将代码转换为张量流的优势

Question

我正在尝试将下面的代码转换为tensorflow。我对tensorflow很陌生，我想知道下面的代码中是否有任何部分可以在tensorflow中更快地运行？有没有办法在张量流中完成相同的事情，并使其运行更快？例如，像最终的点产品一样，像matmul那样效果会更好吗？

代码：

    # creates dataframe with columns from x_df and index from x_df
    astk_df = pd.DataFrame(columns=x_df.columns, index=x_df.index)
    for i, m in enumerate(astk_df.columns):

        # creates array of values from astk_df
        x = x_df[m].values

        # appends 0.0 to the beginning of array
        x_extend = np.concatenate((np.zeros(L-1), x)) #we extend the media spend series with zeros to 
                                                      #compute initial adstock values
        # sets weights equal to final alpha value raised to a range of powers
        weights = alpha[i]**(np.arange(0, L))
        weights /= weights.sum()              #normalization constant
        for t, week in enumerate(astk_df.index):
            astk_df.loc[week, m] = x_extend[t:(t+L)][::-1].dot(weights)