paypal / autosklearn-zeroconf Goto Github PK

possible candidate for replacement fit_transformer()

As it mentioned in the https://github.com/paypal/autosklearn-zeroconf

Data Format
The code uses a pandas dataframe format to manage the data. It is stored in the HDF5 .h5 file for convenience. (Python module "tables")

It is a bit unclear for me. when I check my data with the example data Adult.h5, the structure looks different.

Is there any rules or suggestions, that users have to follow? thank you.

autosklearn.metrics.f1 replaces F1_METRIC

CSV is a bit tricky because it does not have data type metadata. So in order to take that in we will have to guesstimate that or get this metadata in some other way. The default way would be to as the user to specify this information but then it won't be "zeroconf" anymore.
One way of approaching this would be https://github.com/bashalex/datapot

lib/python3.5/site-packages/sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.
"This module will be removed in 0.20.", DeprecationWarning)

freshly installed CentOS7

 python3 ./bin/zeroconf.py -d ./data/Adult.h5

Traceback (most recent call last):
  File "./bin/zeroconf.py", line 22, in <module>
    import d2khelper as d2k
  File "/home/username/autosklearn-zeroconf/bin/d2khelper.py", line 48
    return {**parameter, **param}
             ^
SyntaxError: invalid syntax

@d2k do you you think you could look into this?

@d2k what should we list here instead of a leftover from your environment?

https://github.com/szilard/benchm-ml
https://github.com/rambler-digital-solutions/criteo-1tb-benchmark/

automl/auto-sklearn#263

Really nitpicky remark, but it doesn't make sense to interpolate eval results between two different models, so a line plot is not appropriate (the wild drops/increases in loss make it hard to read as well). A scatter plot would convey the idea better I think!

Currently auto-sklearn=2.0 is required
Test auto-sklearn 3.0 successfull with the test dataset.