Mercurial > repos > george-weingart > lefse
view format_input_selector.py @ 0:e7cd19afda2e draft
Lefse
author | george-weingart |
---|---|
date | Tue, 13 May 2014 21:57:00 -0400 |
parents | |
children |
line wrap: on
line source
from galaxy import datatypes,model import sys,string,time def timenow(): """return current time as a string """ return time.strftime('%d/%m/%Y %H:%M:%S', time.localtime(time.time())) def get_opt(data): return [('r','r',False),('c','c',False)] def red(st,l): if len(st) <= l: return st l1,l2 = l/2,l/2 return st[:l1]+".."+st[len(st)-l2:] def get_row_names(data,t): if data == "": return [] max_len =38 fname = data.dataset.file_name opt = [] rc = '' # lines = [(red(v.split()[0],max_len),'%s' % str(v.split()[0]),False) for i,v in enumerate(open(fname))] if t == 'b': lines = [(red(v.split()[0],max_len),'%d' % (i+1),False) for i,v in enumerate(open(fname)) if len(v.split()) > 3 ] else: lines = [(red(v.split()[0],max_len),'%d' % (i+1),False) for i,v in enumerate(open(fname))] return sorted(opt+lines) def get_cols(data,t,c): if data == "": return [] max_len =32 fname = data.dataset.file_name opt = [] if c != 'cl': opt.append(('no '+c,'%d' % -1,False)) if t == 'c': rc = '' lines = [(red((rc+"#"+str(i+1)+":"+v[0]),max_len),'%d' % (i+1),False) for i,v in enumerate(zip(*[line.split() for line in open(fname)]))] else: rc = '' lines = [(red((rc+"#"+str(i+1)+":"+v.split()[0]),max_len),'%d' % (i+1),False) for i,v in enumerate(open(fname))] return opt+lines """ def get_phecols(i,addNone,hint): hint = hint.lower() fname = i.dataset.file_name try: f = open(fname,'r') except: return [('get_phecols unable to open file "%s"' % fname,'None',False),] header = f.next() h = header.strip().split() dat = [(x,'%d' % i,False) for i,x in enumerate(h)] matches = [i for i,x in enumerate(h) if x.lower().find(hint) <> -1] if len(matches) > 0: sel = matches[0] dat[sel] = (dat[sel][0],dat[sel][1],True) if addNone: dat.insert(0,('None - no Manhattan plot','0', False )) return dat """ """ def exec_after_process(app, inp_data, out_data, param_dict, tool, stdout, stderr): outfile = 'out_html' job_name = param_dict.get( 'name', 'Manhattan QQ plots' ) killme = string.punctuation + string.whitespace trantab = string.maketrans(killme,'_'*len(killme)) newname = '%s.html' % job_name.translate(trantab) data = out_data[outfile] data.name = newname data.info='%s run at %s' % (job_name,timenow()) out_data[outfile] = data app.model.context.flush() """