MAINT: Failing CI

[mgxd]

This updates the base images for our circle tests, and bumps the numpy and scipy versions as the one we're currently using is raising an error.

Given that this is the LTS branch, are we comfortable moving to a different minor of the above packages?

[mgxd]

Okay, I looked into this a little more.

A freeze of 20.2.1 (LTS) python versions

alabaster==0.7.12
annexremote==1.4.3
appdirs==1.4.4
asn1crypto==0.24.0
async-generator==1.10
attrs==20.3.0
Babel==2.8.0
bids-validator==1.5.7
bleach==3.2.1
boto==2.49.0
cached-property==1.5.2
certifi==2020.6.20
cffi==1.11.5
chardet==3.0.4
ci-info==0.2.0
citeproc-py==0.5.1
click==7.1.2
codecov==2.1.10
conda==4.5.11
coverage==5.3
cryptography==2.3.1
cycler==0.10.0
Cython==0.29.21
datalad==0.13.5
decorator==4.4.2
defusedxml==0.6.0
Deprecated==1.2.10
docopt==0.6.2
docutils==0.16
duecredit==0.8.0
entrypoints==0.3
etelemetry==0.2.2
fasteners==0.15
filelock==3.0.12
fmriprep==20.2.1
future==0.18.2
h5py==3.1.0
humanize==3.1.0
idna==2.7
imageio==2.9.0
imagesize==1.2.0
importlib-metadata==2.0.0
indexed-gzip==1.3.2
iniconfig==1.1.1
ipython-genutils==0.2.0
iso8601==0.1.13
isodate==0.6.0
jeepney==0.4.3
Jinja2==2.11.2
joblib==0.17.0
jsmin==2.2.2
jsonschema==3.2.0
jupyter-client==6.1.7
jupyter-core==4.6.3
jupyterlab-pygments==0.1.2
keyring==21.4.0
keyrings.alt==4.0.0
kiwisolver==1.2.0
lockfile==0.12.2
lxml==4.6.1
MarkupSafe==1.1.1
matplotlib==2.2.2
mistune==0.8.4
mkl-fft==1.0.6
mkl-random==1.0.1
monotonic==1.5
msgpack==1.0.0
nbclient==0.5.1
nbconvert==6.0.7
nbformat==5.0.8
nbsphinx==0.8.0
nest-asyncio==1.4.2
networkx==2.5
nibabel==3.2.0
niflow-nipype1-workflows==0.0.4
nilearn==0.6.2
nipype==1.5.1
nitime==0.8.1
nitransforms==20.0.0rc3
niworkflows==1.3.2
num2words==0.5.10
numpy==1.15.4
packaging==20.4
pandas==0.23.4
pandocfilters==1.4.3
patool==1.12
patsy==0.5.1
Pillow==8.0.1
pluggy==0.13.1
pockets==0.9.1
prov==2.0.0
psutil==5.7.3
py==1.9.0
pybids==0.12.3
pycosat==0.6.3
pycparser==2.18
pydot==1.4.1
pydotplus==2.0.2
PyGithub==1.53
Pygments==2.7.2
PyJWT==1.7.1
pyOpenSSL==18.0.0
pyparsing==2.4.7
pyrsistent==0.17.3
PySocks==1.6.8
pytest==6.1.2
pytest-env==0.6.2
python-dateutil==2.8.1
pytz==2020.1
PyWavelets==1.1.1
PyYAML==5.3.1
pyzmq==19.0.2
rdflib==5.0.0
requests==2.19.1
ruamel-yaml==0.15.46
scikit-image==0.17.2
scikit-learn==0.19.1
scipy==1.1.0
sdcflows==1.3.3
seaborn==0.11.0
SecretStorage==3.1.2
sentry-sdk==0.19.2
simplejson==3.17.2
six==1.15.0
sklearn==0.0
smriprep==0.7.0
snowballstemmer==2.0.0
Sphinx==3.3.0
sphinx-argparse==0.2.5
sphinx-rtd-theme==0.5.0
sphinxcontrib-applehelp==1.0.2
sphinxcontrib-devhelp==1.0.2
sphinxcontrib-htmlhelp==1.0.3
sphinxcontrib-jsmath==1.0.1
sphinxcontrib-napoleon==0.7
sphinxcontrib-qthelp==1.0.3
sphinxcontrib-serializinghtml==1.1.4
SQLAlchemy==1.3.20
svgutils==0.3.1
TBB==0.1
tedana==0.0.9a1
templateflow==0.6.3
testpath==0.4.4
threadpoolctl==2.1.0
tifffile==2020.10.1
toml==0.10.2
tornado==6.0.4
tqdm==4.51.0
traitlets==5.0.5
traits==4.6.0
transforms3d==0.3.1
urllib3==1.23
webencodings==0.5.1
Whoosh==2.7.4
wrapt==1.12.1
zipp==3.4.0

Looking at the docker build step more closely, we do have some incompatibilities...

ERROR: scikit-image 0.18.1 has requirement numpy>=1.16.5, but you'll have numpy 1.15.4 which is incompatible.
ERROR: tedana 0.0.9 has requirement scikit-learn>=0.22, but you'll have scikit-learn 0.19.1 which is incompatible.
ERROR: tedana 0.0.9 has requirement scipy>=1.3.3, but you'll have scipy 1.1.0 which is incompatible.

My inclination is to remove these by pinning to the following major/minors:

• numpy == 0.16.*
• scikit-image == 0.18.*
• scikit-learn == 0.22.*
• scipy == 1.3.*

WDYT?

[effigies]

I'm mostly okay relaxing the patch versions, but I'd really rather not keep pushing things. It seems that tedana 0.0.9 is doing most of the pushing, so what about the following?

• numpy == 1.15.*
• scikit-image == 0.17.*
• scikit-learn == 0.19.*
• scipy == 1.1.*
• tedana == 0.0.9a1

[mgxd]

I don't see how that will fix the problem we are currently running into:
https://app.circleci.com/pipelines/github/nipreps/fmriprep/80/workflows/a13b7179-22e9-4b2f-833d-223225d69bf1/jobs/493

Skimming through the sklearn issue, it seems upgrading numpy to 1.16.* is the way.

I'd prefer to get all the incompatibilities out of the way at once, but I'd settle for just numpy to get this (and #2400) in quicker.

[effigies]

I'm proposing using an older skimage that is known to be compatible with the numpy series we have...

[effigies]

I think we should still set tedana==0.0.9a1 because of ERROR: tedana 0.0.9 has requirement scipy>=1.3.3, but you'll have scipy 1.1.0 which is incompatible..

@tsalo Would be interested in your thoughts here.

[effigies]

Nonetheless, okay with merging as is.

[mgxd]

Opened #2402 - let's merge to get green again