def test_calculate_probability(self):
k = 3
model = kNN.train(xs, ys, k)
weights = kNN.calculate(model, [6, -173.143442352])
self.assertAlmostEqual(weights[0], 0.0, places=6)
self.assertAlmostEqual(weights[1], 3.0, places=6)
weights = kNN.calculate(model, [309, -271.005880394])
self.assertAlmostEqual(weights[0], 3.0, places=6)
self.assertAlmostEqual(weights[1], 0.0, places=6)
weights = kNN.calculate(model, [117, -267.13999999999999])
self.assertAlmostEqual(weights[0], 2.0, places=6)
self.assertAlmostEqual(weights[1], 1.0, places=6)
def test_calculate_probability(self):
k = 3
model = kNN.train(xs, ys, k)
weights = kNN.calculate(model, [6, -173.143442352])
self.assertAlmostEqual(weights[0], 0.0, 6)
self.assertAlmostEqual(weights[1], 3.0, 6)
weights = kNN.calculate(model, [309, -271.005880394])
self.assertAlmostEqual(weights[0], 3.0, 6)
self.assertAlmostEqual(weights[1], 0.0, 6)
weights = kNN.calculate(model, [117, -267.13999999999999])
self.assertAlmostEqual(weights[0], 2.0, 6)
self.assertAlmostEqual(weights[1], 1.0, 6)
def classify(self, sound_fft, som, knn_model):
"""
Classify a new sound. Takes in a sound FFT, a SOM model and kNN model.
Returns a tuple containing a highest probability label and a dictionary of
all label probabilities.
"""
sequence = self.sound_fft_to_string(sound_fft, som)
weights = kNN.calculate(knn_model,
self.stringify_sequence(sequence),
weight_fn=self.knn_weight_fn,
distance_fn=self.sound_seq_distance_str)
sum_weights = float(sum(weights.values()))
most_class = None
most_weight = None
for klass, weight in weights.items():
weights[klass] = weight / sum_weights
if most_class is None or weight > most_weight:
most_class = klass
most_weight = weight
# make sure there are no 0's
for k, v in weights.items():
if v == 0: weights[k] = 1.0e-2
# renormalize
tot = sum(weights.values())
for k, v in weights.items():
weights[k] = v / float(tot)
return most_class, weights
def classify(self, sound_fft, som, knn_model):
"""
Classify a new sound. Takes in a sound FFT, a SOM model and kNN model.
Returns a tuple containing a highest probability label and a dictionary of
all label probabilities.
"""
sequence = self.sound_fft_to_string(sound_fft, som)
weights = kNN.calculate(knn_model,
self.stringify_sequence(sequence),
weight_fn=self.knn_weight_fn,
distance_fn=self.sound_seq_distance_str)
sum_weights = float(sum(weights.values()))
most_class = None
most_weight = None
for klass, weight in weights.items():
weights[klass] = weight / sum_weights
if most_class is None or weight > most_weight:
most_class = klass
most_weight = weight
# make sure there are no 0's
for k,v in weights.items():
if v == 0: weights[k] = 1.0e-2
# renormalize
tot = sum(weights.values())
for k,v in weights.items():
weights[k] = v / float(tot)
return most_class, weights
def classify(self, sound_fft):
"""
Classify a new sound. Takes in a sound FFT, a SOM model and kNN model.
Returns a tuple containing a highest probability label and a dictionary of
all label probabilities.
"""
sequence = []
for col in range(sound_fft.shape[1]):
sequence.append(self.som.bmu(sound_fft[:,col]))
weights = kNN.calculate(self.knn_model,
self.stringify_sequence(sequence),
weight_fn=self.knn_weight_fn,
distance_fn=self.sound_seq_distance_str)
sum_weights = float(sum(weights.values()))
most_class = None
most_weight = None
for klass, weight in weights.items():
weights[klass] = weight / sum_weights
if most_class is None or weight > most_weight:
most_class = klass
most_weight = weight
return most_class, weights
def get(self):
offset = int(self.get_argument('o', default='1'))
rowcount = int(self.get_argument('r', default='10'))
offset = (offset - 1) * rowcount
no = self.get_argument('no', default='')
model_id = self.get_argument('model_id', default='')
model_type = self.get_argument('model_type', default='')
package = self.get_argument('model_name', default='')
cur = self.db.getCursor()
rowdata = {}
#查询
if no == '1':
if model_type == '1':
cur.execute(
" select b.name,a.create_id,a.name,a.note,a.beta from public.logistis a "
" left join public.account b on a.create_id = b.id "
"where a.id='%s' " % (model_id))
rows = cur.fetchall()
print(rows)
rowdata['struct'] = "id,create_id,name,note,beta "
rowdata['rows'] = rows
else:
cur.execute(
" select b.name,a.create_id,a.name,a.note,c.name,a.file_name from public.pymodel a "
" left join public.account b on a.create_id = b.id "
" left join public.model c on a.type = c.type "
" where a.id='%s' and a.type='%s' " %
(model_id, model_type))
rows = cur.fetchall()
rowdata['struct'] = "id,create_id,name,note,type,filename "
rowdata['rows'] = rows
self.response(rowdata)
elif no == '2':
if model_type == '1':
beta = self.get_argument('beta', default='')
model_data = self.get_argument('model', default='')
a = []
q = 0
print(model_data)
a = (list(eval(model_data)))
model = LogisticRegression.LogisticRegression()
model.beta = (list(eval(beta)))
rowdata = {}
rowdata['op'] = LogisticRegression.calculate(model, a)
rowdata['rows'] = LogisticRegression.classify(model, a)
elif model_type == '2':
pack = 'data_mining.' + package
import importlib
bb = importlib.import_module(pack)
ma = kNN.kNN()
model = bb.model.knn(ma)
model_data = self.get_argument('model', default='')
a = []
a = (list(eval(model_data)))
rowdata = {}
rowdata['op'] = kNN.calculate(model, a)
rowdata['rows'] = kNN.classify(model, a)
elif model_type == '3':
pack = 'data_mining.' + package
import importlib
bb = importlib.import_module(pack)
ma = NaiveBayes.NaiveBayes()
model = bb.model.bayes(ma)
model_data = self.get_argument('model', default='')
a = []
a = (list(eval(model_data)))
rowdata = {}
rowdata['op'] = NaiveBayes.calculate(model, a)
rowdata['rows'] = NaiveBayes.classify(model, a)
self.response(rowdata)
def get(self):
offset = int(self.get_argument('o',default='1'))
rowcount = int(self.get_argument('r',default='10'))
offset=(offset-1)*rowcount
no = self.get_argument('no', default='')
model_id = self.get_argument('model_id', default='')
model_type = self.get_argument('model_type', default='')
package=self.get_argument('model_name', default='')
cur=self.db.getCursor()
rowdata={}
#查询
if no=='1':
if model_type =='1':
cur.execute(" select b.name,a.create_id,a.name,a.note,a.beta from public.logistis a "
" left join public.account b on a.create_id = b.id "
"where a.id='%s' "% (model_id) )
rows = cur.fetchall()
print(rows)
rowdata['struct']="id,create_id,name,note,beta "
rowdata['rows']= rows
else:
cur.execute(" select b.name,a.create_id,a.name,a.note,c.name,a.file_name from public.pymodel a "
" left join public.account b on a.create_id = b.id "
" left join public.model c on a.type = c.type "
" where a.id='%s' and a.type='%s' "% (model_id,model_type) )
rows = cur.fetchall()
rowdata['struct']="id,create_id,name,note,type,filename "
rowdata['rows']= rows
self.response(rowdata)
elif no=='2':
if model_type=='1':
beta = self.get_argument('beta', default='')
model_data=self.get_argument('model', default='')
a=[]
q=0
print(model_data)
a=(list(eval(model_data)))
model=LogisticRegression.LogisticRegression()
model.beta=(list(eval(beta)))
rowdata={}
rowdata['op']=LogisticRegression.calculate(model,a)
rowdata['rows']=LogisticRegression.classify(model,a)
elif model_type=='2':
pack='data_mining.'+package
import importlib
bb=importlib.import_module(pack)
ma=kNN.kNN()
model=bb.model.knn(ma)
model_data=self.get_argument('model', default='')
a=[]
a=(list(eval(model_data)))
rowdata={}
rowdata['op']=kNN.calculate(model,a)
rowdata['rows']=kNN.classify(model,a)
elif model_type=='3':
pack='data_mining.'+package
import importlib
bb=importlib.import_module(pack)
ma=NaiveBayes.NaiveBayes()
model=bb.model.bayes(ma)
model_data=self.get_argument('model', default='')
a=[]
a=(list(eval(model_data)))
rowdata={}
rowdata['op']=NaiveBayes.calculate(model,a)
rowdata['rows']=NaiveBayes.classify(model,a)
self.response(rowdata)
