Mercurial > repos > bgruening > upload_testing

comparison logistic_regression_vif.py @ 80:c4a3a8999945 draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Uploaded
author bernhardlutz
date Mon, 20 Jan 2014 14:39:43 -0500
parents
children
comparison
equal deleted inserted replaced
79:dc82017052ac 80:c4a3a8999945
1 #!/usr/bin/env python
2
3 #from galaxy import eggs
4 import sys, string
5 #from rpy import *
6 import rpy2.robjects as robjects
7 import rpy2.rlike.container as rlc
8 import rpy2.rinterface as ri
9 r = robjects.r
10 import numpy
11
12 def stop_err(msg):
13 sys.stderr.write(msg)
14 sys.exit()
15
16 infile = sys.argv[1]
17 y_col = int(sys.argv[2])-1
18 x_cols = sys.argv[3].split(',')
19 outfile = sys.argv[4]
20
21
22 print "Predictor columns: %s; Response column: %d" %(x_cols,y_col+1)
23 fout = open(outfile,'w')
24 elems = []
25 for i, line in enumerate( file ( infile )):
26 line = line.rstrip('\r\n')
27 if len( line )>0 and not line.startswith( '#' ):
28 elems = line.split( '\t' )
29 break
30 if i == 30:
31 break # Hopefully we'll never get here...
32
33 if len( elems )<1:
34 stop_err( "The data in your input dataset is either missing or not formatted properly." )
35
36 y_vals = []
37 x_vals = []
38 x_vector = []
39 for k,col in enumerate(x_cols):
40 x_cols[k] = int(col)-1
41 x_vals.append([])
42
43 NA = 'NA'
44 for ind,line in enumerate( file( infile )):
45 if line and not line.startswith( '#' ):
46 try:
47 fields = line.split("\t")
48 try:
49 yval = float(fields[y_col])
50 except:
51 yval = r('NA')
52 y_vals.append(yval)
53 for k,col in enumerate(x_cols):
54 try:
55 xval = float(fields[col])
56 except:
57 xval = r('NA')
58 x_vals[k].append(xval)
59 x_vector.append(xval)
60 except Exception, e:
61 print e
62 pass
63
64 #x_vals1 = numpy.asarray(x_vals).transpose()
65
66 check1=0
67 check0=0
68 for i in y_vals:
69 if i == 1:
70 check1=1
71 if i == 0:
72 check0=1
73 if check1==0 or check0==0:
74 sys.exit("Warning: logistic regression must have at least two classes")
75
76 for i in y_vals:
77 if i not in [1,0,r('NA')]:
78 print >>fout, str(i)
79 sys.exit("Warning: the current version of this tool can run only with two classes and need to be labeled as 0 and 1.")
80
81
82 #dat= r.list(x=array(x_vals1), y=y_vals)
83 novif=0
84 #set_default_mode(NO_CONVERSION)
85 #try:
86 # linear_model = r.glm(r("y ~ x"), data = r.na_exclude(dat),family="binomial")
87 # #r('library(car)')
88 # #r.assign('dat',dat)
89 # #r.assign('ncols',len(x_cols))
90 # #r.vif(r('glm(dat$y ~ ., data = na.exclude(data.frame(as.matrix(dat$x,ncol=ncols))->datx),family="binomial")')).as_py()
91 #
92 #except RException, rex:
93 # stop_err("Error performing logistic regression on the input data.\nEither the response column or one of the predictor columns contain only non-numeric or invalid values.")
94
95 fv = robjects.FloatVector(x_vector)
96 m = r['matrix'](fv, ncol=len(x_cols),byrow=True)
97 # ensure order for generating formula
98 od = rlc.OrdDict([('y',robjects.FloatVector(y_vals)),('x',m)])
99 dat = robjects.DataFrame(od)
100 # convert dat.names: ["y","x.1","x.2"] to formula string: 'y ~ x.1 + x.2'
101 formula = ' + '.join(dat.names).replace('+','~',1)
102 print formula
103 try:
104 linear_model = r.glm(formula, data = r['na.exclude'](dat), family="binomial")
105 except Exception, rex:
106 stop_err("Error performing linear regression on the input data.\nEither the response column or one of the predictor columns contain only non-numeric or invalid values.")
107
108 if len(x_cols)>1:
109 try:
110 r('library(car)')
111 r.assign('dat',dat)
112 r.assign('ncols',len(x_cols))
113 #vif=r.vif(r('glm(dat$y ~ ., data = na.exclude(data.frame(as.matrix(dat$x,ncol=ncols))->datx),family="binomial")'))
114 od2 = rlc.OrdDict([('datx', m)])
115 glm_data_frame = robjects.DataFrame(od2)
116 glm_result = r.glm("dat$y ~ .", data = r['na.exclude'](glm_data_frame),family="binomial")
117 print 'Have glm'
118 vif = r.vif(glm_result)
119 except Exception, rex:
120 print rex
121 else:
122 novif=1
123
124 #set_default_mode(BASIC_CONVERSION)
125
126 #coeffs=linear_model.as_py()['coefficients']
127 coeffs=linear_model.rx2('coefficients')
128 #null_deviance=linear_model.as_py()['null.deviance']
129 null_deviance=linear_model.rx2('null.deviance')[0]
130 #residual_deviance=linear_model.as_py()['deviance']
131 residual_deviance=linear_model.rx2('deviance')[0]
132 #yintercept= coeffs['(Intercept)']
133 yintercept= coeffs.rx2('(Intercept)')[0]
134
135 summary = r.summary(linear_model)
136 #co = summary.get('coefficients', 'NA')
137 co = summary.rx2("coefficients")
138 print co
139 """
140 if len(co) != len(x_vals)+1:
141 stop_err("Stopped performing logistic regression on the input data, since one of the predictor columns contains only non-numeric or invalid values.")
142 """
143
144 try:
145 yintercept = r.round(float(yintercept), digits=10)[0]
146 #pvaly = r.round(float(co[0][3]), digits=10)
147 pvaly = r.round(float(co.rx(1,4)[0]), digits=10)[0]
148 except Exception, e:
149 print str(e)
150 pass
151 print >>fout, "response column\tc%d" %(y_col+1)
152 tempP=[]
153 for i in x_cols:
154 tempP.append('c'+str(i+1))
155 tempP=','.join(tempP)
156 print >>fout, "predictor column(s)\t%s" %(tempP)
157 print >>fout, "Y-intercept\t%s" %(yintercept)
158 print >>fout, "p-value (Y-intercept)\t%s" %(pvaly)
159
160 print coeffs
161 if len(x_vals) == 1: #Simple linear regression case with 1 predictor variable
162 try:
163 #slope = r.round(float(coeffs['x']), digits=10)
164 raw_slope = coeffs.rx2('x')[0]
165 slope = r.round(float(raw_slope), digits=10)[0]
166 except:
167 slope = 'NA'
168 try:
169 #pval = r.round(float(co[1][3]), digits=10)
170 pval = r.round(float(co.rx2(2,4)[0]), digits=10)[0]
171 except:
172 pval = 'NA'
173 print >>fout, "Slope (c%d)\t%s" %(x_cols[0]+1,slope)
174 print >>fout, "p-value (c%d)\t%s" %(x_cols[0]+1,pval)
175 else: #Multiple regression case with >1 predictors
176 ind=1
177 #while ind < len(coeffs.keys()):
178 print len(coeffs.names)
179 while ind < len(coeffs.names):
180 try:
181 #slope = r.round(float(coeffs['x'+str(ind)]), digits=10)
182 raw_slope = coeffs.rx2('x.' + str(ind))[0]
183 slope = r.round(float(raw_slope), digits=10)[0]
184 except:
185 slope = 'NA'
186 print >>fout, "Slope (c%d)\t%s" %(x_cols[ind-1]+1,slope)
187
188 try:
189 #pval = r.round(float(co[ind][3]), digits=10)
190 pval = r.round(float(co.rx2(ind+1, 4)[0]), digits=10)[0]
191 except:
192 pval = 'NA'
193 print >>fout, "p-value (c%d)\t%s" %(x_cols[ind-1]+1,pval)
194 ind+=1
195
196 #rsq = summary.get('r.squared','NA')
197 rsq = summary.rx2('r.squared')
198 if rsq == ri.RNULLType():
199 rsq = 'NA'
200 else:
201 rsq = rsq[0]
202
203
204 try:
205 #rsq= r.round(float((null_deviance-residual_deviance)/null_deviance), digits=5)
206 rsq= r.round(float((null_deviance-residual_deviance)/null_deviance), digits=5)[0]
207 #null_deviance= r.round(float(null_deviance), digits=5)
208 null_deviance= r.round(float(null_deviance), digits=5)[0]
209 #residual_deviance= r.round(float(residual_deviance), digits=5)
210 residual_deviance= r.round(float(residual_deviance), digits=5)[0]
211
212 except:
213 pass
214
215 print >>fout, "Null deviance\t%s" %(null_deviance)
216
217 print >>fout, "Residual deviance\t%s" %(residual_deviance)
218 print >>fout, "pseudo R-squared\t%s" %(rsq)
219 print >>fout, "\n"
220 print >>fout, 'vif'
221
222 if novif==0:
223 #py_vif=vif.as_py()
224 count=0
225 for i in sorted(vif.names):
226 print >>fout,'c'+str(x_cols[count]+1) ,str(vif.rx2(i)[0])
227 count+=1
228 elif novif==1:
229 print >>fout, "vif can calculate only when model have more than 1 predictor"