In [2]:

import numpy as np
from fmskill import PointObservation, TrackObservation
from fmskill import ModelResult, Connector

In [6]:

fn = '../tests/testdata/SW/HKZN_local_2017_DutchCoast.dfsu'
mr = ModelResult(fn, name='HKZN_local', item=0)
mr.dfs

Out[6]:

Dfsu2D
Number of elements: 958
Number of nodes: 570
Projection: LONG/LAT
Number of items: 15
Time: 23 steps with dt=10800.0s
      2017-10-27 00:00:00 -- 2017-10-29 18:00:00

Configuration of comparison, see SW_DutchCoast.ipynb for more details.

In [7]:

o1 = PointObservation('../tests/testdata/SW/HKNA_Hm0.dfs0', item=0, x=4.2420, y=52.6887, name="HKNA")
o2 = PointObservation("../tests/testdata/SW/eur_Hm0.dfs0", item=0, x=3.2760, y=51.9990, name="EPL")
o3 = TrackObservation("../tests/testdata/SW/Alti_c2_Dutch.dfs0", item=3, name="c2")
con = Connector([o1, o2, o3], mr)
cc = con.extract()
cc

Standard set of metrics

In [13]:

cc.skill().style(precision=2)

Out[13]:

	n	bias	rmse	urmse	mae	cc	si	r2
observation
EPL	67	-0.07	0.22	0.21	0.19	0.97	0.08	0.93
HKNA	386	-0.19	0.35	0.29	0.25	0.97	0.09	0.91
c2	113	-0.00	0.35	0.35	0.29	0.97	0.12	0.90

Select a specific metric

In [38]:

cc.skill(metrics="mean_absolute_error")

Out[38]:

	n	mean_absolute_error
observation
EPL	67	0.188513
HKNA	386	0.251839
c2	113	0.294585

Some metrics has parameters, which require a bit special treatment.

In [39]:

from fmskill.metrics import hit_ratio

def hit_ratio_05_pct(obs, model):
    return hit_ratio(obs, model, 0.5) * 100

def hit_ratio_01_pct(obs, model):
    return hit_ratio(obs, model, 0.1) * 100

cc.skill(metrics=[hit_ratio_05_pct, hit_ratio_01_pct]).style(precision=0)

Out[39]:

	n	hit_ratio_05_pct	hit_ratio_01_pct
observation
EPL	67	99	27
HKNA	386	87	30
c2	113	86	17

It is also possible to do it in a single line using a lambda function (anonymous function), with the downside that the anonymous function has no name, and thus no automatic column label :-(

In [37]:

cc.skill(metrics=lambda obs, model : hit_ratio(obs, model, 0.1))

Out[37]:

	n	<lambda>
observation
EPL	67	0.268657
HKNA	386	0.295337
c2	113	0.168142

And you are of course always free to specify your own special metric.

In [33]:

def my_special_metric(obs, model):

    res = obs - model

    res_clipped = np.clip(res,0,np.inf)

    return np.mean(np.abs(res_clipped))

cc.skill(metrics=my_special_metric)

Out[33]:

	n	my_special_metric
observation
EPL	67	0.127555
HKNA	386	0.223049
c2	113	0.147897

In [ ]: