def __init__(self, matrix, mode="dtw", fast=False, radius=20, pu="CPU"):
'''
<calls others modules for the actual calculation, checks if CUDA
is present an organize computations>
Input Parameters:
matrix: a matrix containing time series, each row is a
timeseries and each column a point of the time series.
mode: method by which the distance between series is
calculated
"dtw" for normal DTW,
"ddtw" for the derivative DTW,
"euclidean" for euclidean distance,
"pearson", for pearson distance
fast: specifies whether to run the FastDTW or not, radius
parameter indicates the accuracy of the calculation,
the higher the value, the more accurate the result.
It applies only to the DTW algorithms.
FastDTWs are not implemented on GPU.
radius: see 'fast'. It is suggested to not use more than
100 for this value ignored if fast is flase.
Returns:
None
'''
self.pu = pu
self.matrix = matrix
self.mode = mode
self.fast = fast
self.radius = radius
self.derivative = False
self.euclidean = False
if self.mode == "ddtw":
self.derivative = True
elif self.mode == "euclidean":
self.euclidean = True
if self.pu == "GPU":
try:
from dtw_gpu import GpuDistance
except ImportError:
print "No suitable hardware! Doing DTW on CPU..."
self.pu = "CPU"
import dtw_cpu
self.dtw_cpu = dtw_cpu
else:
self.gpu = GpuDistance(matrix, self.mode, self.derivative)
else:
pu = "CPU"
import dtw_cpu
self.dtw_cpu = dtw_cpu
class Dist:
def __init__(self, matrix, mode="dtw", fast=False, radius=20, pu="CPU"):
'''
<calls others modules for the actual calculation, checks if CUDA
is present an organize computations>
Input Parameters:
matrix: a matrix containing time series, each row is a
timeseries and each column a point of the time series.
mode: method by which the distance between series is
calculated
"dtw" for normal DTW,
"ddtw" for the derivative DTW,
"euclidean" for euclidean distance,
"pearson", for pearson distance
fast: specifies whether to run the FastDTW or not, radius
parameter indicates the accuracy of the calculation,
the higher the value, the more accurate the result.
It applies only to the DTW algorithms.
FastDTWs are not implemented on GPU.
radius: see 'fast'. It is suggested to not use more than
100 for this value ignored if fast is flase.
Returns:
None
'''
self.pu = pu
self.matrix = matrix
self.mode = mode
self.fast = fast
self.radius = radius
self.derivative = False
self.euclidean = False
if self.mode == "ddtw":
self.derivative = True
elif self.mode == "euclidean":
self.euclidean = True
if self.pu == "GPU":
try:
from dtw_gpu import GpuDistance
except ImportError:
print "No suitable hardware! Doing DTW on CPU..."
self.pu = "CPU"
import dtw_cpu
self.dtw_cpu = dtw_cpu
else:
self.gpu = GpuDistance(matrix, self.mode, self.derivative)
else:
pu = "CPU"
import dtw_cpu
self.dtw_cpu = dtw_cpu
def compute(self, li):
'''
<Does the actual calculations.>
Input Parameters:
li: list of tuples containg indices of couples of time
series between which distace has to be calculated.
Indices refears to matrix passed at the init func.
Returns:
returns a numpy array containing distance values for each
couple passed. Result
are in the order of input.
'''
if self.pu == "GPU":
res = self.gpu.launch(li)
elif self.pu == "CPU":
res = numpy.array([])
for qui in li:
tmp = self.dtw_cpu.compute_dtw(self.matrix[qui[0]],
self.matrix[qui[1]],
self.euclidean,
False,
self.derivative,
self.fast,
self.radius)
res = numpy.append(res, numpy.array([tmp]))
return res
