gpu

🚀 Feature

Support to pickle a jitted function (or at least throw a TypeError when using protocol 0 and 1).

Motivation

Trying to pickle a jitted function either raises TypeError: cannot pickle 'torch._C.ScriptFunction' object when protocol>1 or far worse when using protocol=0 or protocol=1 python 3.9.5 dies with:

terminate called after throwing an instance of 'std::runt

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: the approximate method can still be slow for many trees
catboost version: master
Operating System: ubuntu 18.04
CPU: i9
GPU: RTX2080

Would be good to be able to specify how many trees to use for shapley. The model.predict and prediction_type versions allow this. lgbm/xgb allow this.

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.

Our users are often confused by the output from programs such as zip2john sometimes being very large (multi-gigabyte). Maybe we should identify and enhance these programs to output a message to stderr to explain to users that it's normal for the output to be very large - maybe always or maybe only when the output size is above a threshold (e.g., 1 million bytes?)

https://github.com/brendangregg/FlameGraph

For feature engineering tasks, I'd like to be able to determine whether a datetime is the beginning or end of a year, like I can in pandas.

import pandas as pd
​
s = pd.Series(["2021-02-27", "2020-03-31"], dtype="datetime64[ms]")
s.dt.is_year_end
0    False
1    False
dtype: bool

import pandas as pd
​
s = pd.Series(["2021-01-01", "2020-04-01"], dtype="datet

Describe the Problem

plot_model currently has the save argument which can be used to save the plots. It does not provide the functionality to decide where to save the plot and with what name. Right now it saves the plot with predefined names in the current working directory.

Describe the solution you'd like

We can have another argument save_path which is used whenever the `

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're seeing a lot of warnings on Linux, host compiler GCC 9.3.0 with -Wconversion.

For a basic example, compile:

#include <thrust/device_vector.h>
int main() {
  thrust::device_vector<int> a;
  return 0;
}

With: nvcc main.cu -Xcompiler=-Wconversion.

This will result in around 1000 lines of output warnings.
Tested with the Thrust versions in CUDA 11.0, 11.3, and the la