h2o4gpu

h2o4gpu copied to clipboard

Current type checking mechanism is too restrictive. We need something more generic that will allow the user use all the types SKLearn can use.

mdymczyk

And, fallback to CPU or sklearn if n_gpu=0 is set and we have no CPU algorithm except scikit-learn's. E.g, if CPU lib not found for kmeans, revert to sklearn.

pseudotensor

* Add new pipeline step that will run small performance test

navdeep-G

And jenkins results should be used to add asserts. Can time performance with entire fit+predict but compare result with just a fast call to sklearn fit(1 iteration) + predict.

pseudotensor

http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LinearRegression.html#sklearn.linear_model.LinearRegression Parameters: | X : array-like, shape = (n_samples, n_features)Test samples.y : array-like, shape = (n_samples) or (n_samples, n_outputs)True values for X.sample_weight : array-like, shape = [n_samples], optionalSample weights. --...

pseudotensor

Jenkins do all make tests

3

Separate builds for make test, make testbig, make testperf, make testbigperf . Jenkins should use dotest, dotestbig, dotestperf, dotestbigperf

pseudotensor

xgboost wheel should be built and tested and used separately

1

To speed-up build process: 1) Wheel should always be build from https://github.com/h2oai/xgboost/tree/h2oai and pushed to private site. 2) We need a separate jenkins job that then takes that wheel and...

pseudotensor

Need to have make testperf (with it's artifacts correctly grabbed) done. Also need to buy a volta machine.

pseudotensor

isolate openblas/GPU things if only doing CPU scoring.

2

``` We are evaluating DAI and I am having issues getting the scoring pipeline to run on a machine. Using the instructions at http://docs.h2o.ai/driverless-ai/latest-stable/docs/userguide/scoring-package.html I spun up a AWS ec2...

pseudotensor

lightgbm on ppc64le

1

sh1ng