def regression_embedding():
import regression_based
class_labels = unpickle('cifar-10-batches-py/batches.meta')['label_names']
# print class_labels
Y_8_train = np.array(Y_train)
X_8_train = np.array(X_train)
removed_indices = np.where(Y_8_train != 1)
Y_8_train = Y_8_train[removed_indices]
X_8_train = X_8_train[removed_indices]
removed_indices = np.where(Y_8_train != 4)
Y_8_train = Y_8_train[removed_indices]
X_8_train = X_8_train[removed_indices]
Y_8_train = [embeddings[class_labels[i]] for i in Y_8_train]
Y_8_validation = np.array(Y_validation)
X_8_validation = np.array(X_validation)
regression_based.train(X_8_train, Y_8_train)
Y_2_validation = np.array(Y_validation)
X_2_validation = np.array(X_validation)
indices = np.where(np.logical_or(Y_2_validation == 1, Y_2_validation == 4))
Y_2_validation = Y_2_validation[indices]
X_2_validation = X_2_validation[indices]
validaiton_embeddings = regression_based.predict(X_2_validation)
targets_embeddings = []
# for i in class_labels:
for i in [
'airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog',
'horse', 'ship', 'truck'
]:
targets_embeddings.append(embeddings[i])
targets_embeddings = np.array(targets_embeddings, dtype=np.float32)
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from scipy.spatial.distance import cosine
# Y_pred_validation = []
# for i in validaiton_embeddings:
# cos = []
# for j in targets_embeddings:
# val = cosine(i,j)
# cos.append(val)
# Y_pred_validation.append(np.argmax(cos))
#
neigh = KNeighborsClassifier(n_neighbors=1)
neigh.fit(targets_embeddings, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
Y_pred_validation = neigh.predict(validaiton_embeddings)
print(Y_2_validation)
print(Y_pred_validation)
print(accuracy_score(Y_2_validation, Y_pred_validation))
def regression_embedding():
import regression_based
class_labels = unpickle('cifar-10-batches-py/batches.meta')['label_names']
#Y_8_train = np.array(Y_train)
#X_8_train = np.array(X_train)
#removed_indices = np.where(Y_8_train!=1)
#Y_8_train = Y_8_train[removed_indices]
#X_8_train = X_8_train[removed_indices]
#removed_indices = np.where(Y_8_train!=4)
#Y_8_train = Y_8_train[removed_indices]
#X_8_train = X_8_train[removed_indices]
Y_GO_train = np.array(Y_train)
X_GO_train = np.array(X_train)
Y_GO_test = np.array(Y_validation)
X_GO_test = np.array(X_validation)
#Y_8_train = [embeddings[class_labels[i]] for i in Y_8_train]
Y_GO_train = [embeddings[class_labels[i]] for i in np.array(Y_GO_train)]
#Y_8_validation = np.array(Y_validation)
#X_8_validation = np.array(X_validation)
regression_based.train(X_GO_train, Y_GO_train)
#Y_2_validation = np.array(Y_validation)
#X_2_validation = np.array(X_validation)
#indices = np.where(np.logical_or(Y_2_validation==1 ,Y_2_validation==4))
#Y_2_validation = Y_2_validation[indices]
#X_2_validation = X_2_validation[indices]
validaiton_embeddings = regression_based.predict(X_GO_test)
targets_embeddings_words = []
targets_embeddings = []
for i in class_labels:
targets_embeddings.append(embeddings[i])
targets_embeddings_words.append(i)
for k in embeddings:
if k not in class_labels:
targets_embeddings.append(embeddings[k])
targets_embeddings_words.append(k)
targets_embeddings = np.array(targets_embeddings, dtype=np.float32)
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from scipy.spatial.distance import cosine
# Y_pred_validation = []
# for i in validaiton_embeddings:
# cos = []
# for j in targets_embeddings:
# val = cosine(i,j)
# cos.append(val)
# Y_pred_validation.append(np.argmax(cos))
#
neigh = KNeighborsClassifier(n_neighbors=1)
print len(targets_embeddings)
neigh.fit(targets_embeddings, range(0, len(targets_embeddings)))
Y_pred_validation = neigh.predict([validaiton_embeddings[0]])
print Y_pred_validation
neighbors = neigh.kneighbors([validaiton_embeddings[0]],
20,
return_distance=False)
print neighbors
for n in neighbors[0]:
print targets_embeddings_words[n]
removed_indices = np.where(Y_8_train != 1)
Y_8_train = Y_8_train[removed_indices]
X_8_train = X_8_train[removed_indices]
removed_indices = np.where(Y_8_train != 4)
Y_8_train = Y_8_train[removed_indices]
X_8_train = X_8_train[removed_indices]
Y_8_train = [
embeddings[str(class_labels[i], encoding='ascii')] for i in Y_8_train
]
Y_8_validation = np.array(Y_validation)
X_8_validation = np.array(X_validation)
# train with L2 loss regression
regression_based.train(X_8_train, Y_8_train)
# validation using 1, 4 data
Y_2_validation = np.array(Y_validation)
X_2_validation = np.array(X_validation)
indices = np.where(np.logical_or(Y_2_validation == 1, Y_2_validation == 4))
Y_2_validation = Y_2_validation[indices]
X_2_validation = X_2_validation[indices]
# predict image data of label 1, 4
validaiton_embeddings = regression_based.predict(X_2_validation)
# embedding of label 1, 4
targets_embeddings = []
for i in ['automobile', 'deer']:
targets_embeddings.append(embeddings[i])