Upgrade to torch 1.7.1, 1.8.1, 1.9.0 #332
Comments
|
Hi @erickeagle, Torch 1.7 had a known issue that prevented us from upgrading. I'll check whether it has been resolved by now, and else find a workaround. |
ourownstory
changed the title
|
Any updates? Can we help? |
facebook-github-bot
pushed a commit
to facebookresearch/Kats
that referenced
this issue
Sep 20, 2021
Summary: After D30550349 (5f44b20) and D30652610 (f1694ef), which introduced `neuralprophet` to Kats, a number of issues arose with the build, mostly stemming from the known issue that `neuralprophet` only supports older `torch` versions (ourownstory/neural_prophet#332). The dependency web was quite complex, as some issues arose from dependencies of dependencies of dependencies. All my solutions are detailed below: 1. The `pip install -r test_requirements.txt` failed as `pip` could not resolve the added dependency complexities from downgrading the required `torch` version in D30884622 (8d0b005). This is a known issue in newer pip versions (pypa/pip#9215), so this diff forces the Github test to use the legacy resolver. 2. `torch>=1.4.0` was still resolving to `torch==1.8.0`, so this diff forces a version downgrade to `torch<1.7.0` which is compatible with `neuralprophet`. 3. Subsequently, the `gpytorch` dependency was failing to install due to it requiring a more recent `torch` version, so this diff downgrades the required `gpytorch` version to `1.2.1` which will still accept `torch==1.6.0`. 4. The newest version of `ax-platform` has `botorch==0.5.1` as a requirement, and this version of `botorch` requires `gpytorch>=1.5.1`, which is incompatible with #3. As such, this diff forces `ax-platform==0.1.18`, which will install `botorch>=0.3.2`, which accepts `gpytorch>=1.2`. 5. Nonetheless, despite #4, sometimes `pip` would still install `botorch==0.5.1`, as it technically satisfies the `ax-platform==0.1.18` requirement of `botorch>=0.3.2`. As such, this diff explicitly adds `botorch==0.3.2` as a dependency and places it before the `ax-platform` installation, ensuring that the correct version of `botorch` is installed, thus allowing the dependencies to resolve. Reviewed By: michaelbrundage Differential Revision: D31044893 fbshipit-source-id: 9152fe04da199dd0061472ea60b302ed3945238f
|
Hi @erickeagle and @alonme We had relaxed the requirements a while back on master, but had not updated the package on PYPI. Please kindly check whether the issue is resolved for you. Otherwise please re-open issue. Thank you! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
PS C:\Users\kalas> pip install neuralprophet
Collecting neuralprophet
Using cached neuralprophet-0.2.7-py3-none-any.whl (56 kB)
Requirement already satisfied: matplotlib>=2.0.0 in k:\anaconda3\envs\my\lib\site-packages (from neuralprophet) (3.4.2)
Collecting LunarCalendar>=0.0.9
Using cached LunarCalendar-0.0.9-py2.py3-none-any.whl (18 kB)
Collecting torch-lr-finder>=0.2.1
Downloading torch_lr_finder-0.2.1-py3-none-any.whl (11 kB)
Collecting neuralprophet
Using cached neuralprophet-0.2.6-py3-none-any.whl (55 kB)
Using cached neuralprophet-0.2.5-py3-none-any.whl (54 kB)
ERROR: Cannot install neuralprophet==0.2.5, neuralprophet==0.2.6 and neuralprophet==0.2.7 because these package versions have conflicting dependencies.
The conflict is caused by:
neuralprophet 0.2.7 depends on torch<1.7.0 and >=1.4.0
neuralprophet 0.2.6 depends on torch<1.7.0 and >=1.4.0
neuralprophet 0.2.5 depends on torch<1.7.0 and >=1.4.0
To fix this you could try to:
ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/user_guide/#fixing-conflicting-dependencies
PS C:\Users\kalas> pip install torch==1.6.0
ERROR: Could not find a version that satisfies the requirement torch==1.6.0 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2, 1.7.1, 1.8.0, 1.8.1)
ERROR: No matching distribution found for torch==1.6.0
The text was updated successfully, but these errors were encountered: