def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
rn = np.random.randn
# refresh weight
W_in = 0.01 * rn(V, H).astype('f')
W_out = 0.01 * rn(V, H).astype('f')
# make class
self.in_layer = Embedding(W_in)
self.loss_layers = []
for i in range(2 * window_size):
layer = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
self.loss_layers.append(layer)
# put all weights and gradients in one list
layers = [self.in_layer] + self.loss_layers
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# save a word's variation in instance variable
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
rn = np.random.randn
# 重みの初期化
W_in = 0.01 * rn(V, H).astype('f')
W_out = 0.01 * rn(V, H).astype('f')
# レイヤの生成
self.in_layer = Embedding(W_in)
self.loss_layers = []
for i in range(2 * window_size):
layer = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
self.loss_layers.append(layer)
# すべての重みと勾配をリストにまとめる
layers = [self.in_layer] + self.loss_layers
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# メンバ変数に単語の分散表現を設定
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# initialize weight
W_in = 0.01 * np.random.randn(V, H).astype('f')
W_out = 0.01 * np.random.randn(V, H).astype('f')
# create layer
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in)
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
# combine all weights and grads into array
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# set word vector to member variable
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
rn = np.random.randn
# 가중치 초기화
W_in = 0.01 * rn(V, H).astype('f')
W_out = 0.01 * rn(V, H).astype('f')
# 계층 생성
self.in_layer = Embedding(W_in)
self.loss_layers = []
for i in range(2 * window_size):
layer = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
self.loss_layers.append(layer)
# 모든 가중치와 기울기를 리스트에 모은다.
layers = [self.in_layer] + self.loss_layers
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# 인스턴스 변수에 단어의 분산 표현을 저장한다.
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# 가중치 초기화
W_in = 0.01 * np.random.randn(V, H).astype('f')
W_out = 0.01 * np.random.randn(V, H).astype(
'f') # embedding layer를 사용하므로 둘의 형상이 같다.
# 계층 생성
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in) # Embedding 계층 2*window_size만큼 사용
self.in_layers.append(layer) # 배열로 보관
self.ns_loss = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
# 모든 가중치와 기울기를 배열에 모은다.
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# 인스턴스 변수에 단어의 분산 표현을 저장한다.
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# 重みの初期化
W_in = 0.01 * np.random.randn(V, H).astype('f')
W_out = 0.01 * np.random.randn(V, H).astype('f')
# レイヤの生成
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in) # Embeddingレイヤを使用
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
# すべての重みと勾配を配列にまとめる
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# メンバ変数に単語ベクトルを設定
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# initialize of weight
W_in = 0.01 * np.random.randn(V, H).astype("f")
W_out = 0.01 * np.random.randn(V, H).astype("f")
# making layers
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in) # using Embedding layer
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
# conclude all of weights & grads
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# Set word representations as a member var.
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# 重みの初期化
W_in = tf.Variable(
tf.random.normal((V, H), mean=0.0, stddev=0.01, dtype='float'))
W_out = tf.Variable(
tf.random.normal((V, H), mean=0.0, stddev=0.01, dtype='float'))
# レイヤの生成
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in) # Embeddingレイヤを使用
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out,
corpus,
power=0.75,
sample_size=5)
# すべての重みと勾配をリストにまとめる
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# メンバ変数に単語の分散表現を設定
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus): #10000, 100, 5, corpus
V, H = vocab_size, hidden_size # 10000, 100
# 가중치 초기화
W_in = 0.01 * np.random.randn(V, H).astype('f') # 10000, 100
W_out = 0.01 * np.random.randn(V, H).astype('f') # 10000, 100
# 계층 생성
self.in_layers = []
for i in range(2 * window_size): # 10
layer = Embedding(W_in) # Embedding 계층 사용, weight-sharing 통해서 10개 layer 생성.
self.in_layers.append(layer) # 10개 layer 이어줌, 객체 리스트 [객체1, 객체2, ...]
self.ns_loss = NegativeSamplingLoss(W_out, corpus, power=0.75, sample_size=5)
# 모든 가중치와 기울기를 배열에 모은다.
layers = self.in_layers + [self.ns_loss] # 객체들의 묶음, 객체 리스트 [layer객체1, layer객체2, ..., Negativ객체] - 11개 객체
self.params, self.grads = [], []
for layer in layers: # 10개 layer, 1개 negative params,grads를 모음 [array(layer1_weight), array(layer2_weight2), ..., array(negativ_weight(w_out)]
self.params += layer.params
self.grads += layer.grads
# 인스턴스 변수에 단어의 분산 표현을 저장한다.
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
W_in = 0.01 * np.random.randn(V, H).astype('f')
W_out = 0.01 * np.random.randn(V, H).astype('f')
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in)
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out, corpus, power=0.75)
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
self.word_vecs = W_in
def __init__(self, vocab_size, hidden_size, window_size, corpus):
V, H = vocab_size, hidden_size
# 初始化权重
W_in = 0.01 * np.random.randn(V, H).astype('f')
W_out = 0.01 * np.random.randn(V, H).astype('f')
# 生成层
self.in_layers = []
for i in range(2 * window_size):
layer = Embedding(W_in) # 使用Embedding层
self.in_layers.append(layer)
self.ns_loss = NegativeSamplingLoss(W_out, corpus, power=0.75, sample_size=5)
# 将所有的权重和梯度整理到列表中
layers = self.in_layers + [self.ns_loss]
self.params, self.grads = [], []
for layer in layers:
self.params += layer.params
self.grads += layer.grads
# 将单词的分布式表示设置为成员变量
self.word_vecs = W_in