def testFacepointsWithSeq():
print("hi")
import numpy
import sys
sys.path.append("./utils")
from read_write import read_npy
import configparser
cf = configparser.ConfigParser()
cf.read("./config.ini")
from distanceBetweenFacePoints import facePoints68_DistBetweenPoints
from distanceBetweenFacePoints import facePoints68_DistToNNextPoints
from distanceBetweenSeq import distanceBetweenSeq
from DistanceFunction import DistanceSeqFunction
dataSet = read_npy(cf.get("datasets", "npyAllDataTest"))
# seqA = dataSet[0][1]
# seqB = dataSet[2][0]
distFunc = DistanceFunction(facePoints68_DistToNNextPoints,
"normalizedL1")
distSeqFunc = DistanceSeqFunction(distFunc, "normalizedL1",
distanceBetweenSeq)
#print(seqB[1][0][2][1])
print("hi")
pred = predictions(dataSet[0], dataSet[1], distSeqFunc)
print(len([i for i in pred if i[0] == i[1]]))
print("accuracy: ", len([i for i in pred if i[0] == i[1]]) / len(pred))
def _supercounts(graph, template_event_set, subcount_table):
''' Given
graph - a reconciliation graph
template_event_set - a reconciliation in event-set form
subcount_table - a mapping from nodes in the graph to subcounts
Computes the stratified symmetric supercounts of every node in the graph
with respect to the template. Relies on pre-computed subcounts
'''
table = {}
for n in graph.preorder():
if n.isRoot():
table[n] = DistanceFunction.kronicker(0)
elif n.isMap():
def process_parent(event_parent):
shift_amount = 1 if event_parent not in template_event_set else -1
parent_super = table[event_parent]
otherChild = event_parent.otherChild(n)
convolved = parent_super.convolve(subcount_table[otherChild]) \
if otherChild \
else parent_super
return convolved.shift(shift_amount)
sup_count = reduce(lambda x, y: x.sum(y), map(process_parent, n.parents))
table[n] = sup_count
else:
shift_amount = 1 if n not in template_event_set else -1
# Event nodes inherit their single parent's super-count.
# Shallow copy is same because we don't mutatate Fn's
table[n] = table[n.parents[0]]
return table
def _supercounts(graph, template_event_set, subcount_table):
''' Given
graph - a reconciliation graph
template_event_set - a reconciliation in event-set form
subcount_table - a mapping from nodes in the graph to subcounts
Computes the stratified symmetric supercounts of every node in the graph
with respect to the template. Relies on pre-computed subcounts
'''
table = {}
for n in graph.preorder():
if n.isRoot():
table[n] = DistanceFunction.kronicker(0)
elif n.isMap():
def process_parent(event_parent):
shift_amount = 1 if event_parent not in template_event_set else -1
parent_super = table[event_parent]
otherChild = event_parent.otherChild(n)
convolved = parent_super.convolve(subcount_table[otherChild]) \
if otherChild \
else parent_super
return convolved.shift(shift_amount)
sup_count = reduce(lambda x, y: x.sum(y),
map(process_parent, n.parents))
table[n] = sup_count
else:
shift_amount = 1 if n not in template_event_set else -1
# Event nodes inherit their single parent's super-count.
# Shallow copy is same because we don't mutatate Fn's
table[n] = table[n.parents[0]]
return table
def predictions(trData, testData, distFunc: DistanceFunction):
'''
args:
useSeq: whether we treat the sequence as a single datapoint or not
'''
closestIndices = []
for testDp in testData:
print("new test dp")
closestIndex = -1
currDistance = float("inf")
for i, trDp in enumerate(trData):
tempDistance = distFunc.calculateDistance(testDp, trDp)
if (tempDistance < currDistance) and (trData[i][1] != testDp[1]):
closestIndex = i
currDistance = tempDistance
#print([testDp[0],trData[closestIndex][0]])
closestIndices.append([testDp[0],
trData[closestIndex][0]]) #get the name
return closestIndices
def _subcounts(graph, template_event_set):
''' Given
graph - a reconciliation graph
template_event_set - a reconciliation in event-set form
Computes the stratified symmetric subcounts of every node in the graph
with respect to the template
'''
table = {}
for n in graph.postorder():
if n.isLeaf():
table[n] = DistanceFunction.kronicker(-1)
elif n.isMap():
table[n] = reduce(lambda x, y: x.sum(y), \
[table[c] for c in n.children])
else:
shift_amount = 1 if n not in template_event_set else -1
table[n] = reduce(lambda x, y: x.convolve(y), \
[table[c] for c in n.children])
table[n] = table[n].shift(shift_amount)
return table
def _subcounts(graph, template_event_set):
''' Given
graph - a reconciliation graph
template_event_set - a reconciliation in event-set form
Computes the stratified symmetric subcounts of every node in the graph
with respect to the template
'''
table = {}
for n in graph.postorder():
if n.isLeaf():
table[n] = DistanceFunction.kronicker(-1)
elif n.isMap():
table[n] = reduce(lambda x, y: x.sum(y), \
[table[c] for c in n.children])
else:
shift_amount = 1 if n not in template_event_set else -1
table[n] = reduce(lambda x, y: x.convolve(y), \
[table[c] for c in n.children])
table[n] = table[n].shift(shift_amount)
return table
