Use onnxmltools>=1.6.0,<=1.11.0 instead of onnxmltools>=1.6.0

[mshr-h]

import onnxmltools will fail with
onnxconverter_common==1.8.1 & onnxmltools==1.11.1. The failure caused ONNX test skipping.
In this PR, we'll use onnxmltools>=1.6.0,<=1.11.0 instead of onnxmltools>=1.6.0.

Fix #589

[codecov-commenter]

Codecov Report

Merging #592 (6ab11c2) into main (2fe2e35) will increase coverage by 9.82%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##             main     #592      +/-   ##
==========================================
+ Coverage   80.47%   90.30%   +9.82%     
==========================================
  Files          78       78              
  Lines        4569     4569              
  Branches      948      948              
==========================================
+ Hits         3677     4126     +449     
+ Misses        723      252     -471     
- Partials      169      191      +22     

Flag	Coverage Δ
unittests	90.15% <ø> (+9.80%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...ummingbird/ml/operator_converters/_gbdt_commons.py	91.78% <0.00%> (-2.74%)	⬇️
hummingbird/ml/_utils.py	88.06% <0.00%> (+0.56%)	⬆️
hummingbird/ml/supported.py	93.04% <0.00%> (+2.60%)	⬆️
...tor_converters/_one_hot_encoder_implementations.py	96.22% <0.00%> (+3.77%)	⬆️
hummingbird/ml/_topology.py	85.59% <0.00%> (+4.93%)	⬆️
...mmingbird/ml/containers/sklearn/onnx_containers.py	89.84% <0.00%> (+6.25%)	⬆️
hummingbird/ml/_executor.py	84.50% <0.00%> (+7.04%)	⬆️
...erters/_array_feature_extractor_implementations.py	100.00% <0.00%> (+9.52%)	⬆️
...l/operator_converters/_pipeline_implementations.py	78.94% <0.00%> (+10.52%)	⬆️
hummingbird/ml/_parse.py	87.40% <0.00%> (+11.08%)	⬆️
... and 13 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2fe2e35...6ab11c2. Read the comment docs.

[ksaur]

This is great, thank you for finding the issue!! Not even related to your PR/TVM, but just unlucky timing.

This work for me (back to 90%), what do you think @interesaaat ?

[interesaaat]

This is great, thanks for finding this. Let's merge it.

[ksaur]

We had to pin it to that exact version of skltoonnx because of #533. :(

I see that when you merged it, it passed on python 3.7 but failed on python 3.8?

I'm confused how it passed before... was it bumping to skl==1.1.1 that broke things for you? Meaning, if you set everything back to f452692, is it ok then?

[mshr-h]

On Python 3.7, pip install git+https://github.com/onnx/sklearn-onnx.git@9b551d8ecfd9f97ad8771909e34cfffea7b6e99d installs scikit-learn-1.0.2. But on Python 3.8, it installs scikit-learn-1.1.1.
I don't know why...

[mshr-h]

I think bumping to skl==1.1.1 broke something. With skl==1.0.2, CI has passed. Should we revert 2fe2e35?
What do you think? @ksaur @interesaaat

[mshr-h]

Another option is that we first go back to skl=1.0.2, then merge this commit. After that bump to skl=1.1.0.

[ksaur]

Thanks so much for all the time you put into this; I really appreciate it!

I don't want to revert the last commit entirely because it had fixes we'll need moving forward, but we can definitely go back to skl1.0.2 for now.