For viziphant version 0.1.0 the usage of the default parameters for the plot_isi_histogram results in

'numpy.float64' object has no attribute 'rescale'

Here is a minimal breaking example

st = neo.SpikeTrain([1,2,3]*pq.s,t_stop=10*pq.s)
viziphant.statistics.plot_isi_histogram(spiketrains)

Otherwise its easy to pass up this project because the results are unseen.

Currently, Viziphant can plot triple correlations, but the labeling is somewhat off, and only the hash [1 1 1] is correctly plotted. For other hashes non-involved neurons should not get a rectangle, e.g., for [1 0 1] no rectangles in cyan or red should be plotted for the middle neuron.

Generating the plot for UEA occasionally results in multiple instances of the following warning:

RuntimeWarning: invalid value encountered in greater_equal
  sig_idx_win = numpy.where(Js_dict['Js'] >= Js_sig)[0]

To Recreate:

import elephant.unitary_event_analysis as ue
import numpy as np
import quantities as pq
from elephant.spike_train_generation import homogeneous_poisson_process as hpp
from viziphant import unitary_event_analysis

np.random.seed(10)

spiketrains1 = [hpp(10 * pq.Hz, t_start=0.0*pq.ms, t_stop=2000*pq.ms) for _
                in range(5)]
spiketrains2 = [hpp(50 * pq.Hz, t_start=0.0*pq.ms, t_stop=2000*pq.ms) for _
                in range(5)]

data = np.vstack((spiketrains1, spiketrains2)).T
winsize = 100 * pq.ms
bin_size = 5 * pq.ms
winstep = 20 * pq.ms
pattern_hash = [3]
ue_dict = ue.jointJ_window_analysis(data, bin_size, winsize,
                                    winstep, pattern_hash)

significance_level = 0.05
plot_params = {'events': {}}
fig = unitary_event_analysis.plot_ue(
    data, ue_dict, significance_level, bin_size, winsize,
    winstep, pattern_hash, plot_params
)

The plotting of array grids (currently: array_data module) should utilize information of the Neo object to plot itself, e.g., which electrode is which channel, etc. Currently this is not well solved. This is related to work of @rgutzen in trying to find a nice way to encode this (e.g., using array annotations or even modifications in Neo), but also is related to the imagesequence objects introduced in Neo 0.8.

Tested with:

• python: 3.8.10
• viziphant: 0.2.0
• matplotlib: 3.6.2

The following example:

import numpy as np
import quantities as pq
from elephant.gpfa import GPFA
from elephant.spike_train_generation import StationaryPoissonProcess
from viziphant.gpfa import plot_trajectories
from matplotlib import pyplot as plt

data = []
for trial in range(50):
    n_channels = 20
    firing_rates = np.random.randint(low=1, high=100, size=n_channels) * pq.Hz
    spike_times = [StationaryPoissonProcess(rate=rate).generate_spiketrain()
                   for rate in firing_rates]
    data.append(spike_times)

gpfa = GPFA(bin_size=20*pq.ms, x_dim=8)
gpfa.fit(data)

results = gpfa.transform(data, returned_data=['latent_variable_orth',
                                              'latent_variable'])

trial_id_lists = np.arange(50).reshape(5, 10)
trial_group_names = ['A', 'B', 'C', 'D', 'E']

trial_grouping_dict = {}
for trial_group_name, trial_id_list in zip(trial_group_names, trial_id_lists):
    trial_grouping_dict[trial_group_name] = trial_id_list

plot_trajectories(results,
                  gpfa,
                  dimensions=[0, 1, 2],
                  trial_grouping_dict=trial_grouping_dict,
                  plot_group_averages=True)
plt.show()

gives the following error:

line 572, in plot_trajectories
axes = fig.gca(projection=projection, aspect='auto')
TypeError: gca() got an unexpected keyword argument 'projection'

Looking into the code for the rasterplot_rates function, the computation of the population histogram uses the matplotlib hist function (lines 465 and 469 of rasterplot.py).

It would be useful to reuse the time_histogram function from Elephant, which provides more options to control the histogram, especially the ability to normalize by either mean or density.

The plot_patterns function apparently only works for synchronous patterns, but not when there is a lag between the spikes.

The code below illustrates that

import numpy as np
import quantities as pq
import neo
from elephant.spade import spade
from elephant.spike_train_generation import homogeneous_poisson_process
from copy import deepcopy
import viziphant

# Generate two types of pattern: one synchronous and other with lags

np.random.seed(123)
sync_spiketrains = [homogeneous_poisson_process(rate=3*pq.Hz, t_stop=1*pq.s) for _ in range(50)]
lag_spiketrains = deepcopy(sync_spiketrains)

pattern_times = pq.Quantity([212, 574, 876], units=pq.ms)

pattern_neurons = []
pattern = neo.SpikeTrain(pattern_times, t_stop=1*pq.s)
for st in sync_spiketrains[:4]:
    pattern_st = st.merge(pattern)
    pattern_neurons.append(pattern_st)
sync_spiketrains[:4] = pattern_neurons

pattern_neurons = []
for idx, st in enumerate(lag_spiketrains[:4]):
    pattern = neo.SpikeTrain(pattern_times + idx * (10 * pq.ms), t_stop=1*pq.s)
    pattern_st = st.merge(pattern)
    pattern_neurons.append(pattern_st)
lag_spiketrains[:4] = pattern_neurons

Run SPADE using the synchronous patterns and plot:

sync_patterns = spade(sync_spiketrains, bin_size=5*pq.ms, winlen=10, min_spikes=3,
    n_surr=100, dither=5*pq.ms, psr_param=[0,0,0], alpha=0.05)

print(sync_patterns['patterns'])
viziphant.patterns.plot_patterns(sync_spiketrains, sync_patterns['patterns'])

Output:

[{'itemset': (20, 0, 10, 30),
'windows_ids': (42, 114, 175),
'neurons': [2, 0, 1, 3],
'lags': array([0., 0., 0.]) * ms,
'times': array([210., 570., 875.]) * ms,
'signature': (4, 3),
'pvalue': 0.01}]

Run SPADE using the pattern with lags and plot:

lag_patterns = spade(lag_spiketrains, bin_size=5*pq.ms, winlen=10, min_spikes=3,
    n_surr=100, dither=5*pq.ms, psr_param=[0,0,0], alpha=0.05)
print(lag_patterns['patterns'])
viziphant.patterns.plot_patterns(lag_spiketrains, lag_patterns['patterns'])

Output:

[{'itemset': (24, 36, 12, 0),
'windows_ids': (42, 114, 175),
'neurons': [0, 1, 2, 3],
'lags': array([10., 20., 30.]) * ms,
'times': array([210., 570., 875.]) * ms,
'signature': (4, 3),
'pvalue': 0.0}]

Expected output

Spikes in the second plot would be shown with the lags, overlapping with the original spike trains in gray.

Unitary events are displayed outside of the region where the analytical significance crosses the p-value threshold. The issue persists in the old UEA plots, so it's not a regression bug introduced by the recent changes in #32.

The figure is a zoomed-in part of the UEA tutorial.

Updated the plot.

Even if the events are not spurious, they look as such. What should we do about it?

BTW, the bottom axis 5 is not shifted by win_size / 2 in the X direction as in axis 4, for example.