It seems the current implementation will cut the tensor to min or max according to their values, which might be a problem when debugging logical errors or typos. Maybe it would be better to just throw an exception? Also it would be nice to note that in the documentation.

>>> import torch
>>> print(torch.__version__)
1.6.0
>>> a

At this moment relu_layer op doesn't allow threshold configuration, and legacy RELU op allows that.
We should add configuration option to relu_layer.

Problem:
catboost version: 0.23.2
Operating System: all
Tutorial: https://github.com/catboost/tutorials/blob/master/custom_loss/custom_metric_tutorial.md

Impossible to use custom metric (С++).

Code example

from catboost import CatBoost
train_data = [[1, 4, 5, 6],
 

Hi ,

I have tried out both loss.backward() and model_engine.backward(loss) for my code. There are several subtle differences that I have observed , for one retain_graph = True does not work for model_engine.backward(loss) . This is creating a problem since buffers are not being retained every time I run the code for some reason.

Please look into this if you could.

Improve readability of thread id based branches by giving them more descriptive names.

e.g.

if (!t) // is actually a t == 0

and

https://github.com/rapidsai/cudf/blob/57ef76927373d7260b6a0eda781e59a4c563d36e/cpp/src/io/statistics/column_stats.cu#L285
Is actually a lane_id == 0
As demonstrated in rapidsai/cudf#6241 (comment), pr

Current implementation of join can be improved by performing the operation in a single call to the backend kernel instead of multiple calls.

This is a fairly easy kernel and may be a good issue for someone getting to know CUDA/ArrayFire internals. Ping me if you want additional info.

We would like to forward a particular 'key' column which is part of the features to appear alongside the predictions - this is to be able to identify to which set of features a particular prediction belongs to. Here is an example of predictions output using the tensorflow.contrib.estimator.multi_class_head:

{"classes": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"],
 "scores": [0.068196

Would it make sense to add a variadic overload of make_zip_iterator that composes the existing make_zip_iterator with make_tuple? I have this in my own code, and I find that it reduces syntactic overhead.

template<typename... Iterators>
__host__ __device__
  zip_iterator<thrust::tuple<Iterators...>> make_zip_iterator(thrust::tuple<Iterators...> t)
{
    return zip_iterator<thrust::tupl

Hey everyone!

mapd-core-cpu is already available on conda-forge (https://anaconda.org/conda-forge/omniscidb-cpu)

now we should add some instructions on the documentation.

at this moment it is available for linux and osx.

some additional information about the configuration:

1. for now, always install omniscidb-cpu inside a conda environment (also it is a good practice), eg: