Practical scenarios where FrameX fits today.
FrameX Use Cases
1. Local ETL for Analytics Teams
Use FrameX to process Parquet datasets on a single workstation before loading into BI systems.
Why it fits:
- Arrow-native IO
- familiar dataframe API
- efficient local execution for medium-large files
2. Feature Engineering Before Model Training
Use FrameX for cleaning, grouping, and joining tabular data before exporting arrays/dataframes for ML tooling.
Why it fits:
- fast tabular transforms
- direct conversion to Pandas/NumPy
- chunked arrays for large vectors
3. Replacing Slow Single-Threaded Pandas Scripts
If a script is CPU-bound and spending time in repetitive filters/groupbys/sorts, FrameX can be a migration path that keeps dataframe ergonomics.
Migration pattern:
- replace
import pandas as pdwithimport framex as fxfor target segments - keep logic shape similar
- compare outputs via
.to_pandas()during rollout
4. Interop Bridge Between Ecosystems
FrameX can sit between Arrow-heavy and Pandas/NumPy-heavy parts of a pipeline.
Useful boundaries:
from_pandas(...)from_dask(...)from_ray(...)from_dataframe(...).to_arrow()for Arrow-first systems.to_dask()/.to_ray()for pipeline handoff
5. Incremental Adoption in Existing Codebases
Adopt only the heaviest stages first, then expand.
Suggested order:
- IO and wide aggregations
- joins and sorts
- optional lazy pipelines for long transformations
When Not to Use FrameX (Yet)
- strict requirement for complete Pandas API parity
- workloads that require full multi-node distributed orchestration
- pipelines dominated by custom Python object logic where Arrow typing is minimal