def merge_columns(self, groups, grouping_criterion):
""" Returns a new ArrayTable object in which columns are
merged according to a given criterion.
'groups' argument must be a dictionary in which keys are the
new column names, and each value is the list of current
column names to be merged.
'grouping_criterion' must be 'min', 'max' or 'mean', and
defines how numeric values will be merged.
Example:
my_groups = {'NewColumn':['column5', 'column6']}
new_Array = Array.merge_columns(my_groups, 'max')
"""
if grouping_criterion == "max":
grouping_f = get_max_vector
elif grouping_criterion == "min":
grouping_f = get_min_vector
elif grouping_criterion == "mean":
grouping_f = get_mean_vector
else:
raise ValueError, "grouping_criterion not supported. Use max|min|mean "
grouped_array = self.__class__()
grouped_matrix = []
colNames = []
alltnames = set([])
for gname, tnames in groups.iteritems():
all_vectors = []
for tn in tnames:
if tn not in self.colValues:
raise ValueError, str(tn) + " column not found."
if tn in alltnames:
raise ValueError, str(
tn) + " duplicated column name for merging"
alltnames.add(tn)
vector = self.get_column_vector(tn).astype(float)
all_vectors.append(vector)
# Store the group vector = max expression of all items in group
grouped_matrix.append(grouping_f(all_vectors))
# store group name
colNames.append(gname)
for cname in self.colNames:
if cname not in alltnames:
grouped_matrix.append(self.get_column_vector(cname))
colNames.append(cname)
grouped_array.rowNames = self.rowNames
grouped_array.colNames = colNames
vmatrix = numpy.array(grouped_matrix).transpose()
grouped_array._link_names2matrix(vmatrix)
return grouped_array
def merge_columns(self, groups, grouping_criterion):
""" Returns a new ArrayTable object in which columns are
merged according to a given criterion.
'groups' argument must be a dictionary in which keys are the
new column names, and each value is the list of current
column names to be merged.
'grouping_criterion' must be 'min', 'max' or 'mean', and
defines how numeric values will be merged.
Example:
my_groups = {'NewColumn':['column5', 'column6']}
new_Array = Array.merge_columns(my_groups, 'max')
"""
if grouping_criterion == "max":
grouping_f = get_max_vector
elif grouping_criterion == "min":
grouping_f = get_min_vector
elif grouping_criterion == "mean":
grouping_f = get_mean_vector
else:
raise ValueError, "grouping_criterion not supported. Use max|min|mean "
grouped_array = self.__class__()
grouped_matrix = []
colNames = []
alltnames = set([])
for gname,tnames in groups.iteritems():
all_vectors=[]
for tn in tnames:
if tn not in self.colValues:
raise ValueError, str(tn)+" column not found."
if tn in alltnames:
raise ValueError, str(tn)+" duplicated column name for merging"
alltnames.add(tn)
vector = self.get_column_vector(tn).astype(float)
all_vectors.append(vector)
# Store the group vector = max expression of all items in group
grouped_matrix.append(grouping_f(all_vectors))
# store group name
colNames.append(gname)
for cname in self.colNames:
if cname not in alltnames:
grouped_matrix.append(self.get_column_vector(cname))
colNames.append(cname)
grouped_array.rowNames= self.rowNames
grouped_array.colNames= colNames
vmatrix = numpy.array(grouped_matrix).transpose()
grouped_array._link_names2matrix(vmatrix)
return grouped_array
def safe_mean_vector(vectors):
""" Returns mean profile discarding non finite values.
"""
# if only one vector, avg = itself
if len(vectors)==1:
return vectors[0], numpy.zeros(len(vectors[0]))
# Takes the vector length form the first item
length = len(vectors[0])
safe_mean = []
safe_std = []
for pos in xrange(length):
pos_mean = []
for v in vectors:
if numpy.isfinite(v[pos]):
pos_mean.append(v[pos])
safe_mean.append(numpy.mean(pos_mean))
safe_std.append(numpy.std(pos_mean))
return numpy.array(safe_mean), numpy.array(safe_std)
def safe_mean_vector(vectors):
""" Returns mean profile discarding non finite values.
"""
# if only one vector, avg = itself
if len(vectors) == 1:
return vectors[0], numpy.zeros(len(vectors[0]))
# Takes the vector length form the first item
length = len(vectors[0])
safe_mean = []
safe_std = []
for pos in xrange(length):
pos_mean = []
for v in vectors:
if numpy.isfinite(v[pos]):
pos_mean.append(v[pos])
safe_mean.append(numpy.mean(pos_mean))
safe_std.append(numpy.std(pos_mean))
return numpy.array(safe_mean), numpy.array(safe_std)
def read_arraytable(matrix_file, mtype="float", arraytable_object = None):
""" Reads a text tab-delimited matrix from file """
if arraytable_object is None:
from ete2.coretype import arraytable
A = arraytable.ArrayTable()
else:
A = arraytable_object
A.mtype = mtype
temp_matrix = []
rowname_counter = {}
colname_counter = {}
row_dup_flag = False
col_dup_flag = False
# if matrix_file has many lines, tries to read it as the matrix
# itself.
if len(matrix_file.split("\n"))>1:
matrix_data = matrix_file.split("\n")
else:
matrix_data = open(matrix_file)
for line in matrix_data:
# Clean up line
line = line.strip("\n")
#line = line.replace(" ","")
# Skip empty lines
if not line:
continue
# Get fields in line
fields = line.split("\t")
# Read column names
if line[0]=='#' and re.match("#NAMES",fields[0],re.IGNORECASE):
counter = 0
for colname in fields[1:]:
colname = colname.strip()
# Handle duplicated col names by adding a number
colname_counter[colname] = colname_counter.get(colname,0) + 1
if colname in A.colValues:
colname += "_%d" % colname_counter[colname]
col_dup_flag = True
# Adds colname
A.colValues[colname] = None
A.colNames.append(colname)
if col_dup_flag:
print >>stderr, "Duplicated column names were renamed."
# Skip comments
elif line[0]=='#':
continue
# Read values (only when column names are loaded)
elif A.colNames:
# Checks shape
if len(fields)-1 != len(A.colNames):
raise ValueError, "Invalid number of columns. Expecting:%d" % len(A.colNames)
# Extracts row name and remove it from fields
rowname = fields.pop(0).strip()
# Handles duplicated row names by adding a number
rowname_counter[rowname] = rowname_counter.get(rowname,0) + 1
if rowname in A.rowValues:
rowname += "_%d" % rowname_counter[rowname]
row_dup_names = True
# Adds row name
A.rowValues[rowname] = None
A.rowNames.append(rowname)
# Reads row values
values = []
for f in fields:
if f.strip()=="":
f = numpy.nan
values.append(f)
temp_matrix.append(values)
else:
raise ValueError, "Column names are required."
if row_dup_flag:
print >>stderr, "Duplicated row names were renamed."
# Convert all read lines into a numpy matrix
vmatrix = numpy.array(temp_matrix).astype(A.mtype)
# Updates indexes to link names and vectors in matrix
A._link_names2matrix(vmatrix)
return A
def get_several_row_vectors(self, rownames):
""" Returns a list vectors associated to several row names """
vectors = [self.rowValues[rname] for rname in rownames]
return numpy.array(vectors)
def get_several_column_vectors(self, colnames):
""" Returns a list of vectors associated to several column names """
vectors = [self.colValues[cname] for cname in colnames]
return numpy.array(vectors)
def get_min_vector(vlist):
a = numpy.array(vlist)
return numpy.min(a, 0)
def get_max_vector(vlist):
a = numpy.array(vlist)
return numpy.max(a, 0)
def get_median_vector(vlist):
a = numpy.array(vlist)
return numpy.median(a)
def get_several_row_vectors(self,rownames):
""" Returns a list vectors associated to several row names """
vectors = [self.rowValues[rname] for rname in rownames]
return numpy.array(vectors)
def get_several_column_vectors(self,colnames):
""" Returns a list of vectors associated to several column names """
vectors = [self.colValues[cname] for cname in colnames]
return numpy.array(vectors)
def get_min_vector(vlist):
a = numpy.array(vlist)
return numpy.min(a,0)
def get_max_vector(vlist):
a = numpy.array(vlist)
return numpy.max(a,0)
def get_median_vector(vlist):
a = numpy.array(vlist)
return numpy.median(a)