def corr_circular_linear(alpha, X):
# Authors: Jean-Remi King <*****@*****.**>
#
# Licence : BSD-simplified
"""
Parameters
----------
alpha : numpy.array, shape (n_angles,)
The angular data (if n_dims == 1, repeated across all x dimensions)
X : numpy.array, shape (n_angles, n_dims)
The linear data
Returns
-------
R : numpy.array, shape (n_dims)
R values
R2 : numpy.array, shape (n_dims)
R square values
p_val : numpy.array, shape (n_dims)
P values
Adapted from:
Circular Statistics Toolbox for Matlab
By Philipp Berens, 2009
[email protected] - www.kyb.mpg.de/~berens/circStat.html
Equantion 27.47
"""
from scipy.stats import chi2
from jr.utils import pairwise
import numpy as np
# computes correlation for sin and cos separately
# WIP Applies repeated correlation if X is vector
# TODO: deals with non repeated correlations (X * ALPHA)
if alpha.ndim > 1:
rxs = repeated_corr(np.sin(alpha), X)
rxc = repeated_corr(np.cos(alpha), X)
rcs = np.zeros_like(alpha[0, :])
rcs = pairwise(np.sin(alpha), np.cos(alpha), func=_loop_corr,
n_jobs=-1)
else:
# WIP Applies repeated correlation if alpha is vector
rxs = repeated_corr(X, np.sin(alpha))
rxc = repeated_corr(X, np.cos(alpha))
rcs = repeated_corr(np.sin(alpha), np.cos(alpha))
# Adapted from equation 27.47
R = (rxc ** 2 + rxs ** 2 - 2 * rxc * rxs * rcs) / (1 - rcs ** 2)
# JR adhoc way of having a sign....
R = np.sign(rxs) * np.sign(rxc) * R
R2 = np.sqrt(R ** 2)
# Get degrees of freedom
n = len(X)
pval = 1 - chi2.cdf(n * R2, 2)
return R, R2, pval
def corr_circular_linear(alpha, X):
# Authors: Jean-Remi King <*****@*****.**>
#
# Licence : BSD-simplified
"""
Parameters
----------
alpha : numpy.array, shape (n_angles,)
The angular data (if n_dims == 1, repeated across all x dimensions)
X : numpy.array, shape (n_angles, n_dims)
The linear data
Returns
-------
R : numpy.array, shape (n_dims)
R values
R2 : numpy.array, shape (n_dims)
R square values
p_val : numpy.array, shape (n_dims)
P values
Adapted from:
Circular Statistics Toolbox for Matlab
By Philipp Berens, 2009
[email protected] - www.kyb.mpg.de/~berens/circStat.html
Equantion 27.47
"""
from scipy.stats import chi2
from jr.utils import pairwise
import numpy as np
# computes correlation for sin and cos separately
# WIP Applies repeated correlation if X is vector
# TODO: deals with non repeated correlations (X * ALPHA)
if alpha.ndim > 1:
rxs = repeated_corr(np.sin(alpha), X)
rxc = repeated_corr(np.cos(alpha), X)
rcs = np.zeros_like(alpha[0, :])
rcs = pairwise(np.sin(alpha), np.cos(alpha), func=_loop_corr,
n_jobs=-1)
else:
# WIP Applies repeated correlation if alpha is vector
rxs = repeated_corr(X, np.sin(alpha))
rxc = repeated_corr(X, np.cos(alpha))
rcs = repeated_corr(np.sin(alpha), np.cos(alpha))
# Adapted from equation 27.47
R = (rxc ** 2 + rxs ** 2 - 2 * rxc * rxs * rcs) / (1 - rcs ** 2)
# JR adhoc way of having a sign....
R = np.sign(rxs) * np.sign(rxc) * R
R2 = np.sqrt(R ** 2)
# Get degrees of freedom
n = len(X)
pval = 1 - chi2.cdf(n * R2, 2)
return R, R2, pval
def nested_analysis(X,
df,
condition,
function=None,
query=None,
single_trial=False,
y=None,
n_jobs=-1):
""" Apply a nested set of analyses.
Parameters
----------
X : np.array, shape(n_samples, ...)
Data array.
df : pandas.DataFrame
Condition DataFrame
condition : str | list
If string, get the samples for each unique value of df[condition]
If list, nested call nested_analysis.
query : str | None, optional
To select a subset of trial using pandas.DataFrame.query()
function : function
Computes across list of evoked. Must be of the form:
function(X[:], y[:])
y : np.array, shape(n_conditions)
n_jobs : int
Number of core to compute the function. Defaults to -1.
Returns
-------
scores : np.array, shape(...)
The results of the function
sub : dict()
Contains results of sub levels.
"""
import numpy as np
from jr.utils import pairwise
if isinstance(condition, str):
# Subselect data using pandas.DataFrame queries
sel = range(len(X)) if query is None else df.query(query).index
X = X.take(sel, axis=0)
y = np.array(df[condition][sel])
# Find unique conditions
values = list()
for ii in np.unique(y):
if (ii is not None) and (ii not in [np.nan]):
try:
if np.isnan(ii):
continue
else:
values.append(ii)
except TypeError:
values.append(ii)
# Subsubselect for each unique condition
y_sel = [np.where(y == value)[0] for value in values]
# Mean condition:
X_mean = np.zeros(np.hstack((len(y_sel), X.shape[1:])))
y_mean = np.zeros(len(y_sel))
for ii, sel_ in enumerate(y_sel):
X_mean[ii, ...] = np.mean(X[sel_, ...], axis=0)
if isinstance(y[sel_[0]], str):
y_mean[ii] = ii
else:
y_mean[ii] = y[sel_[0]]
if single_trial:
X = X.take(np.hstack(y_sel), axis=0) # ERROR COME FROM HERE
y = y.take(np.hstack(y_sel), axis=0)
else:
X = X_mean
y = y_mean
# Store values to keep track
sub_list = dict(X=X_mean,
y=y_mean,
sel=sel,
query=query,
condition=condition,
values=values,
single_trial=single_trial)
elif isinstance(condition, list):
# If condition is a list, we must recall the function to gather
# the results of the lower levels
sub_list = list()
X_list = list() # FIXME use numpy array
for subcondition in condition:
scores, sub = nested_analysis(
X,
df,
subcondition['condition'],
n_jobs=n_jobs,
function=subcondition.get('function', None),
query=subcondition.get('query', None))
X_list.append(scores)
sub_list.append(sub)
X = np.array(X_list)
if y is None:
y = np.arange(len(condition))
if len(y) != len(X):
raise ValueError('X and y must be of identical shape: ' +
'%s <> %s') % (len(X), len(y))
sub_list = dict(X=X, y=y, sub=sub_list, condition=condition)
# Default function
function = _default_analysis if function is None else function
scores = pairwise(X, y, function, n_jobs=n_jobs)
return scores, sub_list
def nested_analysis(X, df, condition, function=None, query=None,
single_trial=False, y=None, n_jobs=-1):
""" Apply a nested set of analyses.
Parameters
----------
X : np.array, shape(n_samples, ...)
Data array.
df : pandas.DataFrame
Condition DataFrame
condition : str | list
If string, get the samples for each unique value of df[condition]
If list, nested call nested_analysis.
query : str | None, optional
To select a subset of trial using pandas.DataFrame.query()
function : function
Computes across list of evoked. Must be of the form:
function(X[:], y[:])
y : np.array, shape(n_conditions)
n_jobs : int
Number of core to compute the function. Defaults to -1.
Returns
-------
scores : np.array, shape(...)
The results of the function
sub : dict()
Contains results of sub levels.
"""
import numpy as np
from jr.utils import pairwise
if isinstance(condition, str):
# Subselect data using pandas.DataFrame queries
sel = range(len(X)) if query is None else df.query(query).index
X = X.take(sel, axis=0)
y = np.array(df[condition][sel])
# Find unique conditions
values = list()
for ii in np.unique(y):
if (ii is not None) and (ii not in [np.nan]):
try:
if np.isnan(ii):
continue
else:
values.append(ii)
except TypeError:
values.append(ii)
# Subsubselect for each unique condition
y_sel = [np.where(y == value)[0] for value in values]
# Mean condition:
X_mean = np.zeros(np.hstack((len(y_sel), X.shape[1:])))
y_mean = np.zeros(len(y_sel))
for ii, sel_ in enumerate(y_sel):
X_mean[ii, ...] = np.mean(X[sel_, ...], axis=0)
if isinstance(y[sel_[0]], str):
y_mean[ii] = ii
else:
y_mean[ii] = y[sel_[0]]
if single_trial:
X = X.take(np.hstack(y_sel), axis=0) # ERROR COME FROM HERE
y = y.take(np.hstack(y_sel), axis=0)
else:
X = X_mean
y = y_mean
# Store values to keep track
sub_list = dict(X=X_mean, y=y_mean, sel=sel, query=query,
condition=condition, values=values,
single_trial=single_trial)
elif isinstance(condition, list):
# If condition is a list, we must recall the function to gather
# the results of the lower levels
sub_list = list()
X_list = list() # FIXME use numpy array
for subcondition in condition:
scores, sub = nested_analysis(
X, df, subcondition['condition'], n_jobs=n_jobs,
function=subcondition.get('function', None),
query=subcondition.get('query', None))
X_list.append(scores)
sub_list.append(sub)
X = np.array(X_list)
if y is None:
y = np.arange(len(condition))
if len(y) != len(X):
raise ValueError('X and y must be of identical shape: ' +
'%s <> %s') % (len(X), len(y))
sub_list = dict(X=X, y=y, sub=sub_list, condition=condition)
# Default function
function = _default_analysis if function is None else function
scores = pairwise(X, y, function, n_jobs=n_jobs)
return scores, sub_list