def test_RMSE_offset_neg(self):
'''Test that RMSE is correct for constant prediction offset from constant observation (with negatives)'''
val, diff = -5, -2
n_elements = 12
predvec = [val] * n_elements
obsvec = [val - diff] * n_elements
self.assertEqual(verify.RMSE(predvec, obsvec), np.abs(diff))
def test_RMSE_offset(self):
'''Test that RMSE is correct for constant prediction offset from constant observation '''
val, diff = 5, 2
n_elements = 12
predvec = [val] * n_elements
obsvec = [val - diff] * n_elements
self.assertEqual(verify.RMSE(predvec, obsvec), diff)
def run_for_TSS(model, trw):
fdummy = "out/det.dummy.pred.%d.csv" % (trw)
fname = "out/det.%s.pred.%d.csv" % (model, trw)
_od = pd.read_csv(fdummy)
_o = pd.read_csv(fname)
_od = _od[(_od.prob_clsf != -1.) & (_od.y_pred != -1.) & (_od.y_pred >= 0)
& (_od.y_pred <= 9.)]
_o = _o[(_o.prob_clsf != -1.) & (_o.y_pred != -1.) & (_o.y_pred >= 0) &
(_o.y_pred <= 9.)]
_od.dn = pd.to_datetime(_od.dn)
_o.dn = pd.to_datetime(_o.dn)
stime = dt.datetime(1995, 2, 1)
etime = dt.datetime(2016, 9, 20)
d = stime
skill = []
t = []
while (d < etime):
try:
t.append(d)
dn = d + dt.timedelta(days=27)
dum = _od[(_od.dn >= d) & (_od.dn < dn)]
mod = _o[(_o.dn >= d) & (_o.dn < dn)]
rmse_dum = verify.RMSE(dum.y_pred, dum.y_obs)
rmse = verify.RMSE(mod.y_pred, mod.y_obs)
print(d, rmse, rmse_dum, verify.skill(rmse, rmse_dum))
skill.append(verify.skill(rmse, rmse_dum))
d = d + dt.timedelta(days=1)
except:
pass
pass
skill = np.array(skill)
#skill = nan_helper(skill)
fmt = matplotlib.dates.DateFormatter("%d %b\n%Y")
splot.style("spacepy")
fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(10, 6))
ax.xaxis.set_major_formatter(fmt)
ax.plot(t, skill, "k.", label="")
#ax.plot(t,smooth(np.array(skill),101),"r.")
#strx = "RMSE:%.2f\nr:%.2f"%(_eval_details["RMSE"],_eval_details["r"])
#ax.text(0.2,0.8,strx,horizontalalignment='center',verticalalignment='center', transform=ax.transAxes)
ax.set_ylabel(r"$TSS(\%)$")
ax.set_xlabel(r"$Time$")
ax.set_xlim(dt.datetime(1995, 1, 1), dt.datetime(2017, 1, 1))
ax.set_ylim(0, 100)
fig.savefig("out/stat/det.%s.tss.%d.png" % (model, trw))
def run_validation(pred, obs, year, model):
pred, obs = np.array(pred), np.array(obs)
_eval_details = {}
_eval_details["range"] = "N"
if max(pred) > 9. or min(pred) < 0.: _eval_details["range"] = "Y"
try:
_eval_details["bias"] = np.round(verify.bias(pred, obs), 2)
except:
_eval_details["bias"] = np.NaN
try:
_eval_details["meanPercentageError"] = np.round(
verify.meanPercentageError(pred, obs), 2)
except:
_eval_details["meanPercentageError"] = np.NaN
try:
_eval_details["medianLogAccuracy"] = np.round(
verify.medianLogAccuracy(pred, obs), 3)
except:
_eval_details["medianLogAccuracy"] = np.NaN
try:
_eval_details["symmetricSignedBias"] = np.round(
verify.symmetricSignedBias(pred, obs), 3)
except:
_eval_details["symmetricSignedBias"] = np.NaN
try:
_eval_details["meanSquaredError"] = np.round(
verify.meanSquaredError(pred, obs), 2)
except:
_eval_details["meanSquaredError"] = np.NaN
try:
_eval_details["RMSE"] = np.round(verify.RMSE(pred, obs), 2)
except:
_eval_details["RMSE"] = np.NaN
try:
_eval_details["meanAbsError"] = np.round(
verify.meanAbsError(pred, obs), 2)
except:
_eval_details["meanAbsError"] = np.NaN
try:
_eval_details["medAbsError"] = np.round(verify.medAbsError(pred, obs),
2)
except:
_eval_details["medAbsError"] = np.NaN
try:
_eval_details["nRMSE"] = np.round(verify.nRMSE(pred, obs), 2)
except:
_eval_details["nRMSE"] = np.NaN
try:
_eval_details["forecastError"] = np.round(
np.mean(verify.forecastError(pred, obs)), 2)
except:
_eval_details["forecastError"] = np.NaN
try:
_eval_details["logAccuracy"] = np.round(
np.mean(verify.logAccuracy(pred, obs)), 2)
except:
_eval_details["logAccuracy"] = np.NaN
try:
_eval_details["medSymAccuracy"] = np.round(
verify.medSymAccuracy(pred, obs), 2)
except:
_eval_details["medSymAccuracy"] = np.NaN
try:
_eval_details["meanAPE"] = np.round(verify.meanAPE(pred, obs), 2)
except:
_eval_details["meanAPE"] = np.NaN
try:
_eval_details["medAbsDev"] = np.round(verify.medAbsDev(pred), 2)
except:
_eval_details["medAbsDev"] = np.NaN
try:
_eval_details["rSD"] = np.round(verify.rSD(pred), 2)
except:
_eval_details["rSD"] = np.NaN
try:
_eval_details["rCV"] = np.round(verify.rCV(pred), 2)
except:
_eval_details["rCV"] = np.NaN
_eval_details["year"] = year
_eval_details["model"] = model
r, _ = pearsonr(pred, obs)
_eval_details["r"] = r
return _eval_details
def test_RMSE_alt0(self):
'''Test that RMSE is correct for prediction alternating evenly about observation '''
self.assertEqual(verify.RMSE(self.predalt, self.obsalt), 2)
def test_RMSE(self):
'''Test estimation of RMSE for perfect forecast'''
self.assertEqual(verify.RMSE(self.predvec, self.obsvec), 0)
