y = tau[:,0]
#y = np.log(y/1000. + 10.)
#y, _ = stats.boxcox(y)
#for i in y:
# if i in ydel:
# y[i] = np.nan
notNaNs = ~np.any(np.isnan(x),1) & ~np.isnan(y)
X = x[notNaNs,:]
y = y[notNaNs]
X= sm.add_constant(X)
model = sm.OLS(y, X).fit()
print model.summary()
yhat = model.predict()
print "R2 is: %.3f, R2adj is: %.3f" %(mysm.cal_R2(y,yhat,n=model.nobs,p=model.df_model))
print "rmse is %.3f"%(mysm.cal_RMSE(y, yhat))
if plott == 1:
fig, axes = plt.subplots(nrows=1,ncols=1)
#plt.hist(y,30,alpha=0.3, normed=True)
plt.scatter(X[:,1],y)
if plotres == 1:
fig = plt.figure()
ax = fig.add_axes([0.05, 0.05, 0.9, 0.9])
ax.scatter(model.predict(), model.resid)
ax.set_ylabel('residual')
ax.set_xlabel('yhat')
if plot_y_yhat:
fig = plt.figure()
ax = fig.add_axes([0.05, 0.05, 0.9, 0.9])
ax.scatter(y, yhat)
ax.set_ylabel('yhat')
profid = data.index.unique() # index of profile start
d14C = prep.getvarxls(data,'D14C_BulkLayer', profid, ':')
sampleyr = prep.getvarxls(data, 'SampleYear', profid, ':')
layerbot = prep.getvarxls(data, 'Layer_bottom', profid, ':')
tau, cost = C14.cal_tau(d14C, sampleyr, 3, False)
np.savez('./Synthesis_allD14C_tau.npz',tau=tau,cost=cost)
taudata = np.load('./Synthesis_allD14C_tau.npz')
tau = taudata['tau']
cost = taudata['cost']
D14C2000 = np.array(C14.cal_D14Ctosmpyr(tau[:,0], 2000))
is_badcost = cost[:,0]>50
data.D14C_BulkLayer[is_badcost]
a = mysm.cal_RMSE(d14C[~is_badcost], D14C2000[~is_badcost])
D14C2000df = pd.DataFrame(data=D14C2000)
D14C2000df.to_csv('normalizedD14C.csv')
#%% verify the D14C normalization approach
newdata = data.copy()
# index of profiles that have multiple year measurements
def print_normalized(profid, tosmpyr):
prof1 = data.loc[profid,['Layer_bottom','D14C_BulkLayer','SampleYear']]
mod = C14.cal_D14Ctosmpyr(tau[:,0], tosmpyr)
newdata['D14C_normalized'] = mod
prof = newdata.loc[profid,['Layer_bottom','D14C_BulkLayer','D14C_normalized','SampleYear']]
print prof
print_normalized(1, 2013)
ax.set_title('profile ID:' + str(pro) + '\n' + bio)
plt.legend(loc=4)
plt.gca().invert_yaxis()
#%% calculate rmse
rmse = []
for idd in range(len(profid4modeling)):
pro = profid4modeling[idd]
biome = {1:'Boreal Forest',2:'Temperate Forest',3:'Tropical Forest',4:'Grassland', \
5:'Cropland',6:'Shrublands',7:'Peatland',8:'Savannas'}
print 'profid is ',pro
f_i = interp1d(data.loc[pro:pro,'Layer_bottom'].values, obss[idd])
f_x = prep.extrap1d(f_i)
y = f_x(depthh[idd])
y = np.reshape(y,(y.shape[0],1))
yhat = out[idd]
rmse.append(myst.cal_RMSE(y, yhat))
#%% use jobaggy soc to extrapolate missing soc profiles,
# log-fitting interpolation, using cum_pctC, write to extrasitegridid.txt
from scipy.optimize import curve_fit
def func(x, K, I):
return np.exp(K*np.log(x)+I)
out = []
obss = []
depthh = []
outf = open('extrasitegridid.txt','w')
for i in extraprofid:
print 'profile is :',i
jobgypctC = np.array(csvbiome[data.loc[i:i,'VegTypeCode_Local'].values[0]])/100.
popt, pcov = curve_fit(func, np.r_[0,jobgydepth], np.r_[0,np.cumsum(jobgypctC)])