New alpha + minor tweaks of prompt

Author: Breta01

brain/agent.py

@@ -31,11 +31,12 @@ def prompt(
 
 Steps:
 1. Think about the alphas you created so far and the PROS and CONS
-2. If there is some good alpha, try to improve it first
-2. Reason about possible improvements
-3. Check the list of available operators and data fields
-5. Always try uncommon data fields first before using close, volumn, open, close, etc.
-4. Propose the alpha and run the alpha simulation.
+2. If there is some good alpha (passing tests and seft-correlation < 0.5), try to improve it first
+3. Reason about possible improvements
+4  From time to time try to create a new alpha from scratch
+5. Check the list of available operators and data fields
+6. Always try uncommon data fields first before using close, volumn, open, close, etc.
+7. Propose the alpha and run the alpha simulation.
 
 - Use more of ts_corr(x, y, d), ts_regression(y, x, d, lag=0, rettype=0)
 - Try applying different functions and changing parameter values.

brain/agent_config.py

@@ -3,7 +3,8 @@
 DEFAULT_CONFIG = {
     "region": "USA",
     "universe": "TOP3000",
-    "neutralization": "MARKET",
+    "neutralization": "INDUSTRY",
+    "decay": 8,
     "delay": 1,
 }
 

brain/alphas/alpha_22.py

@@ -0,0 +1,18 @@
+# Code
+code = """
+a = ts_corr(ts_zscore(high + ts_delta(open, 10), 30), ts_zscore(low + ts_delta(open, 10), 30), 100);
+trade_when(abs(ts_zscore(volume, 100)) < 0.5, a, -1)
+"""
+
+# Simulation Settings
+config = {
+    "region": "USA",
+    "universe": "TOP3000",
+    "decay": 5,
+    "delay": 1,
+    "truncation": 0.01,
+    "neutralization": "INDUSTRY",
+    "pasteurization": "ON",
+    "nan_handling": "OFF",
+    "unit_handling": "VERIFY",
+}

brain/api.py

@@ -184,7 +184,7 @@ def get_self_corr(self, alpha_id: str):
 
         return self_corr_df
 
-    def check_self_corr_test(self, alpha_id: str, threshold: float = 0.7):
+    def check_self_corr_test(self, alpha_id: str, threshold: float = 0.6):
         """Check if alpha's self_corr test passed and saves result to dataframe"""
         self_corr_df = self.get_self_corr(alpha_id)
         if self_corr_df.empty:

brain/search_algorithm.py

@@ -150,7 +150,9 @@ def main():
 
                 # Start a new alpha simulation
                 if alpha is not None and alpha.alpha_id is not None and alpha.fitness < -0.5:
-                    alpha_config = {**alpha_config, "regular": f"-{alpha.regular}"}
+                    split = alpha.regular.split(";")
+                    regular = f'{";".join(split[:-1])}{";" if len(split) > 1 else ""}-({split[-1]})'
+                    alpha_config = {**alpha_config, "regular": regular}
                     response = BrainAPI.start_simulation(alpha_config)
                 else:
                     response, alpha_config = create_alpha_simulation(alphas_dict, alphas_categories)

brain/tools/datafields.py

@@ -20,7 +20,7 @@ def filter_datafields(data: pd.DataFrame) -> pd.DataFrame:
     return (
         data[(data["coverage"] > 0.5) & (data["alphaCount"] > 100)]
         .sort_values(by="alphaCount", ascending=False)
-        .iloc[:1000]
+        .iloc[:2000]
     )