def svm():
# Training Parameters
# Defines how high the cost is of a misclassification
# versus making the decision plane more complex.
# Low COST makes decisions very simple but creates classification errors
COST = 0.9
# Used for generalisation
# - Low GAMMA means high generalisation
# - High GAMMA is closer to original dataset
GAMMA = 6
KERNEL = RBF
svm_model.predict = lambda self, x: svm_predict([0], [x], self)[0][0]
# Get the data
SPARSE_LENGTH = 16
sparseCodings = sparse_coding.generateFull(SPARSE_LENGTH)
dataset, data, outputs, classes = sparse_coding.toSVMProblem(sparseCodings)
# Set the parameters for the SVM
parameters = svm_parameter()
parameters.kernel_type = KERNEL
parameters.C = COST
parameters.gamma = GAMMA
# Train the SVM
solver = svm_train(dataset, parameters)
# Create the output path if it doesn't exist
generated_dir = path.abspath(
path.join(
"generated",
"Q2Task1-TrainedSVM-{}".format(strftime("%Y-%m-%d_%H-%M-%S"))))
if not path.exists(generated_dir):
makedirs(generated_dir)
uniqueFileName = path.normpath(path.join(generated_dir, "data.pkl"))
svm_save_model(uniqueFileName, solver)
# Compare the results to the extected values
figure = plot.figure()
axis = figure.add_subplot(111)
colors = ['r', 'y', 'g', 'c', 'b', 'k']
for sample in sparseCodings:
classifier = sparse_coding.getClassifier(sample)
activationResult = svm_predict([0.], [sample], solver, '-q')[0][0]
axis.bar(classifier,
activationResult,
color=colors[classifier % len(colors)])
plot.savefig(path.normpath(path.join(generated_dir, "activations.png")))
plot.show()
def valid(self,datasets,opt,opp,method = fold,part_ids = None,seed = None,test_data = None):
if seed is None:
# If seed is not set. UNIX time is used as seed.
seed = time.time()
saving_seed = "%s/log/%s.log.seed" % (self._dir,self._name)
with open(saving_seed,"w") as fp:
# Save used seed value.
fp.write("seed:%f\n" % seed)
if part_ids is None:
part_ids = datasets.pids
groups = [(test,train) for test,train in method(part_ids,seed = seed)]
for cnt,pdtsts in enumerate(groups):
# cnt is number of cluster.
if test_data is None:
test = False
ltest,dtest,itest = test2svm_prob(datasets.mkTest(pdtsts[0]))
else:
test = True
ltest,dtest,itest = test2svm_prob(test_data.mkTest(test_data.pids))
print "start %s validation" % (cnt)
ptrn,itrain = train2svm_prob(datasets.mkTrain(pdtsts[1]))
#opt = svm.svm_parameter(opt)
model = svmutil.svm_train(ptrn,opt)
plbl,pacc,pval = svmutil.svm_predict(ltest,dtest,model,opp)
# create saving direcotry
#self._mkdir(cnt)
# create log files
self._save_log(itest,plbl,pval,cnt,test)
model_name = "%s/model/%s.model.%s" % (self._dir,self._name,cnt)
def plotSVM(solver, lowerBound, upperBound, step):
assert lowerBound < upperBound
X = arange(lowerBound, upperBound, step)
Y = arange(lowerBound, upperBound, step)
X,Y = meshgrid(X,Y)
Z = zeros(X.shape)
for i in range(len(X)):
for j in range(len(Y)):
#Classify the result
result = int( svm_predict([0.], [[ X[i][j], Y[i][j] ]], solver, '-q')[0][0] )
if result == 0:
Z[i][j] = 0 #lower limit
else:
Z[i][j] = 100 #higher limit
plot.imshow(Z)
plot.gcf()
plot.clim()
plot.title("SVM Activation")
return plot
def plotSVM(solver, lowerBound, upperBound, step):
assert lowerBound < upperBound
X = arange(lowerBound, upperBound, step)
Y = arange(lowerBound, upperBound, step)
X, Y = meshgrid(X, Y)
Z = zeros(X.shape)
for i in range(len(X)):
for j in range(len(Y)):
#Classify the result
result = int(
svm_predict([0.], [[X[i][j], Y[i][j]]], solver, '-q')[0][0])
if result == 0:
Z[i][j] = 0 #lower limit
else:
Z[i][j] = 100 #higher limit
plot.imshow(Z)
plot.gcf()
plot.clim()
plot.title("SVM Activation")
return plot
def classify_audio_file(audio_file,cfg, ext=None):
if isinstance(cfg,types.StringTypes):
cfg_path, _name=os.path.split(cfg)
cfg=eval(open(cfg).read())
if cfg_path:
if not cfg['range_file'].startswith(os.sep):
cfg['range_file']=os.path.join(cfg_path, cfg['range_file'])
if not cfg['model_file'].startswith(os.sep):
cfg['model_file']=os.path.join(cfg_path, cfg['model_file'])
classes=cfg.get('classes')
r=load_scaling_range(cfg['range_file'])
a,sr=audiolib.load_audio_sample(audio_file, ext)
logger.debug('Loaded audio sample of length %.2f' % (len(a)/float(sr)))
segs=audiolib.segment_audio(cfg['segalg'],a, sr, limit=cfg['threshold'],
**cfg['params']) [cfg['start_index']:cfg['end_index']]
scaled=[]
for s in segs:
features= audiolib.calc_mfcc(s, sr, nbins=cfg['nbins'])
scaled.append(scale_sample(features, r))
if model_cache.has_key(cfg['model_file']):
m=model_cache[cfg['model_file']]
else:
with module_lock:
m=svmutil.svm_load_model(cfg['model_file'])
model_cache[cfg['model_file']]=m
p_labels, p_acc, p_vals=svmutil.svm_predict([0]*len(scaled), scaled, m, '-q')
#logger.debug('prediction additional info ' +str(p_acc)+str(p_vals))
res=''.join([cls2char(int(n), classes) for n in p_labels])
return res
def classify_audio_file(audio_file, cfg, ext=None):
if isinstance(cfg, types.StringTypes):
cfg_path, _name = os.path.split(cfg)
cfg = eval(open(cfg).read())
if cfg_path:
if not cfg['range_file'].startswith(os.sep):
cfg['range_file'] = os.path.join(cfg_path, cfg['range_file'])
if not cfg['model_file'].startswith(os.sep):
cfg['model_file'] = os.path.join(cfg_path, cfg['model_file'])
classes = cfg.get('classes')
r = load_scaling_range(cfg['range_file'])
a, sr = audiolib.load_audio_sample(audio_file, ext)
logger.debug('Loaded audio sample of length %.2f' % (len(a) / float(sr)))
segs = audiolib.segment_audio(
cfg['segalg'], a, sr, limit=cfg['threshold'],
**cfg['params'])[cfg['start_index']:cfg['end_index']]
scaled = []
for s in segs:
features = audiolib.calc_mfcc(s, sr, nbins=cfg['nbins'])
scaled.append(scale_sample(features, r))
if model_cache.has_key(cfg['model_file']):
m = model_cache[cfg['model_file']]
else:
with module_lock:
m = svmutil.svm_load_model(cfg['model_file'])
model_cache[cfg['model_file']] = m
p_labels, p_acc, p_vals = svmutil.svm_predict([0] * len(scaled), scaled, m,
'-q')
#logger.debug('prediction additional info ' +str(p_acc)+str(p_vals))
res = ''.join([cls2char(int(n), classes) for n in p_labels])
return res
def valid(self,datasets,opt,opp,method = fold,seed = None): # Should groups and ngroups be idch ? groups = [(test,train) for test,train in method(datasets.pids,seed = seed)] ngroups = [(test,train) for test,train in method(datasets.nids,seed = seed)] for cnt,(pdtsts,ndtsts) in enumerate(zip(groups,ngroups)): # cnt is number of cluster. ltest,dtest,itest = test2svm_prob(datasets.mkTest(pdtsts[0],ndtsts[0])) ptrn,itrain = train2svm_prob(datasets.mkTrain(pdtsts[1],ndtsts[1])) print "start %s validation" % (cnt) #opt = svm.svm_parameter(opt) model = svmutil.svm_train(ptrn,opt) plbl,pacc,pval = svmutil.svm_predict(ltest,dtest,model,opp) # create saving direcotry #self._mkdir(cnt) # create log files self._save_log(itest,plbl,pval,cnt) model_name = "%s/model/%s.model.%s" % (self._dir,self._name,cnt)
import sys
import os
path = os.path.abspath(os.path.dirname(sys.argv[0]))
sys.path.append(path)
from svm.svm import *
from svm.svmutil import svm_train, svm_predict, svm_save_model
from support import csv
#from graphpy import SVM2D # importing this doesn't work for some reason
from time import strftime, time
from matplotlib import pyplot as plot
from numpy import arange, meshgrid, zeros
KERNEL = RBF # The kernel in the .dll file to use
svm_model.predict = lambda self, x: svm_predict([0], [x], self)[0][0]
def plotSVM(solver, lowerBound, upperBound, step):
assert lowerBound < upperBound
X = arange(lowerBound, upperBound, step)
Y = arange(lowerBound, upperBound, step)
X,Y = meshgrid(X,Y)
Z = zeros(X.shape)
for i in range(len(X)):
for j in range(len(Y)):
#Classify the result
result = int( svm_predict([0.], [[ X[i][j], Y[i][j] ]], solver, '-q')[0][0] )
import sys
import os
path = os.path.abspath(os.path.dirname(sys.argv[0]))
sys.path.append(path)
from svm.svm import *
from svm.svmutil import svm_train, svm_predict, svm_save_model
from support import csv
#from graphpy import SVM2D # importing this doesn't work for some reason
from time import strftime, time
from matplotlib import pyplot as plot
from numpy import arange, meshgrid, zeros
KERNEL = RBF # The kernel in the .dll file to use
svm_model.predict = lambda self, x: svm_predict([0], [x], self)[0][0]
def plotSVM(solver, lowerBound, upperBound, step):
assert lowerBound < upperBound
X = arange(lowerBound, upperBound, step)
Y = arange(lowerBound, upperBound, step)
X, Y = meshgrid(X, Y)
Z = zeros(X.shape)
for i in range(len(X)):
for j in range(len(Y)):
#Classify the result
