Environment Setup
The repository specifies Python ≥3.10, but does not indicate an upper bound or specific versions for dependencies such as pandas.
Moreover, on systems with Python 3.10+, installing pandas without a version constraint defaults to pandas ≥2.x, which can introduce compatibility issues with code written for the pandas 1.x API (as previously mentioned in Issue #7).
In my testing, the code works reliably with:
- OS: Ubuntu 24.04.4 LTS (with Python 3.12 pre-installed)
- Python: 3.11.10 (installed via pyenv)
- pandas: 1.5.3 (installed via pip)
Python 3.11 satisfies the ≥3.10 requirement while remaining compatible with the pandas 1.x series. I explicitly installed pandas 1.5.3, the final release of the pandas 1.5.x series, which appears to work reliably with the code. The same setup was also tested on WSL (Ubuntu) and works similarly.
RAM Requirements (CPU-Only)
Integration tests of API and CLI tested on CPU-only system with 16GB RAM based on installation setup:
- Default mode (2/5 skipped, 3/5 pass, all OK)
- CPU mode (Resource Exhaustion/ Out of Memory Error)
In practice, we may require at least ~32 GB RAM on CPU-only systems, since tensorflow attempts to allocate a tensor of shape [57600, 28800] (~6.6 GB) and the peak memory usage during loading exceeds the capacity of a 16 GB system, leading to process termination.
Note:
The taxonomy prediction workflow up to vectorization runs as expected. However, loading the subsequent model fails on CPU-only 16 GB RAM systems because of extremely large tensor allocations. The text prediction workflow runs without issues.
@SantiagoSanchezF
I have documented the setup instructions and the workflow path analysis in greater detail in the attachment below. I would really appreciate your insights or feedback.
setup_trapiche.pdf
Environment Setup
The repository specifies Python ≥3.10, but does not indicate an upper bound or specific versions for dependencies such as pandas.
Moreover, on systems with Python 3.10+, installing pandas without a version constraint defaults to pandas ≥2.x, which can introduce compatibility issues with code written for the pandas 1.x API (as previously mentioned in Issue #7).
In my testing, the code works reliably with:
Python 3.11 satisfies the ≥3.10 requirement while remaining compatible with the pandas 1.x series. I explicitly installed pandas 1.5.3, the final release of the pandas 1.5.x series, which appears to work reliably with the code. The same setup was also tested on WSL (Ubuntu) and works similarly.
RAM Requirements (CPU-Only)
Integration tests of API and CLI tested on CPU-only system with 16GB RAM based on installation setup:
In practice, we may require at least ~32 GB RAM on CPU-only systems, since tensorflow attempts to allocate a tensor of shape [57600, 28800] (~6.6 GB) and the peak memory usage during loading exceeds the capacity of a 16 GB system, leading to process termination.
Note:
The taxonomy prediction workflow up to vectorization runs as expected. However, loading the subsequent model fails on CPU-only 16 GB RAM systems because of extremely large tensor allocations. The text prediction workflow runs without issues.
@SantiagoSanchezF
I have documented the setup instructions and the workflow path analysis in greater detail in the attachment below. I would really appreciate your insights or feedback.
setup_trapiche.pdf