[misc] update action loading in validation and preprocess#1143
[misc] update action loading in validation and preprocess#1143H1yori233 wants to merge 6 commits intohao-ai-lab:mainfrom
Conversation
Summary of ChangesHello, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request significantly enhances the flexibility and robustness of action data handling within the Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Changelog
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request updates the action loading mechanism in validation and preprocessing to be more generalized and robust. The changes ensure that the action sequence length matches the number of frames and removes a hardcoded assertion on keyboard dimensions. My review focuses on further improving the robustness of the data validation logic. I've suggested replacing assert statements with ValueError exceptions for data validation checks and adding checks for key existence before accessing dictionary elements to prevent potential KeyError exceptions. These improvements will make the data loading process more resilient to unexpected or malformed data.
| if isinstance(action_data, dict): | ||
| kb_length = len(action_data["keyboard"]) | ||
| ms_length = len(action_data["mouse"]) | ||
| assert kb_length >= num_frames, ( | ||
| f"keyboard length {kb_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| assert ms_length >= num_frames, ( | ||
| f"mouse length {ms_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| sample["keyboard_cond"] = action_data["keyboard"][:num_frames] | ||
| sample["mouse_cond"] = action_data["mouse"][:num_frames] | ||
| else: | ||
| kb_length = len(action_data) | ||
| assert kb_length >= num_frames, ( | ||
| f"keyboard length {kb_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| sample["keyboard_cond"] = action_data[:num_frames] |
There was a problem hiding this comment.
This block has two issues that affect its robustness:
- Unsafe Dictionary Access: It directly accesses
action_data["keyboard"]andaction_data["mouse"]without first checking if these keys exist. This will raise aKeyErrorif the action data dictionary is missing one of these keys, crashing the data loading process. - Non-Robust Validation: It uses
assertfor data validation. Assertions can be disabled in Python with the-Oflag, which would cause these critical length checks to be skipped in optimized runs, potentially leading to silent errors downstream.
To make this code more robust, I suggest checking for key existence before access and using ValueError for validation, which is always active.
if isinstance(action_data, dict):
if "keyboard" in action_data:
kb_data = action_data["keyboard"]
if len(kb_data) < num_frames:
raise ValueError(
f"keyboard length {len(kb_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
sample["keyboard_cond"] = kb_data[:num_frames]
if "mouse" in action_data:
ms_data = action_data["mouse"]
if len(ms_data) < num_frames:
raise ValueError(
f"mouse length {len(ms_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
sample["mouse_cond"] = ms_data[:num_frames]
else:
if len(action_data) < num_frames:
raise ValueError(
f"keyboard length {len(action_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
sample["keyboard_cond"] = action_data[:num_frames]There was a problem hiding this comment.
I'm not sure if this makes sense.
| if "keyboard" in action_dict: | ||
| keyboard_raw = action_dict["keyboard"] | ||
| # Convert 1D bit-flag values to 2D multi-hot encoding | ||
| if isinstance(keyboard_raw, np.ndarray): | ||
| if keyboard_raw.ndim == 1: | ||
| # [T] -> [T, num_bits] | ||
| T = len(keyboard_raw) | ||
| multi_hot = np.zeros((T, num_bits), | ||
| dtype=np.float32) | ||
| action_values = keyboard_raw.astype(int) | ||
| for bit_idx in range(num_bits): | ||
| target_idx = ( | ||
| 2 - | ||
| (bit_idx % 3)) + 3 * (bit_idx // 3) | ||
| if target_idx < num_bits: | ||
| multi_hot[:, target_idx] = ( | ||
| (action_values >> bit_idx) | ||
| & 1).astype(np.float32) | ||
| keyboard_cond_list.append(multi_hot) | ||
| else: | ||
| # If already 2D, pad to num_bits if necessary | ||
| k_data = keyboard_raw.astype(np.float32) | ||
| if k_data.ndim == 2 and k_data.shape[ | ||
| -1] < num_bits: | ||
| padding = np.zeros( | ||
| (k_data.shape[0], | ||
| num_bits - k_data.shape[-1]), | ||
| dtype=np.float32) | ||
| k_data = np.concatenate( | ||
| [k_data, padding], axis=-1) | ||
| keyboard_cond_list.append(k_data) | ||
| else: | ||
| keyboard_cond_list.append(keyboard_raw) | ||
| kb_length = len(action_dict["keyboard"]) | ||
| assert kb_length >= num_frames, ( | ||
| f"keyboard length {kb_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| keyboard_cond_list.append( | ||
| action_dict["keyboard"][:num_frames].astype( | ||
| np.float32)) | ||
| if "mouse" in action_dict: | ||
| mouse_cond_list.append(action_dict["mouse"]) | ||
| ms_length = len(action_dict["mouse"]) | ||
| assert ms_length >= num_frames, ( | ||
| f"mouse length {ms_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| mouse_cond_list.append( | ||
| action_dict["mouse"][:num_frames]) | ||
| else: | ||
| if isinstance(action_data, | ||
| np.ndarray) and action_data.ndim == 1: | ||
| T = len(action_data) | ||
| multi_hot = np.zeros((T, num_bits), | ||
| dtype=np.float32) | ||
| action_values = action_data.astype(int) | ||
| for bit_idx in range(num_bits): | ||
| target_idx = ( | ||
| 2 - (bit_idx % 3)) + 3 * (bit_idx // 3) | ||
| if target_idx < num_bits: | ||
| multi_hot[:, target_idx] = ( | ||
| (action_values >> bit_idx) & 1).astype( | ||
| np.float32) | ||
| keyboard_cond_list.append(multi_hot) | ||
| else: | ||
| # If already 2D, pad to num_bits if necessary | ||
| k_data = action_data.astype(np.float32) | ||
| if k_data.ndim == 2 and k_data.shape[-1] < num_bits: | ||
| padding = np.zeros( | ||
| (k_data.shape[0], | ||
| num_bits - k_data.shape[-1]), | ||
| dtype=np.float32) | ||
| k_data = np.concatenate([k_data, padding], | ||
| axis=-1) | ||
| keyboard_cond_list.append(k_data) | ||
| kb_length = len(action_data) | ||
| assert kb_length >= num_frames, ( | ||
| f"keyboard length {kb_length} is smaller than " | ||
| f"num_frames {num_frames} for {action_path}.") | ||
| keyboard_cond_list.append( | ||
| action_data.astype(np.float32)[:num_frames]) |
There was a problem hiding this comment.
Using assert for data validation is not robust, as assertions can be disabled with Python's -O flag. This could lead to silent failures or incorrect behavior if the action data has an unexpected length. It's safer to raise a ValueError for invalid data to ensure the check is always performed. The suggested change replaces all assert statements in this block with explicit ValueError checks.
if "keyboard" in action_dict:
kb_data = action_dict["keyboard"]
if len(kb_data) < num_frames:
raise ValueError(
f"keyboard length {len(kb_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
keyboard_cond_list.append(
kb_data[:num_frames].astype(
np.float32))
if "mouse" in action_dict:
ms_data = action_dict["mouse"]
if len(ms_data) < num_frames:
raise ValueError(
f"mouse length {len(ms_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
mouse_cond_list.append(ms_data[:num_frames])
else:
if len(action_data) < num_frames:
raise ValueError(
f"keyboard length {len(action_data)} is smaller than "
f"num_frames {num_frames} for {action_path}.")
keyboard_cond_list.append(
action_data.astype(np.float32)[:num_frames])
2 participants
num_frames.