it.
-        Inspirational prompt: {inspiration_prompt}
+        Generate creative scenarios and social goals based on one or more inspirational prompt listed below. The scenario and social goal should be related to at least one of those inspirational prompts, when creating the goals, try to find one point that both sides may not agree upon initially and need to collaboratively resolve it.
+        Inspirational prompt: 
+        {inspiration_prompt}
         Please use the following format and follow that format strictly:
         {format_instructions}
         """,

data_generate/requirments.txt

@@ -1,4 +1,5 @@
 sotopia
+convokit
 rejson
 evaluate
 redis==3.5.3 # for step3 and step4

data_generate/step0_create_new_inspirational_prompt.py

@@ -0,0 +1,66 @@
+import jsonlines
+import csv
+from tqdm import tqdm
+
+def collect_social_iqa(inspirational_prompt_data):
+    with jsonlines.open('./social_src/social_iqa_train.jsonl', 'r') as f:
+        social_iqa_dataset = list(f)
+    for data in social_iqa_dataset:
+        inspirational_prompt_data.append({'prompt': data['context'], 'source': 'social_iqa'})
+    return inspirational_prompt_data
+
+
+def collect_normbank(inspirational_prompt_data):
+    rows = []
+    with open("./social_src/NormBank.csv", 'r') as file:
+        csvreader = csv.reader(file)
+        header = next(csvreader)
+        for row in csvreader:
+            inspirational_prompt_data.append({'prompt': row[1], 'source': 'normbank'})
+    return inspirational_prompt_data
+
+
+def collect_social_chemistry(inspirational_prompt_data):
+    rows = []
+    with open("./social_src/social-chem-101.v1.0.tsv", 'r') as file:
+        csvreader = csv.reader(file, delimiter="\t")
+        header = next(csvreader)
+        for row in csvreader:
+            inspirational_prompt_data.append({'prompt': row[-6], 'source': 'normbank'})
+    return inspirational_prompt_data
+
+def collect_persuation_for_good(inspirational_prompt_data):
+    from convokit import Corpus, download
+    corpus = Corpus(filename=download("persuasionforgood-corpus"))
+    import pdb; pdb.set_trace()
+
+def delete_sotopia_data(inspirational_prompt_data):
+    sotopia_prompts = []
+    with open("./social_src/inspirational_prompt_for_sotopia.csv", 'r') as file:
+        csvreader = csv.reader(file)
+        header = next(csvreader)
+        for row in csvreader:
+            sotopia_prompts.append(row[0])
+    for inspirational_prompt in tqdm(inspirational_prompt_data):
+        if inspirational_prompt['prompt'] in sotopia_prompts:
+            inspirational_prompt_data.remove(inspirational_prompt)
+    print(len(inspirational_prompt_data))
+    return inspirational_prompt_data
+
+
+inspirational_prompt_data = []
+#inspirational_prompt_data = collect_persuation_for_good(inspirational_prompt_data)
+inspirational_prompt_data = collect_social_chemistry(inspirational_prompt_data)
+inspirational_prompt_data = collect_social_iqa(inspirational_prompt_data)
+inspirational_prompt_data = collect_normbank(inspirational_prompt_data)
+inspirational_prompt_data = delete_sotopia_data(inspirational_prompt_data)
+
+fieldnames = inspirational_prompt_data[0].keys()
+
+# Write to a CSV file
+with open('./social_src/inspirational_prompt.csv', 'w', newline='', encoding='utf-8') as csvfile:
+    writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
+    # Write the header (optional, but usually a good idea)
+    writer.writeheader()
+    for row in inspirational_prompt_data:
+        writer.writerow(row)

data_generate/step1_generate_env_profile.py

@@ -19,7 +19,7 @@ def pydantics_to_csv(filename: str, data: list[T]) -> None:
 #random.seed(41)
 
 envs = EnvironmentProfile.find().all()
-ins_prompts = pd.read_csv("./inspirational_prompt_for_env.csv")
+ins_prompts = pd.read_csv("./social_src/inspirational_prompt.csv")
 prompts = [prompt.strip().replace('\"', '') for prompt in ins_prompts["prompt"].tolist()]
 
 # randomly choose 3 prompts
@@ -29,10 +29,10 @@ def pydantics_to_csv(filename: str, data: list[T]) -> None:
 target_num = 500
 
 for i in range(target_num):
-    sampled_envs = random.sample(envs, 5)
-    sampled_prompt = random.sample(prompts, 5)
-    sampled_examples.append(f"1.{sampled_envs[0].json()}\n2.{sampled_envs[1].json()}\n3.{sampled_envs[2].json()}\n4.{sampled_envs[3].json()}\n5.{sampled_envs[4].json()}")
-    sampled_prompts.append(f"1.{sampled_prompt[0]}\n2.{sampled_prompt[1]}\n3.{sampled_prompt[2]}\n4.{sampled_prompt[3]}\n5.{sampled_prompt[4]}")
+    sampled_envs = random.sample(envs, 1)
+    sampled_prompt = random.sample(prompts, 3)
+    sampled_examples.append(f"{sampled_envs[0].json()}")
+    sampled_prompts.append(f"1.{sampled_prompt[0]}\n2.{sampled_prompt[1]}\n3.{sampled_prompt[2]}")
 
 assert len(sampled_examples) == target_num
 assert len(sampled_prompts) == target_num