Generate a special aggregation node for seed evaluation results

Generate the next sub-stage goal based on what has been done so far. Args: main_stage_goal: The overall goal for the current main

Get all leaf nodes in the subtree rooted at node.

Identify systemic issues and challenges from the current stage's results

Run multiple seeds of the same node to get statistical metrics. Returns a list of nodes with different random seeds.

Generate an aggregation node for seed evaluation results

Test if an object can be pickled

Save the current state of the experiment

Select N nodes to process in parallel, balancing between tree exploration and exploitation. Note: - This function runs in the

Determine if we should maximize based on the metric format

Update ablation tracking state based on execution results. Args: result_node: Node containing ablation execution results

Update hyperparam tuning tracking state based on execution results.

Recursively walk a directory (analogous to os.walk but for pathlib.Path)

Absorb the result of executing the code from this node.

Calculate the cost for a specific model based on token usage.

Compile the LaTeX project in a temporary folder, then determine where the "References" section begins using cleaned text extraction. Next, co

Cleanup parallel workers and resources

Copy contents of `src` to `dst`. Unlike shutil.copytree, the dst dir can exist and will be merged. If src is a file, only that file will be c

Create a unified visualization that shows all completed stages in a tabbed interface. This will be placed in the main log directory.

Temporarily copy the latex folder, compile, and detect on which page the phrase "Impact Statement" appears. Returns a tuple (page_number,

Locate the first occurrence of the word "References" (or variations like "R EFERENCES") within the cleaned content extracted from the PDF.

Serialize dataclasses (such as Journals) to JSON.

Edit the bfts_config.yaml file to point to the idea.md file Args: config_path: Path to the bfts_config.yaml file idea_dir: D

Encode image data to base64 string.

Generates a string that summarizes an exception and its stack trace (either in standard python repl or in IPython format).

unzips all .zip files within `path` and cleans up task dir [TODO] handle nested zips

Extract the number of cleaned text lines for each page from first_page to last_page. This uses pdftotext with layout preservation and the cle

Generate a tree structure of files in a directory

Generate visual layout of graph

Get number of available NVIDIA GPUs without using torch

Get all interactions, optionally filtered by model.

Get summary of token usage and costs for all models.

Convert a dictionary into a markdown file. Args: data: Dictionary containing the data to convert output_path: Path where the

Returns True if the line is likely a header or footer. Filters out: - Lines that are too short (< 4 characters after stripping).

Check if a script is a valid Python script.

Load the folders that start with 'stage_' followed by a number. Args: base_path (str): The base directory path where sta

Load task description from markdown file or config str.

Deserialize JSON to AIDE dataclasses.

Normalize layout to [0, 1]

Generate a natural language plan + code in the same LLM call and split them apart.

Setup the agent's workspace and preprocess data if necessary.

Generate a textual preview of a csv file Args: p (Path): the path to the csv file file_name (str): the file name to use in the pr

Generate a textual preview of a json file using a generated json schema

Run the experiment through generated stages