comparison tools/multivariate_stats/pca.py @ 0:9071e359b9a3

Uploaded
author xuebing
date Fri, 09 Mar 2012 19:37:19 -0500
parents
children
comparison
equal deleted inserted replaced
-1:000000000000 0:9071e359b9a3
1 #!/usr/bin/env python
2
3 from galaxy import eggs
4 import sys, string
5 from rpy import *
6 import numpy
7
8 def stop_err(msg):
9 sys.stderr.write(msg)
10 sys.exit()
11
12 infile = sys.argv[1]
13 x_cols = sys.argv[2].split(',')
14 method = sys.argv[3]
15 outfile = sys.argv[4]
16 outfile2 = sys.argv[5]
17
18 if method == 'svd':
19 scale = center = "FALSE"
20 if sys.argv[6] == 'both':
21 scale = center = "TRUE"
22 elif sys.argv[6] == 'center':
23 center = "TRUE"
24 elif sys.argv[6] == 'scale':
25 scale = "TRUE"
26
27 fout = open(outfile,'w')
28 elems = []
29 for i, line in enumerate( file ( infile )):
30 line = line.rstrip('\r\n')
31 if len( line )>0 and not line.startswith( '#' ):
32 elems = line.split( '\t' )
33 break
34 if i == 30:
35 break # Hopefully we'll never get here...
36
37 if len( elems )<1:
38 stop_err( "The data in your input dataset is either missing or not formatted properly." )
39
40 x_vals = []
41
42 for k,col in enumerate(x_cols):
43 x_cols[k] = int(col)-1
44 x_vals.append([])
45
46 NA = 'NA'
47 skipped = 0
48 for ind,line in enumerate( file( infile )):
49 if line and not line.startswith( '#' ):
50 try:
51 fields = line.strip().split("\t")
52 valid_line = True
53 for k,col in enumerate(x_cols):
54 try:
55 xval = float(fields[col])
56 except:
57 skipped += 1
58 valid_line = False
59 break
60 if valid_line:
61 for k,col in enumerate(x_cols):
62 xval = float(fields[col])
63 x_vals[k].append(xval)
64 except:
65 skipped += 1
66
67 x_vals1 = numpy.asarray(x_vals).transpose()
68 dat= r.list(array(x_vals1))
69
70 set_default_mode(NO_CONVERSION)
71 try:
72 if method == "cor":
73 pc = r.princomp(r.na_exclude(dat), cor = r("TRUE"))
74 elif method == "cov":
75 pc = r.princomp(r.na_exclude(dat), cor = r("FALSE"))
76 elif method=="svd":
77 pc = r.prcomp(r.na_exclude(dat), center = r(center), scale = r(scale))
78 except RException, rex:
79 stop_err("Encountered error while performing PCA on the input data: %s" %(rex))
80
81 set_default_mode(BASIC_CONVERSION)
82 summary = r.summary(pc, loadings="TRUE")
83 ncomps = len(summary['sdev'])
84
85 if type(summary['sdev']) == type({}):
86 comps_unsorted = summary['sdev'].keys()
87 comps=[]
88 sd = summary['sdev'].values()
89 for i in range(ncomps):
90 sd[i] = summary['sdev'].values()[comps_unsorted.index('Comp.%s' %(i+1))]
91 comps.append('Comp.%s' %(i+1))
92 elif type(summary['sdev']) == type([]):
93 comps=[]
94 for i in range(ncomps):
95 comps.append('Comp.%s' %(i+1))
96 sd = summary['sdev']
97
98 print >>fout, "#Component\t%s" %("\t".join(["%s" % el for el in range(1,ncomps+1)]))
99 print >>fout, "#Std. deviation\t%s" %("\t".join(["%.4g" % el for el in sd]))
100 total_var = 0
101 vars = []
102 for s in sd:
103 var = s*s
104 total_var += var
105 vars.append(var)
106 for i,var in enumerate(vars):
107 vars[i] = vars[i]/total_var
108
109 print >>fout, "#Proportion of variance explained\t%s" %("\t".join(["%.4g" % el for el in vars]))
110
111 print >>fout, "#Loadings\t%s" %("\t".join(["%s" % el for el in range(1,ncomps+1)]))
112 xcolnames = ["c%d" %(el+1) for el in x_cols]
113 if 'loadings' in summary: #in case of princomp
114 loadings = 'loadings'
115 elif 'rotation' in summary: #in case of prcomp
116 loadings = 'rotation'
117 for i,val in enumerate(summary[loadings]):
118 print >>fout, "%s\t%s" %(xcolnames[i], "\t".join(["%.4g" % el for el in val]))
119
120 print >>fout, "#Scores\t%s" %("\t".join(["%s" % el for el in range(1,ncomps+1)]))
121 if 'scores' in summary: #in case of princomp
122 scores = 'scores'
123 elif 'x' in summary: #in case of prcomp
124 scores = 'x'
125 for obs,sc in enumerate(summary[scores]):
126 print >>fout, "%s\t%s" %(obs+1, "\t".join(["%.4g" % el for el in sc]))
127
128 r.pdf( outfile2, 8, 8 )
129 r.biplot(pc)
130 r.dev_off()