Deep walk模型 詳細解釋

1 資料解釋

資料集：wiki資料集（2405個網頁，17981條網頁間的關系）

輸入樣本：node1 node2 <edge_weight>

輸出：每個node的embedding

根據随機遊走的序列，輸入到word2vec的模型當中，然後就能訓練後表示出該節點的embedding

2 代碼思想

步驟① 建構一個有向圖 ② 進行deepwalk取樣本 ③ 輸入到word2vec當中訓練 ④ 得到了訓練好的word2vec，進行evaluate

⑤ 資料分為x_train,x_test, 使用logistics函數，訓練embedding後x_train對應的label，然後通過logstics函數預測x_test的标簽（标簽是0-16所屬分類）

⑥畫圖

① 建構一個有向圖

G = nx.read_edgelist('../data/wiki/Wiki_edgelist.txt',
                     create_using=nx.DiGraph(), nodetype=None, data=[('weight', int)])
           

② 進行deepwalk取樣本 其中 deepwalk就是根據步長，進行深度優先周遊，如果該節點沒有鄰居節點了，就break出來

def deepwalk_walk(self, walk_length, start_node):
    walk = [start_node]
    while len(walk) < walk_length:
        cur = walk[-1]
        cur_nbrs = list(self.G.neighbors(cur))
        if len(cur_nbrs) > 0:
            walk.append(random.choice(cur_nbrs))
        else:
            break
    return walk
           

③ 輸入到word2vec當中訓練，得到該點的embedding表示

④ 得到了訓練好的word2vec，得到了2405X128的embedding,然後進行模型的評價

class Classifier(object):

    def __init__(self, embeddings, clf):
        self.embeddings = embeddings
        self.clf = TopKRanker(clf)
        self.binarizer = MultiLabelBinarizer(sparse_output=True)   # multi one-hot


    def train(self, X, Y, Y_all):
        self.binarizer.fit(Y_all)
        X_train = [self.embeddings[x] for x in X]
        Y = self.binarizer.transform(Y)
        self.clf.fit(X_train, Y)

    def evaluate(self, X, Y):
        top_k_list = [len(l) for l in Y]
        Y_ = self.predict(X, top_k_list)
        Y = self.binarizer.transform(Y)
        averages = ["micro", "macro", "samples", "weighted"]
        results = {}
        for average in averages:
            results[average] = f1_score(Y, Y_, average=average)
        results['acc'] = accuracy_score(Y,Y_)
        print('-------------------')
        print(results)
        return results
        print('-------------------')

    def predict(self, X, top_k_list):
        X_ = numpy.asarray([self.embeddings[x] for x in X])
        Y = self.clf.predict(X_, top_k_list=top_k_list)
        return Y

    def split_train_evaluate(self, X, Y, train_precent, seed=0):
        state = numpy.random.get_state() # 使随機生成器保持相同狀态

        training_size = int(train_precent * len(X))
        numpy.random.seed(seed)
        shuffle_indices = numpy.random.permutation(numpy.arange(len(X))) # 随機打亂順序
        X_train = [X[shuffle_indices[i]] for i in range(training_size)]
        Y_train = [Y[shuffle_indices[i]] for i in range(training_size)]
        X_test = [X[shuffle_indices[i]] for i in range(training_size, len(X))]
        Y_test = [Y[shuffle_indices[i]] for i in range(training_size, len(X))]
        '''前面部分都是為了打亂順序'''
        self.train(X_train, Y_train, Y)  # 把label轉成Multi hot 形式，把x_train轉成embedding形式
        numpy.random.set_state(state)
        return self.evaluate(X_test, Y_test)
           

⑤ 先使用split_train_evaluate 函數，

(1）打亂資料 （2）把embedding後的x_train與 轉換為mulit-hot的 Y 通過logistics 模型進行訓練

(3) 把測試資料x_test 傳入 logistics模型，得到評估的F1 score

首先測試資料放進logistics模型，并選擇topk的目标值，打上标簽1

然後轉換回label标簽，通過f1 score進行計算

class TopKRanker(OneVsRestClassifier):
    def predict(self, X, top_k_list):
        probs = numpy.asarray(super(TopKRanker, self).predict_proba(X)) # 預測出來NX17，一共有17個label
        all_labels = []
        for i, k in enumerate(top_k_list):
            probs_ = probs[i, :]
            labels = self.classes_[probs_.argsort()[-k:]].tolist()
            probs_[:] = 0
            probs_[labels] = 1
            all_labels.append(probs_)
        return numpy.asarray(all_labels)

def evaluate(self, X, Y):
    top_k_list = [len(l) for l in Y]
    Y_ = self.predict(X, top_k_list)
    Y = self.binarizer.transform(Y)  # 轉換為label值
    averages = ["micro", "macro", "samples", "weighted"]
    results = {}
    for average in averages:
        results[average] = f1_score(Y, Y_, average=average)
    results['acc'] = accuracy_score(Y,Y_)
    print('-------------------')
    print(results)
    return results
    print('-------------------')
           

⑥畫圖

通過TSNE進行降維，然後根據 embedding降維後與 wiki_labels的值對應上，畫出二維效果圖

參考文章：

https://zhuanlan.zhihu.com/p/56380812