https://github.com/kmkolasinski/approxkernel

approxkernel
https://github.com/kmkolasinski/approxkernel

Last synced: 22 days ago
JSON representation

approxkernel

• Host: GitHub
• URL: https://github.com/kmkolasinski/approxkernel
• Owner: kmkolasinski
• Created: 2019-05-26T06:46:04.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-03-30T19:07:56.000Z (almost 5 years ago)
• Last Synced: 2024-10-29T18:23:44.845Z (2 months ago)
• Language: Fortran
• Size: 1.4 MB
• Stars: 1
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# ApproxKernels

## Non Conda way of installation:

* Install python3.6 and pip3, this can be done via:

```bash

sudo apt-get install python3.6

sudo apt install python3-pip

sudo apt-get install python3.6-tk

```

* Then install python lib with `install_python_lib.sh`

## Install with conda envrionment

* Using conda enviroments is more safe approach

* Create or use existing conda envionment. 

* Enable created conda environment e.g. `source activate kernels`

* Enter python_lib folder and run: `pip install .`

## Training custom kernel

This example shows how to train kernel from exising predefined configuration

* Enter `tests/resources/coulomb1d_kernel` folder

* Check content of config.yml and kernel.py files

* `config.yml` - contains grid definition, number of scales and training schedule

* `kernel.py` - contains `def kernel_fn(r: float) -> float:` which contains definition of your kernel function. This function must accept scalar float value and return scalar

value of kernel function. For example f(r) = 1/r. Make sure that there are no singularities.

* Open terminal in this folder. Make sure you are in correct conda environment.

* Just in case run: `source activate kernels` 

* Run `fit_kernel` command, wait ...

* Use the created kernel files in your fortran code:

```fortran

USE modutils

USE modio

USE modapproxkernel2d

INTEGER, PARAMETER :: width = 128, height = 128

REAL(KIND=8), DIMENSION(width, height) :: input_x, output_x

REAL(KIND=8), DIMENSION(:, :, :), ALLOCATABLE :: kernels

TYPE(ApproxDKernel2D) :: kernelOp

CALL read_kernels_2d(&

    "path/kernels_1D_scales=4_size=33_grid=128_loss=X.XXXX.txt", 4, 33, kernels)

CALL initapproxkernel2d(kernelOp, kernels=kernels, input_shape=[width, height])

CALL execapproxkernel2d(kernelOp, input_x, output_x)

CALL deleteapproxkernel2d(kernelOp)

```

## Running fortran examples with multiple threads using MKL threading:

* Set enviromental variable:

```

export MKL_NUM_THREADS=8

export OMP_NUM_THREADS=8 

```

## Profiling code

* Compile main code with C=PROFILE option

```

cd src

make clean

make C=PROFILE

cd -

```

* Compile test with the same option

```

cd tests

make clean && make test_profile C=PROFILE

./test_profile

loopprofileviewer.sh loop_prof_1579625645.xml

```