https://github.com/diegofps/pywup
Parallelize any experiment and collect results from their stdout
https://github.com/diegofps/pywup
cluster-computing multiprocessing plotting regex
Last synced: 28 days ago
JSON representation
Parallelize any experiment and collect results from their stdout
- Host: GitHub
- URL: https://github.com/diegofps/pywup
- Owner: diegofps
- License: mit
- Created: 2019-05-16T23:10:24.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2020-03-11T14:04:24.000Z (almost 6 years ago)
- Last Synced: 2025-09-29T09:02:12.251Z (4 months ago)
- Topics: cluster-computing, multiprocessing, plotting, regex
- Language: Python
- Homepage:
- Size: 428 KB
- Stars: 2
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# pywup
A small set of tools.
TODO: Expand this [markdown](https://guides.github.com/features/mastering-markdown/)
# How to install / update
```bash
pip3 install --upgrade pywup --user --no-cache-dir
```
If you cannot run the wup command after installation, check if the path to the folder .local/bin is in your PATH. If not, add the following to the end of your ~/.bashrc:
```
export PATH=/home//.local/bin:$PATH
```
# The library
## configure
This is a python library for creating template-based configure scripts.
```python
#!/usr/bin/env python3
from pywup.configure import *
mf = TemplateBuilder()
mf.compiler = find_program_or_abort(["clang++", "g++"], "Compiler", "clang")
mf.python = find_header_or_abort("m/Python.h", "Python header", "python3-dev")
mf.highgui = find_header_or_abort("/highgui.hpp", "Opencv's highgui", "opencv-dev")
mf.imgproc = find_header_or_abort("/imgproc.hpp", "Opencv's imgproc", "opencv-dev")
mf.valgrind_py = find_file_or_abort("python3-devel/valgrind-python.supp", "valgrind suppression file for python3", "python3-dev")
mf.libimgcodecs = find_lib("libopencv_imgcodecs.so", "libopencv_imgcodecs")
mf.libs = "-lopencv_core -lopencv_highgui -lopencv_imgproc -I../../wup/cpp/include"
mf.headers = "-DPYTHON_H=$(PYTHON_H) -DHIGHGUI_H=$(HIGHGUI_H) -DIMGPROC_H=$(IMGPROC_H)"
if mf.libimgcodecs:
mf.libs += " -lopencv_imgcodecs"
mf.build("Makefile", """\
CC={compiler}
HIGHGUI_H={highgui}
PYTHON_H={python}
IMGPROC_H={imgproc}
VALGRIND_PYTHON={valgrind_py}
LIBS = {libs}
HEADERS= {headers}
all:
$(CC) -fPIC -shared wup_wrapper.cpp -o libwup.so -Wall -O3 -std=c++11 $(LIBS) $(HEADERS)
run:
$(CC) main.cpp -o main -Wall -O3 -std=c++11 $(LIBS) $(HEADERS)
./main
debug:
$(CC) main.cpp -o main -Wall -g -std=c++11 $(LIBS) $(HEADERS)
gdb main
valgrind:
$(CC) -fPIC -shared wup_wrapper.cpp -o libwup.so -Wall -O1 -g -std=c++11 $(LIBS) $(HEADERS)
valgrind --leak-check=full --show-leak-kinds=all --track-origins=yes --verbose --suppressions=$(VALGRIND_PYTHON) python3 main3.py --model wisard --dataset mnist 2> valgrind.out
""")
```
# wup
This is a command line utility for running experiments, plotting and executing repetitive tasks.
## collect
collect runs a program multiple times and collects outputs.
Example 1 - Permutate variables and collect outputs
* This will permutate the variables THREADS (arithmetic: 1, 3, 5, 7, ...) and JOBS (geometric: 1, 1.6, ...)
* For each configuration, the program will run 10 times
* For each execution, the following attributes will be collected: TRAIN_TIME, TEST_TIME and ACC
* The output will be saved in CSV format at wespa.csv
* The permutations values for each variable are passed to the program in the same order they are declared. Each "{}" is replaced by the variable value.
```bash
wup collect \
--p TRAIN_TIME "Info: Time to train was ([-0-9\.e\+]+) μs" \
--p TEST_TIME "Info: Time to test was ([-0-9\.e\+]+) μs" \
--p ACC "Test Acc = ([0-9\.]+)" \
--runs 10 \
--va THREADS 1 17 2 \
--vg JOBS 1 80000 1.6 \
--o wespa.csv \
--c "./wisard -createModel RamWisard -ramBits 28 -decoder classic -ramType prime \
-train mnist ../data/emnist/byclass/emnist-byclass-train \
-test mnist ../data/emnist/byclass/emnist-byclass-test \
-times -numThreads {} -hashSize 18041 -jobsPerThread {} -bleaching Y -pPredict 2"
```
Example 2 - Collect multiple rows from the same command execution
* The parameter --n defines a line break, allowing us to collect multiple rows for the same execution.
* The parameter --log defines an output file to write extra details during the process.
```bash
wup collect \
--n "[a-zA-Z0-9]+ : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p DS_NAME "([a-zA-Z0-9]+) : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p INTERSECTION "[a-zA-Z0-9]+ : AvgIntersectionScore = ([-0-9\.e\+]+) AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p CENTER_DISTANCE "[a-zA-Z0-9]+ : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = ([-0-9\.e\+]+)" \
--runs 10 \
--o "./collect.csv" \
--log "./collect.log" \
--c "./run_all.sh"
```
Example 3 - Multiple commands and parallelization
* Specify multiple --c to call multiple commands. When multiple commands are presented they will receive the same variables.
* Their output will also be concatenated, you probably want to use --n to separate them in distinct lines.
* The program will parse all their outputs as if they were continuous, one after the other.
* Use --jobs to define the number of parallel processes.
* Parallelization is be applied across parameter permutation and commands. If you have 16 permutations and 2 commands we have 32 parallel tasks.
* You may also use --jobs without multiple commands.
* You may also use --c with just one --jobs.
* Use parallelism to tune parameters and single process to measure times. When you are measuring execution time, multiple processes may slightly slow each other as they compete for resources.
```bash
wup collect \
--n "[a-zA-Z0-9]+ : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p DS_NAME "([a-zA-Z0-9]+) : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p INTERSECTION "[a-zA-Z0-9]+ : AvgIntersectionScore = ([-0-9\.e\+]+) AvgCenterDistanceScore = [-0-9\.e\+]+" \
--p CENTER_DISTANCE "[a-zA-Z0-9]+ : AvgIntersectionScore = [-0-9\.e\+]+ AvgCenterDistanceScore = ([-0-9\.e\+]+)" \
--jobs 4 \
--runs 10 \
--o "./collect.csv" \
--log "./collect.log" \
--c "./run.sh 'Basketball'" \
--c "./run.sh 'Biker'" \
--c "./run.sh 'Bird1'" \
--c "./run.sh 'Bird2'"
```
## heatmap
heatmap receives a csv file and generates a heatmap.
```bash
wup heatmap \
--data ./wespa.csv \
--y "THREADS" \
--x "JOBS" \
--z "TEST_TIME" \
--tz "data[0,z] / data[i,z]" \
--tx "\"%d\" % int(float(data[i,x]))" \
--ty "\"%d\" % int(float(data[i,y]))" \
--tzz "\"%.2f\" % data[i,j]" \
--title "Speedup (threads / blockSize)" \
--size 10 4 \
--o heatmap_wespa.png
```
## bars
Generates a bar graphic using one or more csv files.
```bash
wup bars \
--load ./note.csv \
--line THREADS TEST_TIME "note test" \
--line THREADS TRAIN_TIME "note train" \
\
--load ./out.csv \
--line THREADS TEST_TIME "wespa test" \
--line THREADS TRAIN_TIME "wespa train"\
\
--title "Predict speedups (threads / speedup)" \
--ty "y[0,0] / y[i,j]" \
--ts "y[0,0]*s[i,j] / (y[i,j]**2)" \
--tyy "\"%.2f\" % ty[i,j] if ty[i,j] else ''" \
--tx "int(float(x[i]))" \
--xlabel "Threads" \
--ylabel "Speedup" \
--barwidth 0.9 \
--size 10 4 \
--verbose \
--o bars_parallelPredictSpeedup.png
```
## backup
Example ***~/Dropbox/backups/system/wup.bak*** file
```
file;~/.local/bin/macro_play;./local_bin/
file;~/.local/bin/macro_rec_start;./local_bin/
file;~/.local/bin/macro_rec_stop;./local_bin/
folder;~/.config/compton;./compton
file;~/.vimrc;./vimrc
folder;~/.config/i3;./i3
```
Invoke backup / restore
```bash
# This will copy files from system (left) to backup folder (right), overwriting any change or previous file in the backup folder.
wup backup create ~/Dropbox/backups/system/wup.bak
# This will copy files from the backup folder (right) to the system (left), overwriting any change or previous file in the system.
wup backup restore ~/Dropbox/backups/system/wup.bak
# This will sync all files and backup folders, copying the most recent one to the right direction.
wup backup sync ~/Dropbox/backups/system/wup.bak
```