def main():
dimension = 32
X, y = data_processing.read_data('Data/conmat_240.mat', 'Data/age_240.mat')
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=.8)
# average matrix over train data
avg_matrix = X_train.mean(axis=0)
# generate random walks
walk = random_walk(avg_matrix, steps=1000)
seq = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
seq[i, :] = avg_matrix[pos]
print(seq.shape)
skipgram = Skip_Gram(268, dimension, 2, 0.1)
skipgram.train_from_feature_seq(seq, epochs=200)
embedded_train_matrix = np.zeros((len(X_train), 268 * dimension))
for i in range(len(X_train)):
embedding_train = skipgram.encode(X_train[i])
embedded_train_matrix[i] = np.ndarray.flatten(embedding_train)
embedded_test_matrix = np.zeros((len(X_test), 268 * dimension))
for i in range(len(X_test)):
embedding_test = skipgram.encode(X_test[i])
embedded_test_matrix[i] = np.ndarray.flatten(embedding_test)
lasso = Lasso(100, .01)
lasso.train_coordinate_descent(embedded_train_matrix, y_train)
predicted = lasso.predict(embedded_test_matrix)
print(mean_squared_error(y_test, predicted))
def main():
X, y = data_processing.read_data('maps_conmat.mat', 'maps_age.mat')
#X = data_processing.adjacency_matrix(X)
avg_matrix = X.mean(axis = 0)
print(avg_matrix.shape)
model = AutoEncoder(X.shape[-1], 64, activation = 'relu')
model.train(X, epochs = 200, learning_rate = 0.001, loss = 'mse')
#generate_embedding_vis(avg_matrix, model.encode(avg_matrix), embedding_name='Neural Autoencoder')
walk = random_walk(avg_matrix, steps = 1000)
seq = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
seq[i, :] = avg_matrix[pos]
print(seq.shape)
skipgram = Skip_Gram(268, 64, 2, 0.1)
skipgram.train_from_feature_seq(seq, epochs = 200)
#generate_embedding_vis(avg_matrix, skipgram.encode(avg_matrix), embedding_name='SkipGram')
cbow = CBOW(268, 64, 2, 0.1)
cbow.train_from_feature_seq(seq, epochs = 200)
#generate_embedding_vis(avg_matrix, cbow.encode(avg_matrix), embedding_name='CBOW')
distances = [[avg_matrix, model.encode(avg_matrix)], [skipgram.encode(avg_matrix), cbow.encode(avg_matrix)]]
names = [['Original Distances', 'Autoencoder Distances'], ['SkipGram Distances', 'CBOW Distances']]
generate_embedding_vis_array(distances, names)
def cbow(train, evaluate, embedding_dim, sentence_length, window, epochs, learning_rate):
from embeddings.random_walk import random_walk
from embeddings.word2vec import CBOW
Xm = train.mean(axis = 0)
walk = random_walk(Xm, steps = sentence_length)
one_hot = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
one_hot[i, :] = Xm[pos]
model = CBOW(268, embedding_dim, window, learning_rate)
model.train_from_feature_seq(one_hot, epochs = epochs)
return model.encode(evaluate)
def main():
X, y = data_processing.read_data('maps_conmat.mat', 'maps_age.mat')
Xm = X.mean(axis=0)
walk = random_walk(Xm, steps=1000)
one_hot = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
one_hot[i, :] = Xm[pos]
#Skip-Gram
model = Skip_Gram(268, 64, 2, 0.1)
model.train_from_feature_seq(one_hot, epochs=200)
generate_embedding_vis(Xm, model.encode(Xm), embedding_name="Skip-Gram")
#CBOW
model = CBOW(268, 64, 2, 0.1)
model.train_from_feature_seq(one_hot, epochs=200)
generate_embedding_vis(Xm, model.encode(Xm), embedding_name="CBOW")
def main():
X, y = data_processing.read_data('Data/conmat_240.mat', 'Data/age_240.mat')
Xm = X.mean(axis=0)
EMBEDDING_DIM = 16
#Fully-Connected AutoEncoder
e_x = tf.keras.layers.Input((None, X.shape[-1]))
e_o = tf.keras.layers.TimeDistributed(
tf.keras.layers.Dense(EMBEDDING_DIM, activation='tanh'))(e_x)
e = tf.keras.Model(e_x, e_o)
d_x = tf.keras.layers.Input((None, EMBEDDING_DIM))
d_o = tf.keras.layers.TimeDistributed(
tf.keras.layers.Dense(X.shape[-1], activation='linear'))(d_x)
d = tf.keras.Model(d_x, d_o)
ae_model = AutoEncoder(e, d)
ae_model.train(X, epochs=50, learning_rate=0.001, loss='mse')
#Transformer AutoEncoder
et_x = tf.keras.layers.Input((X.shape[1], X.shape[2]))
et_o = Transformer(EMBEDDING_DIM, heads=8, activation='tanh')(et_x)
et = tf.keras.Model(et_x, et_o)
dt_x = tf.keras.layers.Input((X.shape[1], EMBEDDING_DIM))
dt_o = Transformer(X.shape[2], heads=8, activation='linear')(dt_x)
dt = tf.keras.Model(dt_x, dt_o)
ae_modelt = AutoEncoder(et, dt)
ae_modelt.train(X, epochs=100, learning_rate=0.001, loss='mse')
#Matrix Factorization
mat_factorization = MatrixFactorization(Xm, EMBEDDING_DIM)
mat_factorization.fit(200, 0.0001)
#Tensor Factorization
tens_factorization = TensorFactorization(X, EMBEDDING_DIM)
tens_factorization.fit(50)
walk = random_walk(Xm, steps=1000)
one_hot = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
one_hot[i, :] = Xm[pos]
#Skip-Gram
skipgram = Skip_Gram(268, EMBEDDING_DIM, 3, 0.1)
skipgram.train_from_feature_seq(one_hot, epochs=200)
#CBOW
cbow = CBOW(268, EMBEDDING_DIM, 3, 0.1)
skipgram.train_from_feature_seq(one_hot, epochs=200)
og_distances = calculate_distance_matrix(X.reshape((len(X), -1)))
models = {
'AutoEncoder': ae_model,
'Transformer': ae_modelt,
'Matrix Factorization': mat_factorization,
'Tensor Factorization': tens_factorization,
'Skip-Gram': skipgram,
'CBOW': cbow
}
model_distances = {}
for key, mod in models.items():
x_embed = mod.encode(X)
model_distances[key] = calculate_distance_matrix(
x_embed.reshape((len(x_embed), -1)))
#plot distances
plt.matshow(og_distances, cmap='Blues', vmin=0)
plt.title('Original Distances')
plt.savefig('images/og_distance_matrix.png')
fig, axes = plt.subplots(2, 3)
i = 0
for embedding_name, embedding_distances in model_distances.items():
r, c = i // 3, i % 3
axes[r, c].matshow(embedding_distances, cmap='Blues', vmin=0)
axes[r, c].set_title(embedding_name)
i += 1
fig.savefig('images/embedding_distances_matrix.png')
def main():
# dimensions to test
DIMENSIONS = [64, 32, 16, 8, 4, 2]
X, y = data_processing.read_data('Data/conmat_240.mat', 'Data/age_240.mat')
X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=.8)
# average matrix over train data
avg_matrix = X_train.mean(axis=0)
# generate random walks
walk = random_walk(avg_matrix, steps=1000)
seq = np.zeros((len(walk), 268))
for i, pos in enumerate(walk):
seq[i, :] = avg_matrix[pos]
print(seq.shape)
# train embeddings for each dimension
skipgrams = list()
for dimension in DIMENSIONS:
print(str(dimension) + "-D Embedding Training")
skipgram = CBOW(268, dimension, 2, 0.1)
skipgram.train_from_feature_seq(seq, epochs=300)
skipgrams.append((skipgram, dimension))
# encode train and test data using embeddings, then flatten for prediction
embedded_train_list = list()
embedded_test_list = list()
for skipgram in skipgrams:
embedded_train_matrix = np.zeros((len(X_train), 268 * skipgram[1]))
for i in range(len(X_train)):
embedding_train = skipgram[0].encode(X_train[i])
embedded_train_matrix[i] = np.ndarray.flatten(embedding_train)
embedded_train_list.append(embedded_train_matrix)
embedded_test_matrix = np.zeros((len(X_test), 268 * skipgram[1]))
for i in range(len(X_test)):
embedding_test = skipgram[0].encode(X_test[i])
embedded_test_matrix[i] = np.ndarray.flatten(embedding_test)
embedded_test_list.append(embedded_test_matrix)
# train prediction models on encoded train data, then test on encoded test data and calculate Mean Squared Error
lr_error_list = list()
svr_error_list = list()
mlp_error_list = list()
for i in range(len(embedded_train_list)):
#savemat(f'Data/cbow_{DIMENSIONS[i]}.mat', {'train':embedded_train_list[i] ,'test':embedded_test_list[i]})
lr = Ridge().fit(embedded_train_list[i], y_train)
svr = SVR().fit(embedded_train_list[i], np.reshape(y_train, -1))
mlp = MLPRegressor(hidden_layer_sizes=(100,)).fit(embedded_train_list[i], np.reshape(y_train, -1))
print(mlp.loss_)
predictedLR = lr.predict(embedded_test_list[i])
predictedSV = svr.predict(embedded_test_list[i])
predictedMLP = mlp.predict(embedded_test_list[i])
print(str(embedded_test_list[i].shape[-1] // 268) + "-D Predicted")
lr_error = mean_squared_error(predictedLR, y_test)
svr_error = mean_squared_error(predictedSV, y_test)
mlp_error = mean_squared_error(predictedMLP, y_test)
lr_error_list.append(lr_error)
svr_error_list.append(svr_error)
mlp_error_list.append(mlp_error)
# plot MSE for different embedding dims and prediction methods
width = 0.35
plt.bar(np.arange(len(lr_error_list)), lr_error_list, width, label="LinReg")
plt.bar(np.arange(len(svr_error_list)) + width, svr_error_list, width, label="SVR")
plt.bar(np.arange(len(mlp_error_list)) + 2 * width, mlp_error_list, width, label="MLP")
plt.ylabel("MSE")
plt.xlabel("Dimensions")
plt.title("SkipGram Mean Squared Error by Embedding Dimension")
plt.xticks(np.arange(len(svr_error_list)) + width, list(DIMENSIONS))
plt.legend(loc="best")
plt.show()
