@@ -21,19 +21,38 @@ cpdef np.ndarray[np.float64_t, ndim=1, mode='c'] normalized_histogram(np.ndarray
# This will be kept up to date to work with the script in the examples dir.
# It uses the feature below this.
cdef class VectorFeatureFactory(object ):
cdef object dims
"""
Args:
dims: Numpy array of min_val, max_val where [min_val, max_val) shape of (num_dims, 2)
types: Numpy array where 0: Real, 1: Integer, 2: Categorical shape of (num_dims). Real
generates continuous values and uses a <= feature, integer generates integral values
and uses a <= feature, and categorical generates integral values and uses a != feature.
num_thresh: Number of thresholds per feature
"""
cdef np.ndarray dims, types
cdef int num_thresh
def __init__ self , dims , num_thresh ):
self .dims = dims
self .num_thresh = num_thresh
def __init__ self , dims = None , types = None , num_thresh = None , label_values = None ):
self .dims = np.asarray(dims) if dims is not None else dims
self .num_thresh = num_thresh if num_thresh is not None else 0
self .types = np.asarray(types) if types is not None else types
if label_values:
values = [x[1 ] for x in label_values]
self .dims = np.dstack([np.min(values, 0 ), np.max(values, 0 )])[0 ]
def gen_feature (self ):
dim = random.randint(0 , len (self .dims) - 1 )
cdef int dim = random.randint(0 , len (self .dims) - 1 )
min_val, max_val = self .dims[dim]
threshs = np.array([np.random.uniform(min_val, max_val,
self .num_thresh)]).T
return VectorFeature(dim = dim, threshs = threshs)
cdef int feat_type = self .types[dim]
if feat_type == 0 :
threshs = np.random.uniform(min_val, max_val, self .num_thresh)
elif feat_type == 1 or feat_type == 2 :
threshs = np.random.randint(min_val, max_val, self .num_thresh)
else :
raise ValueError (' Feature type not recognized'
threshs = np.ascontiguousarray(threshs.reshape((threshs.size, 1 )))
return VectorFeature(dim = dim, threshs = threshs, feat_type = feat_type)
def loads (self , feat_ser ):
return VectorFeature(feat_ser = feat_ser)
@@ -59,34 +78,44 @@ cdef class VectorFeatureFactory(object):
cdef class VectorFeature(object ):
cdef object feat_ser
cdef int dim
cdef threshs
cdef np.ndarray threshs
cdef int feat_type
def __init__ self , feat_ser = None , dim = None , threshs = None ):
def __init__ self , feat_ser = None , dim = None , threshs = None , feat_type = None ):
self .feat_ser = feat_ser
if self .feat_ser:
self ._deserialize()
else :
self .dim = dim
self .threshs = threshs
self .feat_type = feat_type
def _deserialize (self ):
data = pickle.loads(self .feat_ser)
self .dim = data[' dim'
self .threshs = data[' threshs'
self .feat_type = data[' feat_type'
def __str__ self ):
if self .threshs.size == 1 :
return ' %s <= x[%d ]' % (self .threshs[0 ][0 ], self .dim)
return ' %s <= x[%d ]' % (self .threshs, self .dim)
t = self .threshs[0 ][0 ]
else :
t = self .threshs
if self .feat_type < 2 :
o = ' <='
else :
o = ' =='
return ' %s %s x[%d ]' % (t, o, self .dim)
def dumps (self ):
return pickle.dumps({' dim' self .dim, ' threshs' self .threshs}, - 1 )
return pickle.dumps({' dim' self .dim, ' threshs' self .threshs, ' feat_type ' : self .feat_type }, - 1 )
def __repr__ self ):
return ' VectorFeature(dim=%r , threshs=%r )' % (self .dim, self .threshs)
return ' VectorFeature(dim=%r , threshs=%r , feat_type= %f )' % (self .dim, self .threshs, self .feat_type )
def __getitem__ self , index ):
return VectorFeature(dim = self .dim,
feat_type = self .feat_type,
threshs = np.array([[self .threshs.flat[int (index)]]]))
def __call__ self , values ):
@@ -99,9 +128,13 @@ cdef class VectorFeature(object):
"""
values = np.asarray(values)
if values.ndim == 1 :
return values[self .dim] >= self .threshs
v = values[self .dim]
else :
v = values[:, self .dim]
if self .feat_type < 2 :
return v >= self .threshs
else :
return values[:, self .dim] > =self .threshs
return v = =self .threshs
def label_histograms (self , labels , values , int num_classes ):
"""