python - Improve performance of events duration computation

Question

I have a Pandas DataFrame that looks like that:

ipdb> df[:4]
                name      thread_id  state     time  uuid
0     CB:Component_Loc    829946510      1  2683817     0
1     CB:Component_Loc    829946510      0  2683874     0
2  CB:Component_Fusion   3025005749      1  2683683     1
3  CB:Component_Fusion   3025005749      0  2683882     1

thread_id is not important here. The data represents time points when an event (identified by uuid was started (state = 1) and stopped (state = 0).

From that, I want to compute the duration and time (defined as mean between start and stop). I can achieve this with this code:

    df = df.groupby("uuid").apply(lambda df:
                                  pd.Series({
                                      "name" : df.iloc[0]["name"],
                                      "time" : df["time"].mean(),
                                      "duration": df["time"].diff().iat[1]
                                  }))
    df = df.pivot(columns="name", values="duration", index="time")

which works fine, but is very slow for the number of events I have. Furthermore, it is not really what I consider elegant.

What are ways to improve that code, mostly for performance?

EDIT: Some additional information, as requested:

name is not unique, i.e., there can be many events named CB:Component_Loc. However uuid is unique for a start/stop cycle.

What I want is the duration (time when state = 0 minus time when state = 1) for each uuid.

Thanks!

Answer 1

等待大神答复