A multi-tasks cross validation scikit-learn SVM recipe on DTIC HPC

boundaryPatternClassification.sh

#!/bin/sh
# Name the process
# ----------------
#$ -N compIden_$JOB_ID
#
# Call from the current working directory; no need to cd; using default.q configuration, HPC will distribute the best choice of the nodes
# ------------------------------------------------------
#$ -cwd
# -q default.q
#
# Max time limits, s_rt soft running time, h_rt hard running time
# ---------------
#$ -l s_rt=5:00:00
#$ -l h_rt=10:00:00
#
# Output/Error Text
# ----------------
#$ -o ./python.$JOB_ID.out
#$ -e ./python.$JOB_ID.err
#
# Create an array job = !!!!!!number of audio in the target folder!!!!!!, create 49 tasks, every task corresponds to a parameter setting for SVM
# ----------------
#$ -t 1-49:1
#
# Send me a mail when processed and when finished:
# ------------------------------------------------
#$ -M yourname@upf.edu
#$ -m bea
#
# Start script
# --------------------------------


printf "Starting execution of job $JOB_ID from user $SGE_O_LOGNAME at `date`\n"

# force UTF 8
export LANG="en_US.utf8"

# change python version
module load python/2.7.5

# enter virtualenv, change this to your virtualenv name, yourenv folder is suppose to be at the previous level of the working folder
source ../yourenv/bin/activate

# test import sklearn
python svm_cv.py ${SGE_TASK_ID}

# Print job Done
printf "Job $JOB_ID done at `date`\n"

svm_cv.py

# python code for parameter search

from sklearn.svm import SVC
from sklearn.model_selection import cross_val_score,StratifiedKFold
import numpy as np
import sys

def svm_cv(fv_train,target_train,C,gamma):
	clf 	= SVC(kernel='rbf', gamma=gamma, C=C)
	mycv 	= StratifiedKFold(n_splits=5)
	scores 	= cross_val_score(clf,fv_train,target_train,cv=mycv,scoring='recall_weighted')
	print("Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))

if __name__ == '__main__':

	# 49 parameter settings
	parameters = [[0.001, 0.001], [0.001, 0.01], [0.001, 0.1], [0.001, 1], [0.001, 10], [0.001, 100], [0.001, 1000], [0.01, 0.001], [0.01, 0.01], [0.01, 0.1], [0.01, 1], [0.01, 10], [0.01, 100], [0.01, 1000], [0.1, 0.001], [0.1, 0.01], [0.1, 0.1], [0.1, 1], [0.1, 10], [0.1, 100], [0.1, 1000], [1, 0.001], [1, 0.01], [1, 0.1], [1, 1], [1, 10], [1, 100], [1, 1000], [10, 0.001], [10, 0.01], [10, 0.1], [10, 1], [10, 10], [10, 100], [10, 1000], [100, 0.001], [100, 0.01], [100, 0.1], [100, 1], [100, 10], [100, 100], [100, 1000], [1000, 0.001], [1000, 0.01], [1000, 0.1], [1000, 1], [1000, 10], [1000, 100], [1000, 1000]]
	
	# pass the first arg as the parameter setting index
	index = int(sys.argv[1])
	index -= 1
	if index<0 or index>48:
		print "this parameter doesn''t exist"

	p 		= parameters[index]
	C 		= p[0]
	gamma 	= p[1]
	print 'C:',C,'gamma:',gamma

	fv_train 		= np.load('fv_train.npy')
	target_train 	= np.load('target_train.npy') 

	# start to run cross validation with parameter C and gamma
	svm_cv(fv_train,target_train,C,gamma)

svm_dtic_hpc.md

We want to do the cross_validation of a binary classification task by using the scikit-learn SVM

1. we log firstly into hpc
   ssh user_name@hpc.dtic.upf.edu

2. change the python environment to PYTHON/2.7.5, for some reason I don't know, I can't install any python packages on default PYTHON 2.7.3 environment
   module load PYTHON/2.7.5

3. make a virtual environment, enter into this environment, and install scikit-learn
   virtualenv yourenv
   source yourenv/bin/activate
   pip install --upgrade pip
   pip install numpy,scipy,scikit-learn

4. there should be four files in your working folder, they are:

• boundaryPatternClassification.sh: bash which execute jobs in HPC
• svm_cv.py: python script for running the cross validation of the scikit-learn SVM
• fv_train.npy: feature vectors for training the SVM model download it
• target_train.npy: target vector for training the SVM model download it

5. the explaination of the .sh and .py are inside the scripts.

6. to run this recipe:
   qsub boundaryPatternClassification.sh

7. to view the stat of the running jobs, use qstat. The error log and output log will appear in your directory as .err and .out.