# nca = NCA()
# nca.fit(X, y, max_iter=10, lr=100)
# print nca.A
import sys; sys.path.append('/home/shaofan/Projects')
from FastML import KNN
print "\nTest 3, LFW"
from sklearn.datasets import fetch_lfw_people
from sklearn.decomposition import PCA
from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import normalize
lfw_people = fetch_lfw_people(min_faces_per_person=70, resize=0.4)
pca = PCA(n_components=100)
X = pca.fit_transform(lfw_people.data)
X = normalize(X)
y = lfw_people.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=42)
knn = KNN(n_neighbors=4)
pred = knn.fit(X_train, y_train).predict(X_test)
print '{}/{}'.format( (pred == y_test).sum(), len(y_test) )
nca = NCA()
X_train = nca.fit(X_train, y_train, max_iter=10, lr=5e-3).transform(X_train)
X_test = nca.transform(X_test)
knn = KNN(n_neighbors=4)
pred = knn.fit(X_train, y_train).predict(X_test)
print '{}/{}'.format( (pred == y_test).sum(), len(y_test) )
# assert all(data['y']==rawdata['y'])
x, y = data["x"], data["y"]
n = x.shape[0]
x = x.reshape(n, -1)
trainx, testx, trainy, testy = train_test_split(x, y, test_size=0.32, random_state=32)
train_ind, test_ind = train_test_split(range(n), range(n), test_size=0.32, random_state=32)[:2]
# assert ((trainx==x[train_ind]).all())
# assert ((testx==x[test_ind]).all())
duration_model = LMNN_duration.load("./BIT/models/[K=300][rs=32]duration.model")
duration_trainx, duration_trainy = expand(trainx, trainy, K, G)
duration_testx, duration_testy = expand(testx, testy, K, G)
duration_trainx = duration_model.transform(duration_trainx)
duration_testx = duration_model.transform(duration_testx)
duration_knn = KNN(n_neighbor).fit(duration_trainx, duration_trainy)
print duration_trainx.shape
print duration_trainy.shape
prediction = duration_knn.predict(duration_testx)
print prediction.tolist()
mlmnn = MLMNN.load("./BIT/models/[K=300][rs=32]mlmnn.model")
mlmnn_knns = []
trunc = lambda x: x.reshape(x.shape[0], G, -1)[:, :g, :].reshape(x.shape[0], -1)
for g in range(1, G + 1):
mlmnn_knn = KNN(mlmnn.K).fit(mlmnn.transform(trunc(trainx), g), trainy)
mlmnn_knns.append(mlmnn_knn)
for g in range(1, G + 1):
# print '='*10
# acc = 0.
# print mcml.A
# print mcml.transform(X)
import sys; sys.path.append('/home/shaofan/Projects')
from FastML import KNN
print "\nTest 3, LFW"
from sklearn.datasets import fetch_lfw_people
from sklearn.decomposition import PCA
from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import normalize
lfw_people = fetch_lfw_people(min_faces_per_person=70, resize=0.4)
pca = PCA(n_components=10)
X = pca.fit_transform(lfw_people.data)
X = normalize(X)
y = lfw_people.target
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=42)
knn = KNN(n_neighbors=2)
pred = knn.fit(X_train, y_train).predict(X_test)
print '{}/{}'.format( (pred == y_test).sum(), len(y_test) )
knn = KNN(n_neighbors=2)
mcml = MCML()
for _ in range(100):
X_train = mcml.fit(X_train, y_train, max_iter=5, lr=1e-6).transform(X_train)
X_test = mcml.transform(X_test)
pred = knn.fit(X_train, y_train).predict(X_test)
print '{}/{}'.format( (pred == y_test).sum(), len(y_test) )
# x = normalize(x)
print sum(pca.explained_variance_ratio_)
trainx, testx, trainy, testy = train_test_split(x, y, test_size=0.10)
print trainx.shape
print trainy.shape
print len(set(trainy))
from sklearn.svm import SVC
# from sklearn.multiclass import OneVsRestClassifier
from sklearn.metrics import classification_report
# svm = OneVsRestClassifier(SVC(C=0.11)) # 0.90
svm = SVC(C=100.11) # 0.90
svm.fit(trainx, trainy)
print classification_report(testy, svm.predict(testx))
import sys; sys.path.append('/home/shaofan/Projects')
from FastML import NCA, KNN, MCML
knn = KNN(n_neighbors=8)
knn.fit(trainx, trainy)
print classification_report(testy, knn.predict(testx))
# nca = NCA()
# for i in range(100):
# trainx = nca.fit(trainx, trainy, max_iter=5, lr=5e-3).transform(trainx)
# testx = nca.transform(testx)
#
# print classification_report(testy, knn.predict(testx))
# sys.stdout.flush()
