Projects/laydi
Projects
/
laydi
Archived
7
0
Fork 0
This commit is contained in:
Arnar Flatberg 2007-07-28 16:05:11 +00:00
parent 9a2e259209
commit 349cab3c51
4 changed files with 297 additions and 131 deletions

View File

@ -197,7 +197,7 @@ class PLS(Model):
Model.__init__(self, id, name)
self._options = PlsOptions()
def validation(self, amax, n_sets, cv_val_method):
def validation(self):
"""Returns rmsep for pls model.
"""
m, n = self.model['E0'].shape
@ -207,7 +207,7 @@ class PLS(Model):
val_engine = pls_val
if self._options['calc_cv']==True:
rmsep, aopt = val_engine(self.model['E0'], self.model['F0'],
amax, n_sets)
self._options['amax'], self._options['n_sets'])
self.model['rmsep'] = rmsep[:,:-1]
self.model['aopt'] = aopt
else:
@ -319,7 +319,7 @@ class PLS(Model):
self.model['E0'] = self._data['X']
self.model['F0'] = self._data['Y']
self.validation(**options.validation_options())
self.validation()
self.make_model(self.model['E0'], self.model['F0'],
**options.make_model_options())
# variance captured

View File

@ -6,81 +6,189 @@ There is almost no typechecking of any kind here, just focus on speed
import math
from scipy.linalg import svd,inv
from scipy import dot,empty,eye,newaxis,zeros,sqrt,diag,\
apply_along_axis,mean,ones,randn,empty_like,outer,c_,\
rand,sum,cumsum,matrix
apply_along_axis,mean,ones,randn,empty_like,outer,r_,c_,\
rand,sum,cumsum,matrix, expand_dims,minimum,where
has_sym=True
try:
import symmeig
from symeig import symeig
except:
has_sym = False
has_sym=False
def pca(a, aopt,scale='scores',mode='normal',center_axis=-1):
""" Principal Component Analysis.
Performs PCA on given matrix and returns results in a dictionary.
:Parameters:
a : array
Data measurement matrix, (samples x variables)
aopt : int
Number of components to use, aopt<=min(samples, variables)
:Returns:
results : dict
keys -- values, T -- scores, P -- loadings, E -- residuals,
lev --leverages, ssq -- sum of squares, expvar -- cumulative
explained variance, aopt -- number of components used
:OtherParameters:
mode : str
Amount of info retained, ('fast', 'normal', 'detailed')
center_axis : int
Center along given axis. If neg.: no centering (-inf,..., matrix modes)
:SeeAlso:
- pcr : other blm
- pls : other blm
- lpls : other blm
Notes
-----
Uses kernel speed-up if m>>n or m<<n.
If residuals turn rank deficient, a lower number of component than given
in input will be used. The number of components used is given in results-dict.
Examples
--------
>>> import scipy,engines
>>> a=scipy.asarray([[1,2,3],[2,4,5]])
>>> dat=engines.pca(a, 2)
>>> dat['expvar']
array([0.,99.8561562, 100.])
def pca(a, aopt, scale='scores', mode='normal'):
""" Principal Component Analysis model
mode:
-- fast : returns smallest dim scaled (T for n<=m, P for n>m )
-- normal : returns all model params and residuals after aopt comp
-- detailed : returns all model params and all residuals
"""
if center_axis>=0:
a = a - expand_dims(a.mean(center_axis), center_axis)
m, n = a.shape
#print "rows: %s cols: %s" %(m,n)
if m>(n+100) or n>(m+100):
u, s, v = esvd(a)
u, e, v = esvd(a)
s = sqrt(e)
else:
u, s, vt = svd(a, 0)
v = vt.T
eigvals = (1./m)*s
e = s**2
tol = 1e-10
eff_rank = sum(s>s[0]*tol)
aopt = minimum(aopt, eff_rank)
T = u*s
s = s[:aopt]
e = e[:aopt]
T = T[:,:aopt]
P = v[:,:aopt]
if scale=='loads':
tnorm = apply_along_axis(vnorm, 0, T)
T = T/tnorm
P = P*tnorm
T = T/s
P = P*s
if mode == 'fast':
return {'T':T, 'P':P}
return {'T':T, 'P':P, 'aopt':aopt}
if mode=='detailed':
"""Detailed mode returns residual matrix for all comp.
That is E, is a three-mode matrix: (amax, m, n) """
E = empty((aopt, m, n))
E = empty((aopt, m, n))
ssq = []
lev = []
expvarx = empty((aopt, aopt+1))
for ai in range(aopt):
e = a - dot(T[:,:ai+1], P[:,:ai+1].T)
E[ai,:,:] = e.copy()
E[ai,:,:] = a - dot(T[:,:ai+1], P[:,:ai+1].T)
ssq.append([(E[ai,:,:]**2).sum(0), (E[ai,:,:]**2).sum(1)])
if scale=='loads':
lev.append([((s*T)**2).sum(1), (P**2).sum(1)])
else:
lev.append([(T**2).sum(1), ((s*P)**2).sum(1)])
expvarx[ai,:] = r_[0, 100*e.cumsum()/e.sum()]
else:
E = a - dot(T,P.T)
# residuals
E = a - dot(T, P.T)
SEP = E**2
ssq = [SEP.sum(0), SEP.sum(1)]
# leverages
if scale=='loads':
lev = [(1./m)+(T**2).sum(1), (1./n)+((P/s)**2).sum(1)]
else:
lev = [(1./m)+((T/s)**2).sum(1), (1./n)+(P**2).sum(1)]
# variances
expvarx = r_[0, 100*e.cumsum()/e.sum()]
return {'T':T, 'P':P, 'E':E}
return {'T':T, 'P':P, 'E':E, 'expvarx':expvarx, 'levx':lev, 'ssqx':ssq, 'aopt':aopt}
def pcr(a, b, aopt, scale='scores',mode='normal',center_axis=0):
""" Principal Component Regression.
Performs PCR on given matrix and returns results in a dictionary.
:Parameters:
a : array
Data measurement matrix, (samples x variables)
b : array
Data response matrix, (samples x responses)
aopt : int
Number of components to use, aopt<=min(samples, variables)
:Returns:
results : dict
keys -- values, T -- scores, P -- loadings, E -- residuals,
levx -- leverages, ssqx -- sum of squares, expvarx -- cumulative
explained variance, aopt -- number of components used
:OtherParameters:
mode : str
Amount of info retained, ('fast', 'normal', 'detailed')
center_axis : int
Center along given axis. If neg.: no centering (-inf,..., matrix modes)
:SeeAlso:
- pcr : other blm
- pls : other blm
- lpls : other blm
Notes
-----
Uses kernel speed-up if m>>n or m<<n.
If residuals turn rank deficient, a lower number of component than given
in input will be used. The number of components used is given in results-dict.
def pcr(a, b, aopt=2, scale='scores', mode='normal'):
"""Principal Component Regression.
Examples
--------
Returns
>>> import scipy,engines
>>> a=scipy.asarray([[1,2,3],[2,4,5]])
>>> dat=engines.pca(a, 2)
>>> dat['expvar']
array([0.,99.8561562, 100.])
"""
m, n = m_shape(a)
B = empty((aopt, n, l))
dat = pca(a, aopt=aopt, scale=scale, mode='normal', center_axis=0)
k, l = m_shape(b)
if center_axis>=0:
b = b - expand_dims(b.mean(center_axis), center_axis)
dat = pca(a, aopt=aopt, scale=scale, mode=mode, center_axis=center_axis)
T = dat['T']
weigths = apply_along_axis(vnorm, 0, T)
weights = apply_along_axis(vnorm, 0, T)
if scale=='loads':
# fixme: check weights
Q = dot(b.T, T*weights)
Q = dot(b.T, T*weights**2)
else:
Q = dot(b.T, T/weights**2)
if mode=='fast':
return {'T', T:, 'P':P, 'Q':Q}
dat.update({'Q':Q})
return dat
if mode=='detailed':
for i in range(1, aopt+1, 1):
F[i,:,:] = b - dot(T[:,i],Q[:,:i].T)
F = empty((aopt, k, l))
for i in range(aopt):
F[i,:,:] = b - dot(T[:,:i+1], Q[:,:i+1].T)
else:
F = b - dot(T, Q.T)
#fixme: explained variance in Y + Y-var leverages
dat.update({'Q',Q, 'F':F})
dat.update({'Q':Q, 'F':F})
return dat
def pls(a, b, aopt=2, scale='scores', mode='normal', ab=None):
@ -271,7 +379,6 @@ def nipals_lpls(X, Y, Z, a_max, alpha=.7, mean_ctr=[2, 0, 1], mode='normal', sca
X, mnX = center(X, xctr)
Y, mnY = center(Y, xctr)
Z, mnZ = center(Z, zctr)
print Z.mean(1)
varX = pow(X, 2).sum()
varY = pow(Y, 2).sum()
@ -365,7 +472,7 @@ def nipals_lpls(X, Y, Z, a_max, alpha=.7, mean_ctr=[2, 0, 1], mode='normal', sca
def m_shape(array):
return matrix(array).shape
def esvd(data):
def esvd(data, amax=None):
"""SVD with the option of economy sized calculation
Calculate subspaces of X'X or XX' depending on the shape
of the matrix.
@ -378,17 +485,30 @@ def esvd(data):
m, n = data.shape
if m>=n:
kernel = dot(data.T, data)
u, s, vt = svd(kernel)
u = dot(data, vt.T)
v = vt.T
if has_sym:
if not amax:
amax = n
pcrange = [n-amax, n]
s, v = symeig(kernel, range=pcrange, overwrite=True)
s = s[::-1]
v = v[:,arange(n, -1, -1)]
else:
u, s, vt = svd(kernel)
v = vt.T
u = dot(data, v)
for i in xrange(n):
s[i] = vnorm(u[:,i])
u[:,i] = u[:,i]/s[i]
else:
kernel = dot(data, data.T)
#data = (data + data.T)/2.0
u, s, vt = svd(kernel)
v = dot(u.T, data)
if has_sym:
if not amax:
amax = m
pcrange = [m-amax, m]
s, u = symeig(kernel, range=pcrange, overwrite=True)
else:
u, s, vt = svd(kernel)
v = dot(u.T, data)
for i in xrange(m):
s[i] = vnorm(v[i,:])
v[i,:] = v[i,:]/s[i]

View File

@ -3,32 +3,14 @@ import scipy
import rpy
silent_eval = rpy.with_mode(rpy.NO_CONVERSION, rpy.r)
def get_term_sim(termlist, method = "JiangConrath", verbose=False):
"""Returns the similariy matrix between go-terms.
Arguments:
termlist: character vector of GO terms
method: one of
("JiangConrath","Resnik","Lin","CoutoEnriched","CoutoJiangConrath","CoutoResnik","CoutoLin")
verbose: print out various information or not
"""
_methods = ("JiangConrath","Resnik","Lin","CoutoEnriched","CoutoJiangConrath","CoutoResnik","CoutoLin")
assert(method in _methods)
assert(termlist[0][:2]=='GO')
rpy.r.library("GOSim")
return rpy.r.getTermSim(termlist, method = method, verbose = verbose)
def get_gene_sim(genelist, similarity='OA',
distance="Resnick"):
rpy.r.library("GOSim")
rpy.r.assign("ids", genelist)
silent_eval('a<-getGeneSim(ids)', verbose=FALSE)
def goterms_from_gene(genelist, ontology=['BP'], garbage = ['IEA', 'ISS', 'ND']):
def goterms_from_gene(genelist, ontology='BP', garbage=None):
""" Returns the go-terms from a specified genelist (Entrez id).
Recalculates the information content if needed based on selected evidence codes.
"""
rpy.r.library("GO")
rpy.r.library("GOSim")
_CODES = {"IMP" : "inferred from mutant phenotype",
"IGI" : "inferred from genetic interaction",
"IPI" :"inferred from physical interaction",
@ -42,25 +24,46 @@ def goterms_from_gene(genelist, ontology=['BP'], garbage = ['IEA', 'ISS', 'ND'])
"IC" : "inferred by curator"
}
_ONTOLOGIES = ['BP', 'CC', 'MF']
assert(scipy.all([(code in _CODES) for code in garbage]))
assert(scipy.all([(ont in _ONTOLOGIES) for ont in ontology]))
have_these = rpy.r('as.list(GOTERM)').keys()
goterms = {}
#assert(scipy.all([(code in _CODES) for code in garbage]) or garbage==None)
assert(ontology in _ONTOLOGIES)
dummy = rpy.r.setOntology(ontology)
ddef = False
if ontology=='BP' and garbage!=None:
# This is for ont=BP and garbage =['IEA', 'ISS', 'ND']
rpy.r.load("ICsBPIMP_IGI_IPI_ISS_IDA_IEP_TAS_NAS_IC.rda")
ic = rpy.r.assign("IC",rpy.r.IC, envir=rpy.r.GOSimEnv)
print len(ic)
else:
ic = rpy.r('get("IC", envir=GOSimEnv)')
print "loading GO definitions environment"
gene2terms = {}
for gene in genelist:
goterms[gene] = []
info = rpy.r('GOENTREZID2GO[["' + str(gene) + '"]]')
#print info
if info:
skip=False
for term, desc in info.items():
if term not in have_these:
print "GO miss:"
print term
if desc['Ontology'] in ontology and desc['Evidence'] not in garbage:
goterms[gene].append(term)
if ic.get(term)==scipy.isinf:
print "\nIC is Inf on this GO term %s for this gene: %s" %(term,gene)
skip=True
if ic.get(term)==None:
#print "\nHave no IC on this GO term %s for this gene: %s" %(term,gene)
skip=True
if desc['Ontology']!=ontology:
#print "\nThis GO term %s belongs to: %s:" %(term,desc['Ontology'])
skip = True
if not skip:
if gene2terms.has_key(gene):
gene2terms[gene].append(term)
else:
gene2terms[gene] = [term]
else:
print "\nHave no Annotation on this gene: %s" %gene
return goterms
return gene2terms
def genego_matrix(goterms, tmat, gene_ids, term_ids, func=min):
def genego_matrix(goterms, tmat, gene_ids, term_ids, func=max):
ngenes = len(gene_ids)
nterms = len(term_ids)
gene2indx = {}
@ -71,23 +74,46 @@ def genego_matrix(goterms, tmat, gene_ids, term_ids, func=min):
term2indx[id]=i
#G = scipy.empty((nterms, ngenes),'d')
G = []
newindex = []
new_gene_index = []
for gene, terms in goterms.items():
g_ind = gene2indx[gene]
if len(terms)>0:
t_ind = []
newindex.append(g_ind)
new_gene_index.append(g_ind)
for term in terms:
if term2indx.has_key(term): t_ind.append(term2indx[term])
print t_ind
subsim = tmat[t_ind, :]
gene_vec = scipy.apply_along_axis(func, 0, subsim)
G.append(gene_vec)
return scipy.asarray(G), newindex
return scipy.asarray(G), new_gene_index
def genego_sim(gene2go, gene_ids, all_go_terms, STerm, go_term_sim="OA", term_sim="Lin", verbose=False):
"""Returns go-terms x genes similarity matrix.
:input:
- gene2go: dict: keys: gene_id, values: go_terms
- gene_ids: list of gene ids (entrez ids)
- STerm: (go_terms x go_terms) similarity matrix
- go_terms_sim: similarity measure between a gene and multiple go terms (max, mean, OA)
- term_sim: similarity measure between two go-terms
- verbose
"""
rpy.r.library("GOSim")
#gene_ids = gene2go.keys()
GG = scipy.empty((len(all_go_terms), len(gene_ids)), 'd')
for j,gene in enumerate(gene_ids):
for i,go_term in enumerate(all_go_terms):
if verbose:
print "\nAssigning similarity from %s to terms(gene): %s" %(go_term,gene)
GG_ij = rpy.r.getGSim(go_term, gene2go[gene], similarity=go_term_sim,
similarityTerm=term_sim, STerm=STerm, verbose=verbose)
GG[i,j] = GG_ij
return GG
def goterm2desc(gotermlist):
"""Returns the go-terms description keyed by go-term
"""Returns the go-terms description keyed by go-term.
"""
rpy.r.library("GO")
term2desc = {}

View File

@ -23,7 +23,7 @@ data = DX.asarray().T
rpy.r.assign("data", data)
cl = dot(DY.asarray(), diag([1,2,3])).sum(1)
rpy.r.assign("cl", cl)
rpy.r.assign("B", 100)
rpy.r.assign("B", 20)
# Perform a SAM analysis.
print "Starting SAM"
sam = rpy.r('sam.out<-sam(data=data,cl=cl,B=B,rand=123)')
@ -32,63 +32,74 @@ print "SAM done"
qq = rpy.r('qobj<-qvalue(sam.out@p.value)')
qvals = asarray(qq['qvalues'])
# cut off
co = 0.001
index = where(qvals<0.01)[0]
cutoff = 2
index = where(qvals<cutoff)[0]
# Subset data
X = DX.asarray()
Xr = X[:,index]
gene_ids = DX.get_identifiers('gene_ids', index)
print "\nWorkiing on subset with %s genes " %len(gene_ids)
### Build GO data ####
print "\nWorking on subset with %s genes " %len(gene_ids)
#gene2ind = {}
#for i, gene in enumerate(gene_ids):
# gene2ind[gene] = i
print "Go terms ..."
goterms = rpy_go.goterms_from_gene(gene_ids)
terms = set()
for t in goterms.values():
terms.update(t)
terms = list(terms)
print "Number of go-terms: %s" %len(terms)
### Build GO data ####
print "\n\nFiltering genes by Go terms "
gene2goterms = rpy_go.goterms_from_gene(gene_ids)
all_terms = set()
for t in gene2goterms.values():
all_terms.update(t)
terms = list(all_terms)
print "\nNumber of go-terms: %s" %len(terms)
# update genelist
gene_ids = gene2goterms.keys()
print "\nNumber of genes: %s" %len(gene_ids)
rpy.r.library("GOSim")
# Go-term similarity matrix
methods = ("JiangConrath","Resnik","Lin","CoutoEnriched","CoutoJiangConrath","CoutoResnik","CoutoLin")
meth = methods[0]
meth = methods[3]
print "Term-term similarity matrix (method = %s)" %meth
if meth=="CoutoEnriched":
rpy.r('setEnrichmentFactors(alpha=0.1,beta=0.5)')
print "Calculating term-term similarity matrix"
tmat = rpy.r.getTermSim(terms, verbose=False, method=meth)
print "\nCalculating term-term similarity matrix"
rpytmat1 = rpy.with_mode(rpy.NO_CONVERSION, rpy.r.getTermSim)(terms, method=meth,verbose=False)
tmat1 = rpy.r.assign("haha", rpytmat1)
# check if all terms where found
nanindex = where(isnan(tmat1[:,0]))[0]
if len(nanindex)>0:
raise valueError("NANs in tmat")
# Z-matrix
#Z, newind = rpy_go.genego_matrix(terms, tmat, gene_ids, terms,func=mean)
#Z = Z.T
Z1 = rpy_go.genego_sim(gene2goterms,gene_ids,terms,rpytmat1,go_term_sim="OA",term_sim=meth)
#### do another
meth = methods[4]
rpytmat = rpy.with_mode(rpy.NO_CONVERSION, rpy.r.getTermSim)(terms, method=meth,verbose=False)
tmat = rpy.r.assign("haha", rpytmat)
# check if all terms where found
nanindex = where(isnan(tmat[:,0]))[0]
keep=[]
has_miss = False
if len(nanindex)>0:
has_miss = True
print "Some terms missing in similarity matrix"
keep = where(isnan(tmat[:,0])!=True)[0]
print "Number of nans: %d" %len(nanindex)
tmat_new = tmat[:,keep][keep,:]
new_terms = [i for ind,i in enumerate(terms) if ind in keep]
bad_terms = [i for ind,i in enumerate(terms) if ind not in keep]
# update go-term dict
for gene,trm in goterms.items():
for t in trm:
if t in bad_terms:
trm.remove(t)
if len(trm)==0:
print "Removing gene: %s" %gene
goterms[gene]=trm
terms = new_terms
tmat = tmat_new
raise valueError("NANs in tmat")
# Z-matrix
# func (min, max, median, mean, etc),
# func decides on the representation of gene-> goterm when multiple
# goterms exist for one gene
Z, newind = rpy_go.genego_matrix(goterms, tmat, gene_ids, terms,func=mean)
Z = Z.T
# update X matrix (no go-terms available)
Xr = Xr[:,newind]
new_gene_ids = asarray(gene_ids)[newind]
#Z, newind = rpy_go.genego_matrix(terms, tmat, gene_ids, terms,func=mean)
#Z = Z.T
Z = rpy_go.genego_sim(gene2goterms,gene_ids,terms,rpytmat,go_term_sim="OA",term_sim=meth)
# update data (X) matrix
#newind = [gene2ind[gene] for gene in gene_ids]
newind = DX.get_indices('gene_ids', gene_ids)
Xr = X[:,newind]
#new_gene_ids = asarray(gene_ids)[newind]
######## LPLSR ########
@ -112,11 +123,14 @@ if alpha_check:
rmsep , yhat, ce = cv_lpls(Xr, Y, Z, a_max, alpha=alpha)
Rmsep.append(rmsep)
Yhat.append(yhat)
CE.append(yhat)
CE.append(ce)
Rmsep = asarray(Rmsep)
Yhat = asarray(Yhat)
CE = asarray(CE)
figure(200)
# Significance Hotellings T
Wx, Wz, Wy, = jk_lpls(Xr, Y, Z, aopt)
@ -135,7 +149,13 @@ for a in range(m):
ylim([rmsep.min()-.05, rmsep.max()+.05])
title('RMSEP')
figure(2) # Hypoid correlations
figure(2)
for a in range(m):
bar(arange(a_max)+a*bar_w+.1, class_error[:,a], width=bar_w, color=bar_col[a])
ylim([class_error.min()-.05, class_error.max()+.05])
title('Classification accuracy')
figure(3) # Hypoid correlations
plot_corrloads(Rz, pc1=0, pc2=1, s=tsqz/10.0, c='b', zorder=5, expvar=evz, ax=None)
ax = gca()
ylabels = DY.get_identifiers('_cat', sorted=True)