snippets/python-68min-analysis-the-function.py

def to_tframe(version, df, trades, start):
    d = {'bid': {}, 'ask': {}}
    cursor = 0
    n = 0
    n_periods = 40
    xs = np.concatenate([periods(n_periods)[:0:-1] * -1, periods(n_periods)]) * 1000000 # mult to convert to nanos
    mask = df['version'] == version
    my_trades = sorted(list(zip(df.loc[mask].index.values.astype(np.int64), df.loc[mask, 'side'], df.loc[mask, 'gid'])))
    idx = trades.index.values.astype(np.int64)
    amts = trades['amount']
    totals = trades['total']
    assert len(idx) == len(amts)
    assert len(idx) == len(totals)
    for tm, side, gid in my_trades:
        print '{} to_tfame {} {} (cursor = {})'.format(time.time() - start, version, n, cursor)
        min_time = tm + xs[0]
        max_time = tm + xs[1]
        if idx[cursor] > min_time:
            print 'warning: idx[cursor] ({}) > min_time ({})'.format(idx[cursor], min_time)
            while idx[cursor] > min_time and cursor > 0:
                cursor -= 1
        else:
            while idx[cursor] < min_time and cursor < len(idx) - 1:
                cursor += 1
        i = 1
        j = cursor
        d[side][gid] = {}
        while i < len(xs) - 1:
            wsum = 0.0
            w = 0.0
            while idx[j] < max_time:
                wsum += totals[j]
                w += amts[j]
                j += 1
            if w > 0.0:
                d[side][gid][xs[i]] = wsum / w
            else:
                d[side][gid][xs[i]] = np.nan
            i += 1
            min_time = max_time
            max_time = tm + xs[i]
        n += 1
    d['bid'] = sort_cols(pd.DataFrame.from_dict(d['bid'], orient='index'))
    d['ask'] = sort_cols(pd.DataFrame.from_dict(d['ask'], orient='index'))
    return d


# code used to perform `to_tframe` on all versions:

start = time.time()
tresults = {}
for version in version_counts.index:
    tresults[version] = to_tframe(version, df, trades, start)
print '{} finished'.format(time.time() - start)