Description
Definitive Next Step: Contact Coiled Support
You have a very clear case to present to them:
Goal: Install a recent version of pandas-ta (e.g., latest stable, or at least >=0.3.70) in a Python 3.12 Coiled software environment.
Problem: The old pandas-ta==0.3.14b0 causes an ImportError: cannot import name 'NaN' from 'numpy' because it’s incompatible with modern NumPy’s numpy.nan convention.
Attempts Made:
Specified "pandas-ta" (for latest) in the pip section of coiled.create_software_environment.
Ensured no local requirements.txt is pinning the version.
Used new, unique software environment names for each build attempt.
Consistent Undesired Outcome: Coiled’s build process repeatedly installs pandas-ta==0.3.14b0, even when latest is requested and no local files should be interfering.
Provide Logs:
The successful build log that shows + pandas-ta==0.3.14b0 being installed (like the one you just posted: https://cloud.coiled.io/software/alias/156461/build/151710?account=alex-martin-bee&tab=logs).
The failed build log when you explicitly requested pandas-ta>=0.3.70 which stated “only pandas-ta==0.3.14b0 is available” (https://cloud.coiled.io/software/alias/156459/build/151708?account=alex-martin-bee&tab=logs).
Question for Coiled: “Why does the build system consistently resolve pandas-ta to version 0.3.14b0 even when a newer/latest version is requested and no local requirements.txt is present? Is there an internal index or configuration that
52EC
is pinning or limiting the available versions of pandas-ta for pip installs?”