Question: How do we decide serve placement from court-vision raw data or pandas df?

[kiszk]

While I quickly browsed a json file with court vision results at ATP site, I cannot find how to know the serve placement from a json file or processed pandas df.

The use cases of the serve replacements are as follows:

• To determine the course (e.g. T, middle, or wide)
• To draw the placement on the court using matplot like your analysis

Although I thought x_bounce or y_bounce at stroke_idx==1 in the processed df may be related to the serve placement, I think that it does not make sense.

@glad94 Could you please share your approach?

[kiszk]

Sorry for bothering you.

Is the function categorise_serve_direction() in the jupyter file still valid for this pipeline?

[glad94]

Hi @kiszk thanks for the interest and question , I didn't use the above function from Peter's repo but wrote something separately that's probably similar:

def serve_dir(y_bounce):
    '''
    Returns serve direction
    '''
    width = 8.23/2
    if abs(y_bounce) < width/3:
        dir = "T"
    elif abs(y_bounce) <= width*2/3:
        dir = "M"
    else:
        dir = "W"
    return dir

My bad for not documenting the processed ball coordinate (x,y,z) columns as well, this is what they're meant to be:

• hit: when the ball is struck
• peak_pre: the peak of the ball trajectory before passing the net
• net: when the ball crosses the net
• bounce: when the ball bounces
• peak_post: the peak of the ball trajectory after bouncing (before the next shot)

If the value is -999, then either the value is missing in the raw data, or that part of the ball trajectory didn't happen e.g. bounce values could be -999 if the next shot was a volley.

[kiszk]

@glad94 Thank you for your response. It makes sense. And, you function is equal to Peter' function. Also, thank you for documenting the ball coordinate columns.

This pipeline is a very useful framework to process court vision data on pandas.