40a9641385ff510d42513efd0140a6f51ec46dd9
mmaddren
  Thu Jun 9 11:22:16 2011 -0700
deleted unnecessary files
diff --git python/ucscgenomics/rafile/raInspect.py python/ucscgenomics/rafile/raInspect.py
deleted file mode 100644
index 9600d6b..0000000
--- python/ucscgenomics/rafile/raInspect.py
+++ /dev/null
@@ -1,113 +0,0 @@
-import sys
-import re
-import raFile
-import filterFile
-
-path = sys.argv[1]
-fname = sys.argv[2]
-filter = list()
-
-filter = filterFile.FilterFile()
-filter.read(fname)
-
-ra = raFile.RaFile()
-ra.read(path)
-newra = raFile.RaFile()
-
-# create a dictionary to keep track of which RaEntries have what terms
-keyDict = dict()
-for entry in ra.iterValues():
-    if entry == None:
-        continue
-    for elem in entry.iterKeys():
-        if elem not in keyDict:
-            keyDict[elem] = list()
-        keyDict[elem].append(entry.getValueAt(0))
-
-cellDict = dict()
-for entry in ra.iterValues():
-    elem = entry.getValue('cell')
-    if elem not in keyDict:
-        cellDict[elem] = list()
-    cellDict[elem].append(entry.getValueAt(0))
-
-#print cellDict
-
-# create a dictionary to keep track of what entries match what filters
-matchDict = dict()
-for entry in filter.iterValues():
-    matchDict[entry.getValueAt(0)] = list()
-
-# a dictionary to keep track of user made groups based on filter matches
-groupDict = dict()
-
-for i in range(ra.count()):
-    
-    if ra.getKeyAt(i).startswith('#'):
-        continue
-
-    r = ra.getValueAt(i)
-    for j in range(filter.count()):
-
-        match = True
-        f = filter.getValueAt(j) 
-        
-        if f.getKeyAt(0).startswith('_filter'):
-            for k in range(f.count()):
-                if f.getKeyAt(k).startswith('_'):
-                    continue
-                if r.getValue(f.getKeyAt(k)) == None or not re.match(f.getValueAt(k), r.getValue(f.getKeyAt(k))):
-                    match = False
-                    break
-
-            if match == True:
-                matchDict[f.getValueAt(0)].append(ra.getKeyAt(i))
-                for m in f.iterMatches():
-                    if m[1] == 'add':
-                        if m[0] not in groupDict:
-                            groupDict[m[0]] = list()
-                        #print 'adding ' + ra.getKeyAt(i) + ' to ' +  m[0]
-                        groupDict[m[0]].append(ra.getKeyAt(i))
-                    elif m[1] == 'remove':
-                        if m[0] in groupDict:
-                            #print 'removing ' + ra.getKeyAt(i) + ' from ' +  m[0]
-                            groupDict[m[0]].remove(ra.getKeyAt(i))
-
-#print groupDict
-
-for j in range(filter.count()):
-    f = filter.getValueAt(j)
-    if f.getKeyAt(0).startswith('_check'):
-        checkDict = dict()
-        #print ra
-        for entry in ra.iterValues():
-            elem = entry.getValue(f.getValue('_key'))
-            #print entry.getValueAt(0)
-            if elem not in checkDict:
-                checkDict[elem] = list()
-            checkDict[elem].append(entry.getValueAt(0))
-        #print checkDict
-        for key in checkDict.iterkeys():
-            if checkDict[key] == None:
-                continue
-            entry = checkDict[key]
-            if len(entry) > 1:
-                for m in f.iterMatches():
-                    if m[1] == 'add':
-                        #print entry
-                        hashstr = m[0] + ' (' + str(ra.getValue(entry[0]).getValue(f.getValue('_key'))) + ')'
-                        if hashstr not in groupDict:
-                            groupDict[hashstr] = list()
-                        groupDict[hashstr].extend(entry)
-                    elif m[1] == 'remove':
-                        if hashstr in groupDict:
-                            for rem in entry:
-                                groupDict[hashstr].remove(rem)
-
-for i in groupDict.iterkeys():
-    s = ''
-    print s + i + ' (' + str(len(groupDict[i])) + '):'
-    for j in groupDict[i]:
-        print j
-    print ''
-