def permute_w_replacement(frame, axis=0):
'''
Permute the frame values across the given axis.
Create simulated dataset were the counts of each component (column)
in each sample (row), are randomly sampled from the all the
counts of that component in all samples.
Parameters
----------
frame : DataFrame
Frame to permute.
axis : {0, 1}
- 0 - Permute row values across columns
- 1 - Permute column values across rows
Returns
-------
Permuted DataFrame (new instance).
'''
from numpy.random import randint
axis = 1-_get_axis(axis)
s = frame.shape[axis]
fun = lambda x: x.values[randint(0,s,(1,s))][0]
perm = frame.apply(fun, axis=axis)
return perm
def alr(frame, ref=None, axis=0):
'''
Compute the additive log-ratio (alr) transformation
with respect to the component given in ref.
Parameters
----------
frame : DataFrame
Frame to be transformed
ref : valid label | None
Label of component to be used as the normalization reference.
i.e. values of other component will be divided by values of
this reference component.
IF None is passed (default), the last col/row is used as ref.
axis : {0, 1}
0 : transform each row (default)
1 : transform each colum
'''
if not isinstance(frame, DF):
return alr_for_array(frame, ref, axis)
axis = _get_axis(axis)
if ref is None:
label = frame._get_axis(1 - axis)[-1]
else:
label = ref
if axis == 0:
norm = 1. * frame[label]
elif axis == 1:
norm = 1. * frame.xs(label)
temp = frame.apply(lambda x: log(x / norm), axis=axis)
return temp.drop(label, 1 - axis)
def alr_for_array(frame, ref=None, axis=0):
axis = _get_axis(axis)
if ref is None:
label = -1
else:
label = ref
if axis == 0:
norm = 1. * frame[:, label]
elif axis == 1:
norm = 1. * frame[label, :]
temp = np.apply_along_axis(lambda x: log(x / norm), axis, frame)
return np.delete(temp, label, 1 - axis)
def correlation(frame, method='pearson', axis=0):
'''
Calculate the correlation between all rows/cols.
Return frames of correlation values and p-values.
Parameters
----------
frame : DataFrame
Frame containing data.
method : {pearson (default) | spearman | kendall}
Type of correlations to be computed
axis : {0, 1}
- 0 - Compute correlation between columns
- 1 - Compute correlation between rows
Returns
-------
c : frame
DataFrame of symmetric pairwise correlation coefficients.
Labels are the rows/column labels of the input frame.
p : frame
DataFrame of p-values associated with correlation values.
Labels are the rows/column labels of the input frame.
'''
import scipy.stats as stats
axis = _get_axis(axis)
method = method.lower()
if method not in set(['pearson', 'kendall', 'spearman']):
raise ValueError('Correlation of method %s is not supported.' %method)
if method == 'spearman' :
c_mat, p_mat = stats.spearmanr(frame.values, axis=axis)
if not np.shape(c_mat):
c_mat = np.array([[1, c_mat],[c_mat,1]])
p_mat = np.array([[1, p_mat],[p_mat,1]])
labels = frame._get_axis(1-axis)
c = DF(c_mat, index=labels, columns=labels)
p = DF(p_mat, index=labels, columns=labels)
else:
if method == 'pearson': corr_fun = stats.pearsonr
elif method == 'kendall': corr_fun = stats.kendalltau
if axis == 0: data = frame.T
elif axis == 1: data = frame
mat = data.values
row_labels = data.index
n = len(row_labels)
c_mat = np.zeros((n, n))
p_mat = np.zeros((n, n))
for i in range(n):
for j in range(i, n):
if i == j:
c_mat[i][i] = 1
p_mat[i][i] = 1
continue
c_temp, p_temp = corr_fun(mat[i, :], mat[j, :])
c_mat[i][j] = c_temp
c_mat[j][i] = c_temp
p_mat[i][j] = p_temp
p_mat[j][i] = p_temp
c = DF(c_mat, index=row_labels, columns=row_labels)
p = DF(p_mat, index=row_labels, columns=row_labels)
return c, p