def ar_testerr_non_uniform(dataLst, settings):
testFrac = settings['testfrac'] if 'testfrac' in settings.keys() else 0.1
dataLst2D = _preprocess_ar_non_uniform(dataLst, settings)
x, y = drop_nan_rows(
splitter.split2D_non_unifrom(dataLst2D, settings['hist']))
return _ar_2D_testerr(x, y, testFrac)
def plot_metric_bulk_1D(dataDB, ds, metricName, nameSuffix, prepFunc=None, xlim=None, ylim=None, yscale=None,
verbose=True, xFunc=None, haveTimeLabels=False):#, dropCols=None):
# 1. Extract all results for this test
dfAll = ds.list_dsets_pd().fillna('None')
# if dropCols is not None:
# dfAll = dfAll.drop(dropCols, axis=1)
dfAnalysis = pd_query(dfAll, {'metric' : metricName, "name" : nameSuffix})
dfAnalysis = pd_move_cols_front(dfAnalysis, ['metric', 'name', 'mousename']) # Move leading columns forwards for more informative printing/saving
dfAnalysis = dfAnalysis.drop(['target_dim', 'datetime', 'shape'], axis=1)
# Loop over all other columns except mousename
colsExcl = list(set(dfAnalysis.columns) - {'mousename', 'dset'})
for colVals, dfSub in dfAnalysis.groupby(colsExcl):
fig, ax = plt.subplots(figsize=(4, 4))
if verbose:
print(list(colVals))
for idxMouse, rowMouse in dfSub.sort_values(by='mousename').iterrows():
print(list(rowMouse.values))
dataThis = ds.get_data(rowMouse['dset'])
assert dataThis.ndim == 1, 'Only using 1D data for this plot function'
if prepFunc is not None:
dataThis = prepFunc(dataThis)
# if datatype == 'raw':
# nTrialThis = dataDB.get_ntrial_bytype({'mousename' : row['mousename']}, trialType=trialType, performance=performance)
# dataThis *= np.sqrt(48*nTrialThis)
# print('--', row['mousename'], nTrialThis)
x = np.arange(len(dataThis)) if xFunc is None else np.array(xFunc(rowMouse['mousename'], len(dataThis)))
x, dataThis = drop_nan_rows([x, dataThis])
ax.plot(x, dataThis, label=rowMouse['mousename'])
if yscale is not None:
ax.set_yscale(yscale)
if haveTimeLabels:
dataDB.label_plot_timestamps(ax, linecolor='y', textcolor='k', shX=-0.5, shY=0.05)
dataName = rowMouse.drop(['dset', 'mousename'])
dataName = '_'.join([str(el) for el in dataName])
prefixPath = 'pics/bulk/' + metricName + '/'
make_path(prefixPath)
ax.legend()
ax.set_xlim(xlim)
ax.set_ylim(ylim)
ax.set_xlabel(nameSuffix)
ax.set_ylabel(metricName)
plt.savefig(prefixPath + dataName + '.png', dpi=200)
plt.close()
def average_predictive_info(data, settings):
x, y = drop_nan_rows(split3D(data, settings['max_lag']))
nSample, nProcess = x.shape
if nSample < 5 + 5 * nProcess:
# If there are too few samples, there is no point to calculate anything
return np.array(np.nan)
else:
return ee.mi(x, y) / nProcess
def _preprocess_mar_inp(data, inp, nHist):
x, y = splitter.split3D(data, nHist)
assert inp.ndim == 3, "Input matrix must be a 3D matrix"
assert np.prod(inp.shape) != 0, "Input matrix is degenerate"
nTr, nCh, nT = data.shape
nTrInp, nChInp, nTInp = inp.shape
assert nTr == nTrInp, "Input shape must be consistent with data shape"
assert nT == nTInp, "Input shape must be consistent with data shape"
# Convert input into the form (rps) -> (r*s, p)
inpCanon = numpy_transpose_byorder(inp, 'rps', 'rsp')
u = numpy_merge_dimensions(inpCanon[:, nHist:], 0, 2)
# Drop any nan rows that are present in the data or input
return drop_nan_rows([x, y, u])
def _preprocess_mar_inp_non_uniform(dataLst, inpLst, nHist):
x, y = splitter.split3D_non_uniform(dataLst, nHist)
assert len(dataLst) == len(
inpLst), "Input must have same number of trials as data"
for data, inp in zip(dataLst, inpLst):
assert inp.ndim == 2, "Input must be a list of 2D matrices"
assert inp.shape[1] == data.shape[
1], "Input must have same number of timesteps as data"
# Test that input has the same number of features for each trial
nChInp = list_assert_get_uniform_shape(inpLst, axis=1)
# shape transform for y :: (rps) -> (r*s, p)
u = [inp[:, nHist:].T for inp in inpLst] # (rps) -> (rsp)
u = np.concatenate(u, axis=0) # (rsp) -> (r*s, p)
# Drop any nan rows that are present in the data or input
return drop_nan_rows([x, y, u])
def average_predictive_info_non_uniform(dataLst, settings):
# Test that all trials have sufficient timesteps for lag estimation
nSampleMin = np.min(set_list_shapes(dataLst, axis=1))
if nSampleMin <= settings['max_lag']:
raise ValueError('lag', settings['max_lag'], 'cannot be estimated for number of timesteps', nSampleMin)
xLst = []
yLst = []
for dataTrial in dataLst:
x, y = drop_nan_rows(split3D(dataTrial, settings['max_lag']))
xLst += [x]
yLst += [y]
xArr = np.vstack(xLst)
yArr = np.vstack(yLst)
nSample, nProcess = xArr.shape
if nSample < 4 * nProcess:
# If there are too few samples, there is no point to calculate anything
return np.array(np.nan)
else:
return ee.mi(xArr, yArr) / nProcess
def mar_testerr_non_uniform(dataLst, settings):
testFrac = settings['testfrac'] if 'testfrac' in settings.keys() else 0.1
x, y = drop_nan_rows(
splitter.split3D_non_uniform(dataLst, settings['hist']))
return _mar3D_testerr(x, y, testFrac)
def mar1_coeff_non_uniform(dataLst, settings):
x, y = drop_nan_rows(splitter.split3D_non_uniform(dataLst, 1))
return _mar3D_alpha(x, y)
def ar1_coeff_non_uniform(dataLst3D, settings):
dataLst2D = _preprocess_ar_non_uniform(dataLst3D, settings)
x, y = drop_nan_rows(splitter.split2D_non_unifrom(dataLst2D, 1))
return _ar_2D_alpha(x, y)
def mar1_testerr(data, settings):
testFrac = settings['testfrac'] if 'testfrac' in settings.keys() else 0.1
x, y = drop_nan_rows(splitter.split3D(data, 1))
return _mar3D_testerr(x, y, testFrac)
def mar1_coeff(data, settings):
x, y = drop_nan_rows(splitter.split3D(data, 1))
return _mar3D_alpha(x, y)
def ar1_testerr(data, settings):
testFrac = settings['testfrac'] if 'testfrac' in settings.keys() else 0.1
data2D = _preprocess_ar(data, settings)
x, y = drop_nan_rows(splitter.split2D(data2D, 1))
return _ar_2D_testerr(x, y, testFrac)
def ar1_coeff(data, settings):
data2D = _preprocess_ar(data, settings)
x, y = drop_nan_rows(splitter.split2D(data2D, 1))
return _ar_2D_alpha(x, y)
def poly_fit_transform(x, y, ord):
xEff, yEff = drop_nan_rows([x, y])
coeff = np.polyfit(xEff, yEff, ord) # Fit to data without nans
p = np.poly1d(coeff)
return p(x) # Evaluate for original data
def binary_classifier(data1,
data2,
classifier,
method="kfold",
k=10,
balancing=False,
pcaThr=None,
havePVal=False):
# Convert data to labeled form
labels = [-1, 1]
x, y = label_binary_data(data1, data2, *labels)
# Drop NAN values
xNoNan, yNoNan = drop_nan_rows([x, y])
if pcaThr is not None:
xNoNan = dim_reduction(xNoNan, pcaThr)
print('Reduced number of dimensions to', xNoNan.shape[1])
# map labels to binary variable
nData = len(yNoNan)
if nData == 0:
print("Warning: dataset had zero non-nan rows")
return {"acc_train": 0, "acc_test": 0, "acc_naive": 0, "p-value": 1}
nA = np.sum(yNoNan == 1) # Number of points with label 1
nB = nData - nA # Number of points with label -1
if (nA < 2) or (nB < 2):
print("Warning: unexpected number of labels", nA, nB,
"; aborting classification")
return 0, 0
# Add extra dimension if X is 1D
if xNoNan.ndim == 1:
xNoNan = xNoNan[:, None]
print('Warning: Got 1D data, had to add extra dimension')
cmTrain = np.zeros((2, 2), dtype=int)
cmTest = np.zeros((2, 2), dtype=int)
cvfunc = select_cv_iterator(method, xNoNan, yNoNan, k)
for xTrain, yTrain, xTest, yTest in cvfunc:
if balancing:
xTrainEff, yTrainEff = balance_oversample(xTrain, yTrain, labels)
else:
xTrainEff, yTrainEff = xTrain, yTrain
clf = classifier.fit(xTrainEff, yTrainEff)
# LogisticRegression(max_iter=1000)
cmTrain += confusion_matrix(clf.predict(xTrain), yTrain, labels)
cmTest += confusion_matrix(clf.predict(xTest), yTest, labels)
# print('cmTrain\n', cmTrain)
# print('cmTest\n', cmTest)
# Accuracy
accTrain = weighted_accuracy(cmTrain)
accTest = weighted_accuracy(cmTest)
rez = {"accTrain": accTrain, "accTest": accTest}
if havePVal:
rez = {**rez, **test_classifier_significance(nA, nB, cmTest)}
# rez = {**rez, **test_classifier_significance(nA, nB, len(yTest), accTest)}
return rez
def scatter_metric_bulk(ds, metricName, nameSuffix, prepFunc=None, xlim=None, ylim=None, yscale=None,
verbose=True, xFunc=None, haveRegression=False):#, dropCols=None):
# 1. Extract all results for this test
dfAll = ds.list_dsets_pd().fillna('None')
# if dropCols is not None:
# dfAll = dfAll.drop(dropCols, axis=1)
dfAnalysis = pd_query(dfAll, {'metric' : metricName, "name" : nameSuffix})
dfAnalysis = pd_move_cols_front(dfAnalysis, ['metric', 'name', 'mousename']) # Move leading columns forwards for more informative printing/saving
dfAnalysis = dfAnalysis.drop(['target_dim', 'datetime', 'shape'], axis=1)
if 'performance' in dfAnalysis.columns:
dfAnalysis = dfAnalysis[dfAnalysis['performance'] == 'None'].drop(['performance'], axis=1)
# Loop over all other columns except mousename
colsExcl = list(set(dfAnalysis.columns) - {'mousename', 'dset'})
for colVals, dfSub in dfAnalysis.groupby(colsExcl):
fig, ax = plt.subplots()
if verbose:
print(list(colVals))
xLst = []
yLst = []
for idxMouse, rowMouse in dfSub.sort_values(by='mousename').iterrows():
print(list(rowMouse.values))
dataThis = ds.get_data(rowMouse['dset'])
if prepFunc is not None:
dataThis = prepFunc(dataThis)
# if datatype == 'raw':
# nTrialThis = dataDB.get_ntrial_bytype({'mousename' : row['mousename']}, trialType=trialType, performance=performance)
# dataThis *= np.sqrt(48*nTrialThis)
# print('--', row['mousename'], nTrialThis)
x = np.arange(len(dataThis)) if xFunc is None else np.array(xFunc(rowMouse['mousename'], len(dataThis)))
print(dataThis.shape)
x, dataThis = drop_nan_rows([x, dataThis])
print(dataThis.shape)
ax.plot(x, dataThis, '.', label=rowMouse['mousename'])
xLst += [x]
yLst += [dataThis]
if yscale is not None:
plt.yscale(yscale)
dataName = rowMouse.drop(['dset', 'mousename'])
dataName = '_'.join([str(el) for el in dataName])
ax.legend()
ax.set_xlim(xlim)
ax.set_ylim(ylim)
if haveRegression:
sns.regplot(ax=ax, x=np.hstack(xLst), y=np.hstack(yLst), scatter=False)
prefixPath = 'pics/bulk/' + metricName + '/'
make_path(prefixPath)
fig.savefig(prefixPath + dataName + '.png')
plt.close()