def fcnn(x_train, y_train,k,r,w,v):
#k, r, w, v are dummies to integrate all reduction algs
#Vectorized implementation for a more interpretable implementation see below
n_classes = int(y_train.max() + 1)
n_samples = x_train.shape[0]
T = np.concatenate([x_train, np.reshape(y_train, [n_samples, 1])], axis=1)
ind_delta_S = np.array([centroid(x_train[y_train == i]) for i in range(n_classes)])
X = T[:, :-1]
Y = T[:, -1].astype(int)
ind_T = np.array([i for i in range(n_samples)])
ind_nearest = np.ones([n_classes + n_samples], dtype=int)
min_distances = np.inf * np.ones([n_classes + n_samples])
ind_S = np.array([])
while ind_delta_S.shape[0] != 0:
ind_S = np.array(list(set(ind_S.tolist()).union(set(ind_delta_S.tolist()))))
ind_Q = np.array(list(set(ind_T.tolist()) - set(ind_S.tolist())))
distances = np.min(dist(X[ind_Q], X[ind_delta_S]), axis=1)
ind_distances = np.array([ind_delta_S[i] for i in np.argmin(dist(X[ind_Q], X[ind_delta_S]), axis=1)]).astype(
int)
ind_change_nearest = min_distances[ind_Q] > distances
min_distances[ind_Q[ind_change_nearest]] = distances[ind_change_nearest]
ind_nearest[ind_Q[ind_change_nearest]] = ind_distances[ind_change_nearest]
# We do the same for representatives
ind_missmatch = ind_Q[np.logical_not(np.equal(Y[ind_Q], Y[ind_nearest[ind_Q]]))] # Different labels
ind_rep = np.array([], dtype=int)
def_rep = np.unique(ind_nearest[ind_Q])
delete=[]
for m,n in enumerate(def_rep):
n_indexes = ind_Q[np.where(ind_nearest[ind_Q] == n)[0]]
n_indexes = np.array(list(set(n_indexes.tolist()).intersection(set(ind_missmatch.tolist()))))
if n_indexes.size == 0:
delete.append(m)
continue
rep_distances = np.linalg.norm(X[ind_nearest[n_indexes]] - X[n_indexes], axis=1)
ind_rep = np.concatenate([ind_rep, [n_indexes[np.argmin(rep_distances)]]])
def_rep=np.delete(def_rep,delete)
ind_delta_S = ind_rep[np.in1d(def_rep, ind_S)]
# Return final subset
x_train = np.stack(X[ind_S])
y_train = np.array(Y[ind_S])
return x_train, y_train
def K_medoids_min_cost(x, k, iteration):
a = []
b = []
c = []
for i in range(iteration):
medoids = int_medoids(x, k)
d = dist(x, medoids)
a.append(d)
target_class = d.argmin(axis=1)
b.append(target_class)
iteration_cost = sum(d.min(axis=1))
c.append(iteration_cost)
return a, b, c
def incremental_farthest_search(points, k):
remaining_points = points[:]
solution_set = []
solution_set.append(
remaining_points.pop(random.randint(0,
len(remaining_points) - 1)))
for _ in range(k - 1):
distances = [dist(p, solution_set[0]) for p in remaining_points]
for i, p in enumerate(remaining_points):
for j, s in enumerate(solution_set):
distances[i] = min(distances[i], distance(p, s))
solution_set.append(
remaining_points.pop(distances.index(max(distances))))
return solution_set
def Predict_Ngram(Inpath="../test/test.txt",
Outpath="../Output/8_1.csv",
Train_Trump="../train/trump.txt",
Train_Obama="../train/obama.txt"):
'''
Input:
Inpath : file path for test data
Outpath : file path for output csv file
Train_Trump : file path to train Trump's bigram model
Train_Obama : file path to train Obama's bigram model
Output:
Return None, Output should go straight to .csv file
'''
f = open(Outpath, 'w')
f.write('Id,Prediction\n')
#Preprocess the test set
Paragraphs_Trump = NGram.corpora_preprocess(Train_Trump)
P_Vecs_Trump = [Get_Vector(p) for p in Paragraphs_Trump]
Paragraphs_Obama = NGram.corpora_preprocess(Train_Obama)
P_Vecs_Obama = [Get_Vector(p) for p in Paragraphs_Obama]
Paragraphs_test = NGram.corpora_preprocess(Inpath)
P_Vecs_test = [Get_Vector(p) for p in Paragraphs_test]
for idx, pvec in enumerate(P_Vecs_test):
max_cosine = -50
isTrump = True
for pv_trump in P_Vecs_Trump:
max_cosine = max(max_cosine, dist(pvec, pv_trump))
for pv_obama in P_Vecs_Obama:
if dist(pvec, pv_obama) > max_cosine:
isTrump = False
break
f.write(str(idx) + ',')
if isTrump:
f.write('1')
else:
f.write('0')
f.write('\n')
def find_d_embedding(data: list, maxm: int) -> int:
RT = 15.0
AT = 2
sigmay = np.std(data, ddof=1)
nyr = len(data)
m = maxm
EM = lagmat(data, maxlag=m - 1)
EEM = np.asarray([EM[j, :] for j in range(m - 1, EM.shape[0])])
embedm = maxm
for k in range(AT, EEM.shape[1] + 1):
fnn1 = []
fnn2 = []
Ma = EEM[:, range(k)]
D = dist(Ma)
for i in range(1, EEM.shape[0] - m - k):
d = D[i, :]
pdnz = np.where(d > 0)
dnz = d[pdnz]
Rm = np.min(dnz)
l = np.where(d == Rm)
l = l[0]
l = l[len(l) - 1]
if l + m + k - 1 < nyr:
fnn1.append(
np.abs(data[i + m + k - 1] - data[l + m + k - 1]) / Rm)
fnn2.append(
np.abs(data[i + m + k - 1] - data[l + m + k - 1]) / sigmay)
Ind1 = np.where(np.asarray(fnn1) > RT)
Ind2 = np.where(np.asarray(fnn2) > AT)
if len(Ind1[0]) / float(len(fnn1)) < 0.1 and len(Ind2[0]) / float(
len(fnn2)) < 0.1:
embedm = k
break
return embedm
def fnn(data, maxm):
"""
Compute the embedding dimension of a time series data to build the phase space using the false neighbors criterion
data--> time series
maxm--> maximmum embeding dimension
"""
RT = 15.0
AT = 2
sigmay = np.std(data, ddof=1)
nyr = len(data)
m = maxm
EM = lagmat(data, maxlag=m - 1)
EEM = np.asarray([EM[j, :] for j in range(m - 1, EM.shape[0])])
embedm = maxm
for k in range(AT, EEM.shape[1] + 1):
fnn1 = []
fnn2 = []
Ma = EEM[:, range(k)]
D = dist(Ma)
for i in range(1, EEM.shape[0] - m - k):
#print D.shape
#print(D[i,range(i-1)])
d = D[i, :]
pdnz = np.where(d > 0)
dnz = d[pdnz]
Rm = np.min(dnz)
l = np.where(d == Rm)
l = l[0]
l = l[len(l) - 1]
if l + m + k - 1 < nyr:
fnn1.append(
np.abs(data[i + m + k - 1] - data[l + m + k - 1]) / Rm)
fnn2.append(
np.abs(data[i + m + k - 1] - data[l + m + k - 1]) / sigmay)
Ind1 = np.where(np.asarray(fnn1) > RT)
Ind2 = np.where(np.asarray(fnn2) > AT)
if len(Ind1[0]) / float(len(fnn1)) < 0.1 and len(Ind2[0]) / float(
len(fnn2)) < 0.1:
embedm = k
break
return embedm
def Dim_Corr(datas, Tao, m, graph=False):
"""
Compute the correlation dimension of a time series with a time-lag Tao and an embedding dimension m
datas--> time series to compute the correlation dimension
Tao--> time lag computed using the first zero crossing of the auto-correlation function (see Tao func)
m--> embeding dimension of the time-series, computed using the false neighbors method (see fnn func)
graph (optional)--> plot the phase space (attractor) in 3D
"""
x = PhaseSpace(datas, m, Tao, graph)
ED2 = dist(x.T)
posD = np.triu_indices_from(ED2, k=1)
ED = ED2[posD]
max_eps = np.max(ED)
min_eps = np.min(ED[np.where(ED > 0)])
max_eps = np.exp(math.floor(np.log(max_eps)))
n_div = int(math.floor(np.log(max_eps / min_eps)))
n_eps = n_div + 1
eps_vec = range(n_eps)
unos = np.ones([len(eps_vec)]) * -1
eps_vec1 = max_eps * np.exp(unos * eps_vec - unos)
Npairs = ((len(x[1, :])) * ((len(x[1, :]) - 1)))
C_eps = np.zeros(n_eps)
for i in eps_vec:
eps = eps_vec1[i]
N = np.where(((ED < eps) & (ED > 0)))
S = len(N[0])
C_eps[i] = float(S) / Npairs
omit_pts = 1
k1 = omit_pts
k2 = n_eps - omit_pts
xd = np.log(eps_vec1)
yd = np.log(C_eps)
xp = xd[k1:k2]
yp = yd[k1:k2]
p = np.polyfit(xp, yp, 1)
return p[0]
def fnn(data, maxm):
"""
Compute the embedding dimension of a time series data to build the phase space using the false neighbors criterion
data--> time series
maxm--> maximmum embeding dimension
"""
RT=15.0
AT=2
sigmay=np.std(data, ddof=1)
nyr=len(data)
m=maxm
EM=lagmat(data, maxlag=m-1)
EEM=np.asarray([EM[j,:] for j in range(m-1, EM.shape[0])])
embedm=maxm
for k in range(AT,EEM.shape[1]+1):
fnn1=[]
fnn2=[]
Ma=EEM[:,range(k)]
D=dist(Ma)
for i in range(1,EEM.shape[0]-m-k):
#print D.shape
#print(D[i,range(i-1)])
d=D[i,:]
pdnz=np.where(d>0)
dnz=d[pdnz]
Rm=np.min(dnz)
l=np.where(d==Rm)
l=l[0]
l=l[len(l)-1]
if l+m+k-1<nyr:
fnn1.append(np.abs(data[i+m+k-1]-data[l+m+k-1])/Rm)
fnn2.append(np.abs(data[i+m+k-1]-data[l+m+k-1])/sigmay)
Ind1=np.where(np.asarray(fnn1)>RT)
Ind2=np.where(np.asarray(fnn2)>AT)
if len(Ind1[0])/float(len(fnn1))<0.1 and len(Ind2[0])/float(len(fnn2))<0.1:
embedm=k
break
return embedm
def Dim_Corr(datas, Tao, m, graph=False):
"""
Compute the correlation dimension of a time series with a time-lag Tao and an embedding dimension m
datas--> time series to compute the correlation dimension
Tao--> time lag computed using the first zero crossing of the auto-correlation function (see Tao func)
m--> embeding dimension of the time-series, computed using the false neighbors method (see fnn func)
graph (optional)--> plot the phase space (attractor) in 3D
"""
x=PhaseSpace(datas, m, Tao, graph)
ED2=dist(x.T)
posD=np.triu_indices_from(ED2, k=1)
ED=ED2[posD]
max_eps=np.max(ED)
min_eps=np.min(ED[np.where(ED>0)])
max_eps=np.exp(math.floor(np.log(max_eps)))
n_div=int(math.floor(np.log(max_eps/min_eps)))
n_eps=n_div+1
eps_vec=range(n_eps)
unos=np.ones([len(eps_vec)])*-1
eps_vec1=max_eps*np.exp(unos*eps_vec-unos)
Npairs=((len(x[1,:]))*((len(x[1,:])-1)))
C_eps=np.zeros(n_eps)
for i in eps_vec:
eps=eps_vec1[i]
N=np.where(((ED<eps) & (ED>0)))
S=len(N[0])
C_eps[i]=float(S)/Npairs
omit_pts=1
k1=omit_pts
k2=n_eps-omit_pts
xd=np.log(eps_vec1)
yd=np.log(C_eps)
xp=xd[k1:k2]
yp=yd[k1:k2]
p = np.polyfit(xp, yp, 1)
return p[0]
def correlation_dim(datas: List[np.ndarray], tau: int, d: int) -> float:
x = phase_space(datas, d, tau)
print('Finding correlation dimension ...')
ED2 = dist(x.T)
posD = np.triu_indices_from(ED2, k=1)
ED = ED2[posD]
max_eps = np.max(ED)
min_eps = np.min(ED[np.where(ED > 0)])
max_eps = np.exp(math.floor(np.log(max_eps)))
n_div = int(math.floor(np.log(max_eps / min_eps)))
n_eps = n_div + 1
eps_vec = range(n_eps)
unos = np.ones([len(eps_vec)]) * -1
eps_vec1 = max_eps * np.exp(unos * eps_vec - unos)
Npairs = ((len(x[1, :])) * ((len(x[1, :]) - 1)))
C_eps = np.zeros(n_eps)
for i in eps_vec:
eps = eps_vec1[i]
N = np.where(((ED < eps) & (ED > 0)))
S = len(N[0])
C_eps[i] = float(S) / Npairs
omit_pts = 1
k1 = omit_pts
k2 = n_eps - omit_pts
xd = np.log(eps_vec1)
yd = np.log(C_eps)
xp = xd[k1:k2]
yp = yd[k1:k2]
p = np.polyfit(xp, yp, 1)
return p[0]
def kernel(x, y, lengthscale):
return np.exp(-dist(x, y, squared=True) / (lengthscale**2)) + 1e-300
def get_distance(points):
"""points assumed to be latitue and longitude in m"""
return dist(np.radians(points)) * 6371 * 1000
def centroid(class_x_train):
mean = np.mean(class_x_train, axis=0, keepdims=True)
distances = dist(class_x_train, mean)
return np.argmin(distances)
query_radius = 15 x_ref = tracker.particles.mean(axis=0)[:2] skip=1 ref_points = s_init.points[s_init.tree.query_ball_point(x_ref,query_radius)][::skip] train_size = min(1000,ref_points.shape[0]) rand_ind = np.random.choice(range(len(ref_points)),train_size,replace=False) ref_points = ref_points[rand_ind] l = 0.1 sigma2 = 0.01 r_mean = ref_points.mean(axis=0) r_std = ref_points.std(axis=0) x_train = (ref_points - r_mean)/r_std K = np.exp(-dist(x_train[:,:2],x_train[:,:2],squared=True)/l**2) + sigma2*np.eye(x_train.shape[0]) Kinv = np.linalg.inv(K) particles_init = tracker.particles.copy() init_index = np.linspace(0,len(particles_init)-1,len(particles_init)).astype(int) particle_list = [] mean_list = [] fig,axs = plt.subplots() fig.set_size_inches(12,12) s0 = scanset.index(0) particles_0 = tracker.particles.copy() w = np.ones(len(particles_0)) w/=w.sum()