"""
loadings plot;
shows, how flu on different freq-s affect PC;
axis: em, ex, loading; figure per PC; ',' and '.' goes to next PC;
lines should be smooth, so that flu on next freq-s affect PC in similar way;
but they are not in large areas; there is an error;
then we disable normalizing (/=std) data, and lines become smooth;
scaling can prevent spectroscopic interpretation of loadings;
we see peaks on em==ex; is it right, or we'd preprocess data in this area?
idea: may be we'd normalize not by var (freq pair), but by em or ex;
"""
normalize=0
from numpy import load, array, arange
from linre_tools import pcs_calc
from linre_explorer import freq3d_save
l = load('linre_big.npz')
max_components = 7
pco = pcs_calc(
exa=l['exa'], flum=l['flum'], max_components=max_components,
no_peaks=1, recalc_wo_outliers=1, normalize=normalize,
ret_loadings=1,
)
freq3d_save('out14/loadings',
ema=l['ema'], exa=l['exa'], data=pco.loadings, zlabel='Scaling coefficient',
pga=['L'+str(i+1) for i in range(max_components)]
)
"""
residuals plot by variables (ex-em);
"""
max_components = 9
from numpy import load, array, arange, power, sqrt
from linre_tools import pcs_calc
from linre_explorer import freq3d_save
l = load('linre_big.npz')
pco = pcs_calc(
exa=l['exa'], flum=l['flum'], max_components=max_components,
no_peaks=1, recalc_wo_outliers=1, normalize=0,
ret_X=1, ret_pca=1, ret_PC=1
)
def evar(X_fwd,X_back): return power(X_fwd-X_back,2).sum(axis=0) ## variance
X = pco.X
var0 = evar( X, X.mean(axis=0) ) ## mean in already 0
var = evar( X, pco.pca.inverse_transform(pco.PC) )
freq3d_save('out15/residuals',
ema=l['ema'], exa=l['exa'], data=sqrt(var/var0)[None], pga=['residuals']
) #sqrt(var/var0)
print pco.pca.explained_variance_ratio_
"""
pls: loadings explorer
loadings are like cutten, what about making EX even less?
higher component: more noise
"""
preprocess = 'pp_' #|''
from numpy import load, hstack
from linre_tools import AttrDict
from linre_explorer import freq3d_save
l = AttrDict(load('linre_big.npz'))
pls = AttrDict(load("out21/"+preprocess+"loadings.npz"))
#print pls.y_loadings, l.ema.shape, l.exa.shape, pls.x_loadings.shape
def annotate(pf,d): return map(lambda v:pf+str(v), range(d))
data = hstack((pls.x_loadings,pls.x_rotations)).T
pga = annotate('L ',pls.x_loadings.shape[1]) + annotate('R ',pls.x_rotations.shape[1])
print pga
freq3d_save("out22/"+preprocess+"loadings",ema=l.ema,exa=l.exa,data=data,pga=pga)
def show3d(preprocess):
coefs, x_mean, y_mean = load_pls(preprocess)
print y_mean
data = hstack((coefs, x_mean)).T
pga = ['coefs','x_mean']
freq3d_save("out27/a"+preprocess,ema=ema,exa=exa,data=data,pga=pga) #not final plot
from numpy import load
from linre_tools import AttrDict
from linre_explorer import freq3d_save
for fn in ['linre_big','linre_big2']:
l = AttrDict(load(fn+'.npz'))
freq3d_save(fn,
ema=l.ema, exa=l.exa, data=l.flum.T, pga=l.expa, zlabel='Fluorescence'
)
