preprocessing.py 文件源码

def make_subset(coocc_features, x_axis, y_axis):
    logsource = np.log(coocc_features.ix[x_axis][y_axis]+1)
    x_sorted = logsource.ix[logsource.sum(axis=1).sort_values(ascending=False).index]
    y_sorted = x_sorted.T.ix[x_sorted.T.sum(axis=1).sort_values(ascending=False).index]
    logsource = y_sorted.T.ix[:25, :25]
    n_cols = len(logsource.columns)
    n_rows = len(logsource.index)
    df = pd.DataFrame()
    df["x"] = list(itertools.chain.from_iterable(list(itertools.repeat(i, times=n_cols)) for i in logsource.index))
    df["y"] = list(itertools.chain.from_iterable(list(itertools.repeat(logsource.stack().index.levels[1].values, times=n_rows))))
    df["counts"] = logsource.stack().values
    df["raw"] = df["counts"].map(np.exp)-1
    df.sort_values("counts", ascending=False, inplace=True)

    new_axis_factors = logsource.index.values.tolist()

    return df, new_axis_factors, new_axis_factors