fix: refine details (#979)

* feat: add parquet preview and extract common DataFrame preview logic

* refactor: improve error messages, prompts, regex, and session loading

* lint

rdagent/utils/agent/ret.py

@@ -31,7 +31,8 @@ class PythonAgentOut(AgentOut):
 
     @classmethod
     def extract_output(cls, resp: str):
-        match = re.search(r".*```[Pp]ython\n(.*)\n```.*", resp, re.DOTALL)
+        # We use lazy mode (.*?) to only extract the first code block in the response.
+        match = re.search(r".*```[Pp]ython\n(.*?)\n```.*", resp, re.DOTALL)
         if match:
             code = match.group(1)
             code = re.sub(r"</?code>", "", code, flags=re.IGNORECASE)

rdagent/utils/env.py

@@ -116,6 +116,7 @@ def pull_image_with_progress(image: str) -> None:
 
 
 class EnvConf(ExtendedBaseSettings):
+    # TODO: add prefix ....
     default_entry: str
     extra_volumes: dict = {}
     running_timeout_period: int = 3600  # 10 minutes

rdagent/utils/workflow/loop.py

@@ -250,7 +250,11 @@ class LoopBase:
                         current_step = self.step_idx[li]
                         self.pbar.n = current_step
                         next_step = self.step_idx[li] % len(self.steps)
-                        self.pbar.set_postfix(loop_index=li, step_index=next_step, step_name=self.steps[next_step])
+                        self.pbar.set_postfix(
+                            loop_index=li + next_step_idx // len(self.steps),
+                            step_index=next_step,
+                            step_name=self.steps[next_step],
+                        )
                         self._check_exit_conditions_on_step()
                     else:
                         logger.warning(f"Step forward {si} of loop {li} is skipped.")
@@ -411,6 +415,18 @@ class LoopBase:
             An instance of LoopBase with the loaded session.
         """
         path = Path(path)
+        # if the path is a directory, load the latest session
+        if path.is_dir():
+            if path.name != "__session__":
+                path = path / "__session__"
+
+            if not path.exists():
+                raise FileNotFoundError(f"No session file found in {path}")
+
+            # iterate the dump steps in increasing order
+            files = sorted(path.glob("*/*_*"), key=lambda f: (int(f.parent.name), int(f.name.split("_")[0])))
+            path = files[-1]
+            logger.info(f"Loading latest session from {path}")
         with path.open("rb") as f:
             session = cast(LoopBase, pickle.load(f))