def get_capacity(data_df, qtiles, time_fmt, tm_col, tps_col, rt_col, capacity_window, xexp=1.0, name=None):
"""
Daily capacity
:param data_df: DF with all data
:param qtiles: number of quantiles used in the capacity model
:param capacity_window: number of workdays used to compute the capacity.
:param xexp: exponent for generalized power function
:param time_fmt: str format of time col
:param tm_col: time column name
:param tps_col: throughput columns name
:param rt_col: response time column name
:param name: name of the system
:return:
"""
cap_df = pd.DataFrame(columns=['day', tps_col, rt_col, 'traffic', 'rsq_adj'])
data_df['day'] = data_df[tm_col].apply(lambda x: tm_ut.change_format(x, in_format=time_fmt, out_format='%Y-%m-%d'))
dates = data_df['day'].sort_values().unique()
row, last_result = 0, None
for date in dates:
start_dates = list(dates[dates <= date])
if len(start_dates) >= capacity_window:
start_date = start_dates[-capacity_window]
elif len(start_dates) > 0:
start_date = start_dates[0]
else:
start_date = date
df = data_df[(start_date <= data_df['day']) & (data_df['day'] <= date)].copy() # tm_col, users_col, tput_col, rtm_col, traffic
if len(df) > 0:
result = tpe_capacity.tput_engine_cap(df, tps_col, rt_col, qtiles, xexp=xexp, date=date, name=name)
cap_df.loc[row] = [date] + result
row += 1
cap_df.columns = ['day', tps_col + '_cap', rt_col + '_cap', 'traffic_cap', 'rsq_adj']
return cap_df
def df_check(a_df, x_col, r_col, n_col, u_col, agg=None):
a_df.reset_index(inplace=True, drop=True)
if x_col not in a_df.columns or r_col not in a_df.columns or n_col not in a_df.columns:
print 'invalid columns: ' + str(x_col) + ' ' + str(r_col) + ' ' + str(n_col) + ' ' + str(a_df.columns)
sys.exit(0)
for c in [x_col, r_col]:
if all(a_df[c].isnull()) is True:
print 'invalid values in ' + str(c)
sys.exit(0)
if any(a_df[n_col].isnull()) is True:
print 'invalid values in ' + str(n_col)
sys.exit(0)
# find time col and secs between rows
time_fmt, time_col = None, None
s = a_df.ix[a_df.index[0],]
for c in s.index:
time_fmt = tm_ut.get_date_format(s[c])
if time_fmt is not None:
time_col = c
break
time_secs = get_time_secs(a_df[time_col].copy(), time_fmt)
if time_fmt is None or time_secs is None:
print 'invalid time info'
print a_df.head()
sys.exit(0)
a_df[r_col] /= 1000.0 # all our response times are in msecs
a_df[x_col] /= time_secs # tput always in tps
a_df[u_col] /= time_secs # look at users per sec also, mainly to avoid huge values
# time aggregation
agg_val = None if agg not in ['hour', 'day'] else agg
if agg_val is None:
out_df = a_df[[n_col, time_col, x_col, r_col, u_col]].copy()
t_fmt = time_fmt
else: # hour or day aggregation
t_fmt = '%Y-%m-%d-%H' if agg_val == 'hour' else '%Y-%m-%d'
a_df[time_col] = a_df[time_col].apply(lambda x: tm_ut.change_format(x, in_format=time_fmt, out_format=t_fmt))
a_df['x*r'] = a_df[x_col] * a_df[r_col]
agg_df = a_df.groupby([time_col, n_col]).agg({
x_col: {x_col + '_avg': np.mean, x_col + '_sum': np.sum},
u_col: {u_col + '_avg': np.mean, u_col + '_sum': np.sum},
'x*r': np.sum
}).reset_index()
out_df = pd.DataFrame(columns=[n_col, time_col, x_col, r_col, u_col])
out_df[r_col] = agg_df[('x*r', 'sum')] / agg_df[(x_col, x_col + '_sum')] # response time is an weighted avg
out_df[x_col] = agg_df[(x_col, x_col + '_avg')]
out_df[u_col] = agg_df[(u_col, u_col + '_avg')]
out_df[time_col] = agg_df[time_col]
out_df[n_col] = agg_df[n_col]
out_df['time_col'] = time_col
out_df['time_fmt'] = t_fmt
return out_df
