def FL_sparse(data): #Case1 user only provides data and specifies
obj = FacilityLocationFunction(n=43,
data=data,
mode="sparse",
metric="euclidean",
num_neigh=10)
obj.maximize(10, 'NaiveGreedy', False, False, False)
def FL_clustered_case1(data): #Case1 user only provides data
obj = FacilityLocationFunction(n=43,
data=data,
mode="clustered",
metric="euclidean",
num_cluster=10)
obj.maximize(10, 'NaiveGreedy', False, False, False)
def test_4_3(self): # X not a subset of ground set for evaluate()
M = np.array([[1, 2], [3, 4]])
obj = FacilityLocationFunction(n=2, sijs=M)
X = {0, 2}
try:
obj.evaluate(X)
except Exception as e:
assert str(e) == "ERROR: X is not a subset of ground set"
def test_4_4(self): # X not a subset of ground set for marginalGain()
M = np.array([[1, 2], [3, 4]])
obj = FacilityLocationFunction(n=2, sijs=M)
X = {0, 2}
try:
obj.marginalGain(X, 1)
except Exception as e:
assert str(e) == "ERROR: X is not a subset of ground set"
def FL_clustered_case2(
data, lab): #Case2 user also provides cluster info along with data
obj = FacilityLocationFunction(n=43,
data=data,
cluster_lab=lab,
mode="clustered",
metric="euclidean",
num_cluster=10)
obj.maximize(10, 'NaiveGreedy', False, False, False)
def fl_dense_cpp_kernel():
obj = FacilityLocationFunction(n=num_samples,
mode="dense",
data=dataArray,
metric="euclidean")
obj.maximize(budget=budget,
optimizer=optimizer,
stopIfZeroGain=False,
stopIfNegativeGain=False,
verbose=False)
def fl_dense_py_kernel():
K_dense = create_kernel(dataArray, mode='dense', metric='euclidean')
obj = FacilityLocationFunction(n=num_samples,
mode="dense",
sijs=K_dense,
separate_rep=False)
obj.maximize(budget=budget,
optimizer=optimizer,
stopIfZeroGain=False,
stopIfNegativeGain=False,
verbose=False)
def fl_dense_py_kernel_np_numba_array64():
K_dense = helper.create_kernel_dense_np_numba(dataArray, 'euclidean')
obj = FacilityLocationFunction(n=num_samples,
mode="dense",
sijs=K_dense,
separate_rep=False,
pybind_mode="array64")
obj.maximize(budget=budget,
optimizer=optimizer,
stopIfZeroGain=False,
stopIfNegativeGain=False,
verbose=False)
def f_3(): #more realistic test case
M = np.array([[-0.78569, 0.75, 0.9, -0.56, 0.005],
[0.00006, 0.400906, -0.203, 0.9765, -0.9],
[0.1, 0.3, 0.5, 0.0023, 0.9],
[-0.1, 0.1, 0.1405, 0.0023, 0.3],
[-0.123456, 0.0789, 0.00456, 0.001, -0.9]])
obj = FacilityLocationFunction(n=5, sijs=M)
return obj
def test_4_2(self): #Inconsistency between n and no of examples in M
M = np.array([[1, 2, 3], [4, 5, 6]])
try:
FacilityLocationFunction(n=1, sijs=M)
except Exception as e:
assert str(
e
) == "ERROR: Inconsistentcy between n and no of examples in the given similarity matrix"
def test_4_1(self): #Non-square dense similarity matrix
M = np.array([[1, 2, 3], [4, 5, 6]])
try:
FacilityLocationFunction(n=2, sijs=M)
except Exception as e:
assert str(
e
) == "ERROR: Dense similarity matrix should be a square matrix if ground and master datasets are same"
def f_5():
data = np.array([[100, 21, 365, 5], [57, 18, -5, -6], [16, 255, 68, -8],
[2, 20, 6, 2000], [12, 20, 68, 200]])
obj = FacilityLocationFunction(n=5,
data=data,
mode="dense",
metric="cosine")
return obj
def test_4_6(self): # n==0
data = np.array([[1, 2], [3, 4]])
num_neigh, M = create_kernel(data, 'sparse', 'euclidean', num_neigh=1)
try:
FacilityLocationFunction(n=0, sijs=M, num_neigh=num_neigh)
except Exception as e:
assert str(
e) == "ERROR: Number of elements in ground set can't be 0"
def test():
num_clusters = 3 #10 #100 #3
cluster_std_dev = 1 #4 #4 #1
num_samples = 9 #500 #5000 #9
budget = 4 #10 #10 #4
points, cluster_ids, centers = make_blobs(n_samples=num_samples,
centers=num_clusters,
n_features=2,
cluster_std=cluster_std_dev,
center_box=(0, 100),
return_centers=True,
random_state=4)
data = list(map(tuple, points))
xs = [x[0] for x in data]
ys = [x[1] for x in data]
#plt.scatter(xs, ys, s=25, color='black', label="Images")
#plt.show()
dataArray = np.array(data)
from submodlib.functions.facilityLocation import FacilityLocationFunction
obj = FacilityLocationFunction(n=num_samples,
mode="dense",
data=dataArray,
metric="euclidean")
print("Testing FacilityLocation's maximize")
# from submodlib.functions.disparitySum import DisparitySumFunction
# obj = DisparitySumFunction(n=num_samples, mode="dense", data=dataArray, metric="euclidean")
# print("Testing DisparitySum's maximize")
#greedyList = obj.maximize(budget=budget, optimizer='NaiveGreedy', stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
greedyList = obj.maximize(budget=budget,
optimizer='LazyGreedy',
stopIfZeroGain=False,
stopIfNegativeGain=False,
verbose=False)
#greedyList = obj.maximize(budget=budget, optimizer='StochasticGreedy', stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
#greedyList = obj.maximize(budget=budget, optimizer='LazierThanLazyGreedy', stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
print(f"Greedy vector: {greedyList}")
greedyXs = [xs[x[0]] for x in greedyList]
greedyYs = [ys[x[0]] for x in greedyList]
def f_7():
data = np.array([[100, 21, 365, 5], [57, 18, -5, -6], [16, 255, 68, -8],
[2, 20, 6, 2000], [12, 20, 68, 200]])
num_cluster = 2
obj = FacilityLocationFunction(n=5,
data=data,
mode="clustered",
metric="euclidean",
num_cluster=num_cluster)
return obj
def test_4_5(
self
): # If sparse matrix is provided but without providing number of neighbors that were used to create it
data = np.array([[1, 2], [3, 4]])
num_neigh, M = create_kernel(data, 'sparse', 'euclidean', num_neigh=1)
try:
FacilityLocationFunction(
n=2, sijs=M
) #its important for user to pass num_neigh with sparse matrix because otherwise
#there is no way for Python FL and C++ FL to know how many nearest neighours were
#reatined in sparse matrix
except Exception as e:
assert str(
e) == "ERROR: num_neigh for given sparse matrix not provided"
def create_fl_dense_py_kernel(num_samples, pyDenseKernel):
return FacilityLocationFunction(n=num_samples,
mode="dense",
sijs=pyDenseKernel,
separate_rep=False)
def fl_dense_py_kernel_other_array():
K_dense = helper.create_kernel(dataArray, mode="dense", metric='euclidean', method="other")
obj = FacilityLocationFunction(n=num_samples, mode="dense", sijs=K_dense, separate_rep=False,pybind_mode="array")
obj.maximize(budget=budget,optimizer=optimizer, stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False, show_progress=False)
def create_fl_dense_cpp_kernel(num_samples, dataArray):
return FacilityLocationFunction(n=num_samples,
mode="dense",
data=dataArray,
metric="euclidean")
def test():
num_clusters = 3
cluster_std_dev = 1
num_samples = 8
num_set = 3
budget = 4
points, cluster_ids, centers = make_blobs(n_samples=num_samples, centers=num_clusters, n_features=2, cluster_std=cluster_std_dev, center_box=(0,100), return_centers=True, random_state=4)
data = list(map(tuple, points))
xs = [x[0] for x in data]
ys = [x[1] for x in data]
# get num_set data points belonging to cluster#1
random.seed(1)
cluster1Indices = [index for index, val in enumerate(cluster_ids) if val == 1]
subset1 = random.sample(cluster1Indices, num_set)
subset1xs = [xs[x] for x in subset1]
subset1ys = [ys[x] for x in subset1]
# plt.scatter(xs, ys, s=25, color='black', label="Images")
# plt.scatter(subset1xs, subset1ys, s=25, color='red', label="Subset1")
# plt.show()
set1 = set(subset1[:-1])
# get num_set data points belonging to different clusters
subset2 = []
for i in range(num_set):
#find the index of first point that belongs to cluster i
diverse_index = cluster_ids.tolist().index(i)
subset2.append(diverse_index)
subset2xs = [xs[x] for x in subset2]
subset2ys = [ys[x] for x in subset2]
# plt.scatter(xs, ys, s=25, color='black', label="Images")
# plt.scatter(subset2xs, subset2ys, s=25, color='red', label="Subset2")
# plt.show()
set2 = set(subset2[:-1])
dataArray = np.array(data)
from submodlib.functions.facilityLocation import FacilityLocationFunction
# start = time.process_time()
obj5 = FacilityLocationFunction(n=num_samples, data=dataArray, mode="clustered", metric="euclidean", num_clusters=num_clusters)
# print(f"Time taken by instantiation = {time.process_time() - start}")
print(f"Subset 1's FL value = {obj5.evaluate(set1)}")
print(f"Subset 2's FL value = {obj5.evaluate(set2)}")
print(f"Gain of adding another point ({subset1[-1]}) of same cluster to {set1} = {obj5.marginalGain(set1, subset1[-1])}")
print(f"Gain of adding another point ({subset2[-1]}) of different cluster to {set1} = {obj5.marginalGain(set1, subset2[-1])}")
obj5.setMemoization(set1)
print(f"Subset 1's Fast FL value = {obj5.evaluateWithMemoization(set1)}")
print(f"Fast gain of adding another point ({subset1[-1]}) of same cluster to {set1} = {obj5.marginalGainWithMemoization(set1, subset1[-1])}")
# start = time.process_time()
greedyList = obj5.maximize(budget=budget,optimizer='NaiveGreedy', stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
print(f"Greedy vector: {greedyList}")
# print(f"Time taken by maximization = {time.process_time() - start}")
# greedyXs = [xs[x[0]] for x in greedyList]
# greedyYs = [ys[x[0]] for x in greedyList]
# plt.scatter(xs, ys, s=25, color='black', label="Images")
# plt.scatter(greedyXs, greedyYs, s=25, color='blue', label="Greedy Set")
from submodlib import ClusteredFunction
obj7 = ClusteredFunction(n=num_samples, mode="multi", f_name='FacilityLocation', metric='euclidean', data=dataArray, num_clusters=num_clusters)
# print(f"Time taken by instantiation = {time.process_time() - start}")
print(f"Subset 1's FL value = {obj7.evaluate(set1)}")
print(f"Subset 2's FL value = {obj7.evaluate(set2)}")
print(f"Gain of adding another point ({subset1[-1]}) of same cluster to {set1} = {obj7.marginalGain(set1, subset1[-1])}")
print(f"Gain of adding another point ({subset2[-1]}) of different cluster to {set1} = {obj7.marginalGain(set1, subset2[-1])}")
obj7.setMemoization(set1)
print(f"Subset 1's Fast FL value = {obj7.evaluateWithMemoization(set1)}")
print(f"Fast gain of adding another point ({subset1[-1]}) of same cluster to {set1} = {obj7.marginalGainWithMemoization(set1, subset1[-1])}")
# start = time.process_time()
greedyList = obj7.maximize(budget=budget,optimizer='NaiveGreedy', stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
print(f"Greedy vector: {greedyList}")
import pytest
import numpy as np
from scipy import sparse
import scipy
from submodlib.functions.facilityLocation import FacilityLocationFunction
from submodlib.helper import create_kernel
from submodlib_cpp import FacilityLocation
data=np.array([
[100, 21, 365, 5],
[57, 18, -5, -6],
[16, 255, 68, -8],
[2,20,6, 2000],
[12,20,68, 200]
])
s = {1}
obj = FacilityLocationFunction(n=5, data=data, mode="sparse", metric="cosine")
print(obj.maximize(3,'NaiveGreedy', False, False, False))
def create_fl_mode_user():
return FacilityLocationFunction(n=num_samples, mode="clustered", data=dataArray, metric="euclidean", num_clusters=num_clusters, cluster_labels=cluster_ids.tolist())
def create_fl_mode_birch():
return FacilityLocationFunction(n=num_samples, mode="clustered", data=dataArray, metric="euclidean", num_clusters=num_clusters)
#A dryrun of implemented code with dummy data
import numpy as np
from submodlib.functions.facilityLocation import FacilityLocationFunction
from submodlib.helper import create_kernel
data = np.array([[1, 2, 3], [3, 4, 5], [4, 5, 6]])
#dryrun of create_kernel
n_, K_dense = create_kernel(data, 'dense', 'euclidean')
print(K_dense)
n_, K_sparse = create_kernel(data, 'sparse', 'euclidean', num_neigh=2)
print(K_sparse)
#dryrun of C++ FL and Python FL when user provides similarity matrix
#1) with dense matrix
obj = FacilityLocationFunction(n=3, sijs=K_dense)
X = {1}
print(obj.evaluate(X))
X = {1, 2}
print(obj.evaluate(X))
X = {1}
print(obj.marginalGain(X, 2))
#2) with sparse matrix
obj = FacilityLocationFunction(n=3, sijs=K_sparse, num_neigh=2)
#dryrun of C++ FL and Python FL when user provides data
#1) with dense mode
obj = FacilityLocationFunction(n=3,
data=data,
mode="dense",
def f_1(): # A simple easy to calculate test case
M = np.array([[1, 3, 2], [5, 4, 3], [4, 7, 5]])
obj = FacilityLocationFunction(n=3, sijs=M)
return obj
def create_fl_sparse_cpp_kernel(num_samples, dataArray, num_neighbors):
return FacilityLocationFunction(n=num_samples,
mode="sparse",
data=dataArray,
metric="euclidean",
num_neighbors=num_neighbors)
def f_2(): #Boundary case of just one element
M = np.array([-0.78569])
obj = FacilityLocationFunction(n=1, sijs=M)
return obj
def FL_case2(M): #Case2 user directly provides kernel
obj = FacilityLocationFunction(n=43, sijs=M, num_neigh=10)
obj.maximize(10, 'NaiveGreedy', False, False, False)
def create_fl_sparse_py_kernel(num_samples, pySparseKernel, num_neighbors):
return FacilityLocationFunction(n=num_samples,
mode="sparse",
sijs=pySparseKernel,
num_neighbors=num_neighbors)
def fl_mode_user():
obj = FacilityLocationFunction(n=num_samples, mode="clustered", data=dataArray, metric="euclidean", num_clusters=num_clusters, cluster_labels=cluster_ids.tolist())
obj.maximize(budget=budget,optimizer=optimizer, stopIfZeroGain=False, stopIfNegativeGain=False, verbose=False)
