Core methods#
Show code cell content
import pynapple as nap
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
custom_params = {"axes.spines.right": False, "axes.spines.top": False}
sns.set_theme(style="ticks", palette="colorblind", font_scale=1.5, rc=custom_params)
Time series method#
Show code cell content
tsdframe = nap.TsdFrame(t=np.arange(100), d=np.random.randn(100, 3), columns=['a', 'b', 'c'])
group = {
0: nap.Ts(t=np.sort(np.random.uniform(0, 100, 10))),
1: nap.Ts(t=np.sort(np.random.uniform(0, 100, 20))),
2: nap.Ts(t=np.sort(np.random.uniform(0, 100, 30))),
}
tsgroup = nap.TsGroup(group)
epochs = nap.IntervalSet([10, 65], [25, 80])
tsd = nap.Tsd(t=np.arange(0, 100, 1), d=np.sin(np.arange(0, 10, 0.1)))
restrict#
restrict is used to get time points within an IntervalSet. This method is available
for TsGroup, Tsd, TsdFrame, TsdTensor and Ts objects.
tsdframe.restrict(epochs)
Time (s) a b c
---------- -------- -------- --------
10.0 0.58014 1.09246 -1.89885
11.0 1.73859 1.7945 -0.36476
12.0 -0.16382 1.28021 -0.1422
13.0 0.32024 -0.50318 1.75336
14.0 0.12109 0.0179 -0.42167
15.0 0.62793 0.28765 -0.18939
16.0 -2.73951 0.54743 1.45959
... ... ... ...
74.0 -0.06583 1.89037 -1.83012
75.0 0.2276 0.89381 -0.62577
76.0 2.22353 -2.04517 2.10752
77.0 0.01721 0.05412 -2.14968
78.0 1.13387 0.21417 -2.56013
79.0 -1.10748 0.8861 2.00899
80.0 -1.23751 -0.98309 0.81676
dtype: float64, shape: (32, 3)
Show code cell source
plt.figure()
plt.plot(tsdframe.restrict(epochs))
[plt.axvspan(s, e, alpha=0.2) for s, e in epochs.values]
plt.xlabel("Time (s)")
plt.title("tsdframe.restrict(epochs)")
plt.xlim(0, 100)
plt.show()
This operation update the time support attribute accordingly.
print(epochs)
print(tsdframe.restrict(epochs).time_support)
index start end
0 10 25
1 65 80
shape: (2, 2), time unit: sec.
index start end
0 10 25
1 65 80
shape: (2, 2), time unit: sec.
count#
count the number of timestamps within bins or epochs of an IntervalSet object.
This method is available for TsGroup, Tsd, TsdFrame, TsdTensor and Ts objects.
With a defined bin size:
count1 = tsgroup.count(bin_size=1.0, time_units='s')
print(count1)
Time (s) 0 1 2
------------ --- --- ---
4.149925309 0 0 1
5.149925309 0 0 0
6.149925309 0 0 0
7.149925309 2 0 0
8.149925309 0 0 0
9.149925309 0 0 1
10.149925309 0 0 0
... ... ... ...
93.149925309 0 0 0
94.149925309 1 0 1
95.149925309 0 0 1
96.149925309 0 0 1
97.149925309 0 0 0
98.149925309 0 0 0
99.149925309 0 0 1
dtype: int64, shape: (96, 3)
Show code cell source
plt.figure()
plt.plot(count1[:,2], 'o-')
plt.title("tsgroup.count(bin_size=1.0)")
plt.plot(tsgroup[2].fillna(-1), '|', markeredgewidth=2)
[plt.axvline(t, linewidth=0.5, alpha=0.5) for t in np.arange(0, 21)]
plt.xlabel("Time (s)")
plt.xlim(0, 20)
plt.show()
With an IntervalSet:
count_ep = tsgroup.count(ep=epochs)
print(count_ep)
Time (s) 0 1 2
---------- --- --- ---
17.5 0 5 2
72.5 1 3 5
dtype: int64, shape: (2, 3)
bin_average#
bin_average downsample time series by averaging data point falling within a bin.
This method is available for Tsd, TsdFrame and TsdTensor.
tsdframe.bin_average(3.5)
Time (s) a b c
---------- -------- -------- --------
1.75 -0.2212 -0.32999 -0.78947
5.25 -0.46703 -0.30685 -0.05399
8.75 0.43073 0.93692 -0.20976
12.25 0.63167 0.85718 0.41547
15.75 -0.66011 0.12261 0.54071
19.25 0.46742 0.54481 -0.28054
22.75 1.41642 -0.52017 -0.82971
... ... ... ...
75.25 0.7951 0.24634 -0.11612
78.75 -0.29848 0.04282 -0.47101
82.25 0.25444 0.26292 1.88096
85.75 0.56476 0.1837 -0.98933
89.25 -0.84359 0.21206 0.3507
92.75 -0.18257 -0.35566 0.25774
96.25 0.94685 -0.08755 0.31382
dtype: float64, shape: (28, 3)
Show code cell source
bin_size = 3.5
plt.figure()
plt.plot(tsdframe[:,0], '.--', label="tsdframe[:,0]")
plt.plot(tsdframe[:,0].bin_average(bin_size), 'o-', label="new_tsdframe[:,0]")
plt.title(f"tsdframe.bin_average(bin_size={bin_size})")
[plt.axvline(t, linewidth=0.5, alpha=0.5) for t in np.arange(0, 21,bin_size)]
plt.xlabel("Time (s)")
plt.xlim(0, 20)
plt.legend(bbox_to_anchor=(1.0, 0.5, 0.5, 0.5))
plt.show()
interpolate#
Theinterpolate method of Tsd, TsdFrame and TsdTensor can be used to fill gaps in a time series. It is a wrapper of numpy.interp.
Show code cell content
tsd = nap.Tsd(t=np.arange(0, 25, 5), d=np.random.randn(5))
ts = nap.Ts(t=np.arange(0, 21, 1))
new_tsd = tsd.interpolate(ts)
Show code cell source
plt.figure()
plt.plot(new_tsd, '.-', label="new_tsd")
plt.plot(tsd, 'o', label="tsd")
plt.plot(ts.fillna(0), '+', label="ts")
plt.title("tsd.interpolate(ts)")
plt.xlabel("Time (s)")
plt.legend(bbox_to_anchor=(1.0, 0.5, 0.5, 0.5))
plt.show()
value_from#
By default, value_from assign to timestamps the closest value in time
from another time series. Let’s define the time series we want to assign values from.
For every timestamps in tsgroup, we want to assign the closest value in time from tsd.
Show code cell content
tsd = nap.Tsd(t=np.arange(0, 100, 1), d=np.sin(np.arange(0, 10, 0.1)))
tsgroup_from_tsd = tsgroup.value_from(tsd)
We can display the first element of tsgroup and tsgroup_sin.
Show code cell source
plt.figure()
plt.plot(tsgroup[0].fillna(0), "|", label="tsgroup[0]", markersize=20, mew=3)
plt.plot(tsd, linewidth=2, label="tsd")
plt.plot(tsgroup_from_tsd[0], "o", label = "tsgroup_from_tsd[0]", markersize=20)
plt.title("tsgroup.value_from(tsd)")
plt.xlabel("Time (s)")
plt.yticks([-1, 0, 1])
plt.legend(bbox_to_anchor=(1.0, 0.5, 0.5, 0.5))
plt.show()
The argument mode can control if the nearest target time is taken before or
after the reference time.
Show code cell content
tsd = nap.Tsd(t=np.arange(0, 10, 1), d=np.arange(0, 100, 10))
ts = nap.Ts(t=np.arange(0.5, 9, 1))
In this case, the variable ts receive data from the time point before.
new_ts_before = ts.value_from(tsd, mode="before")
Show code cell source
plt.figure()
plt.plot(ts.fillna(-1), "|", label="ts", markersize=20, mew=3)
plt.plot(tsd, "*-", linewidth=2, label="tsd")
plt.plot(new_ts_before, "o-", label = "new_ts_before", markersize=10)
plt.title("ts.value_from(tsd, mode='before')")
plt.xlabel("Time (s)")
plt.legend(bbox_to_anchor=(1.0, 0.5, 0.5, 0.5))
plt.show()
Show code cell source
new_ts_after = ts.value_from(tsd, mode="after")
plt.figure()
plt.plot(ts.fillna(-1), "|", label="ts", markersize=20, mew=3)
plt.plot(tsd, "*-", linewidth=2, label="tsd")
plt.plot(new_ts_after, "o-", label = "new_ts_after", markersize=10)
plt.title("ts.value_from(tsd, mode='after')")
plt.xlabel("Time (s)")
plt.legend(bbox_to_anchor=(1.0, 0.5, 0.5, 0.5))
plt.show()
If there is no time point found before or after or within the interval, the function assigns Nans.
tsd = nap.Tsd(t=np.arange(1, 10, 1), d=np.arange(10, 100, 10))
ep = nap.IntervalSet(start=0, end = 10)
ts = nap.Ts(t=[0, 9])
# First ts is at 0s. First tsd is at 1s.
ts.value_from(tsd, ep=ep, mode="before")
Time (s)
---------- ---
0 nan
9 90
dtype: float64, shape: (2,)
threshold#
The method threshold of Tsd returns a new Tsd with all the data above or
below a certain threshold. Default is above. The time support
of the new Tsd object get updated accordingly.
Show code cell content
tsd = nap.Tsd(t=np.arange(0, 100, 1), d=np.sin(np.arange(0, 10, 0.1)))
tsd_above = tsd.threshold(0.5, method='above')
This method can be used to isolate epochs for which a signal is above/below a certain threshold.
epoch_above = tsd_above.time_support
Show code cell source
plt.figure()
plt.plot(tsd, label="tsd")
plt.plot(tsd_above, 'o-', label="tsd_above")
[plt.axvspan(s, e, alpha=0.2) for s, e in epoch_above.values]
plt.axhline(0.5, linewidth=0.5, color='grey')
plt.legend()
plt.xlabel("Time (s)")
plt.title("tsd.threshold(0.5)")
plt.show()
to_trial_tensor#
Tsd, TsdFrame, and TsdTensor all have the method to_trial_tensor, which creates a numpy array from an IntervalSet by slicing the time series. The resulting tensor has shape (shape of time series, number of trials, number of time points), where the first dimension(s) is dependent on the object.
ep = nap.IntervalSet([0, 10, 30, 50], metadata={'trials':[1, 2]})
print(tsgroup, "\n", ep)
Index rate
------- -------
0 0.10449
1 0.20898
2 0.31347
index start end trials
0 0 10 1
1 30 50 2
shape: (2, 2), time unit: sec.
The following example returns a tensor with shape (2, 21), for 2 trials and 21 time points, where the first dimension is dropped due to this being a Tsd object.
tensor = tsd.to_trial_tensor(ep)
print(tensor, "\n")
print("Tensor shape = ", tensor.shape)
[[ 0. 0.09983342 0.19866933 0.29552021 0.38941834 0.47942554
0.56464247 0.64421769 0.71735609 0.78332691 0.84147098 nan
nan nan nan nan nan nan
nan nan nan]
[ 0.14112001 0.04158066 -0.05837414 -0.15774569 -0.2555411 -0.35078323
-0.44252044 -0.52983614 -0.61185789 -0.68776616 -0.7568025 -0.81827711
-0.87157577 -0.91616594 -0.95160207 -0.97753012 -0.993691 -0.99992326
-0.99616461 -0.98245261 -0.95892427]]
Tensor shape = (2, 21)
Since trial 2 is twice as long as trial 1, the array is padded with NaNs. The padding value can be changed by setting the parameter padding_value.
tensor = tsd.to_trial_tensor(ep, padding_value=-1)
print(tensor, "\n")
print("Tensor shape = ", tensor.shape)
[[ 0. 0.09983342 0.19866933 0.29552021 0.38941834 0.47942554
0.56464247 0.64421769 0.71735609 0.78332691 0.84147098 -1.
-1. -1. -1. -1. -1. -1.
-1. -1. -1. ]
[ 0.14112001 0.04158066 -0.05837414 -0.15774569 -0.2555411 -0.35078323
-0.44252044 -0.52983614 -0.61185789 -0.68776616 -0.7568025 -0.81827711
-0.87157577 -0.91616594 -0.95160207 -0.97753012 -0.993691 -0.99992326
-0.99616461 -0.98245261 -0.95892427]]
Tensor shape = (2, 21)
By default, time series are aligned to the start of each trial. To align the time series to the end of each trial, the optional parameter align can be set to “end”.
tensor = tsd.to_trial_tensor(ep, align="end")
print(tensor, "\n")
print("Tensor shape = ", tensor.shape)
[[ nan nan nan nan nan nan
nan nan nan nan 0. 0.09983342
0.19866933 0.29552021 0.38941834 0.47942554 0.56464247 0.64421769
0.71735609 0.78332691 0.84147098]
[ 0.14112001 0.04158066 -0.05837414 -0.15774569 -0.2555411 -0.35078323
-0.44252044 -0.52983614 -0.61185789 -0.68776616 -0.7568025 -0.81827711
-0.87157577 -0.91616594 -0.95160207 -0.97753012 -0.993691 -0.99992326
-0.99616461 -0.98245261 -0.95892427]]
Tensor shape = (2, 21)
trial_count#
TsGroup and Ts objects each have the method trial_count, which builds a trial-based count tensor from an IntervalSet object. Similar to count, this function requires a bin_size parameter which determines the number of time bins within each trial. The resulting tensor has shape (number of group elements, number of trials, number of time bins) for TsGroup objects, or (number of trials, number of time bins) for Ts objects.
tensor = tsgroup.trial_count(ep, bin_size=1)
print(tensor, "\n")
print("Tensor shape = ", tensor.shape)
[[[ 0. 0. 0. 0. 0. 0. 1. 1. 0. 0. nan nan nan nan nan nan nan
nan nan nan]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.
0. 0. 0.]]
[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. nan nan nan nan nan nan nan
nan nan nan]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0.]]
[[ 0. 0. 0. 1. 0. 0. 0. 0. 0. 1. nan nan nan nan nan nan nan
nan nan nan]
[ 0. 0. 1. 0. 0. 0. 0. 1. 2. 1. 0. 0. 0. 1. 0. 0. 1.
0. 0. 1.]]]
Tensor shape = (3, 2, 20)
Similar to to_trial_tensor, the array is padded with NaNs when the trials have uneven durations, where the padding value can be controled using the parameter padding_value. Additionally, the parameter align can change whether the count is aligned to the “start” or “end” of each trial.
tensor = tsgroup.trial_count(ep, bin_size=1, align="end", padding_value=-1)
print(tensor, "\n")
print("Tensor shape = ", tensor.shape)
[[[-1. -1. -1. -1. -1. -1. -1. -1. -1. -1. 0. 0. 0. 0. 0. 0. 1.
1. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.
0. 0. 0.]]
[[-1. -1. -1. -1. -1. -1. -1. -1. -1. -1. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0.]]
[[-1. -1. -1. -1. -1. -1. -1. -1. -1. -1. 0. 0. 0. 1. 0. 0. 0.
0. 0. 1.]
[ 0. 0. 1. 0. 0. 0. 0. 1. 2. 1. 0. 0. 0. 1. 0. 0. 1.
0. 0. 1.]]]
Tensor shape = (3, 2, 20)
Mapping between TsGroup and Tsd#
It’s is possible to transform a TsGroup to Tsd with the method
to_tsd and a Tsd to TsGroup with the method to_tsgroup.
This is useful to flatten the activity of a population in a single array.
tsd = tsgroup.to_tsd()
print(tsd)
Time (s)
------------ --
3.649925309 2
6.949448636 0
7.48259419 0
9.28212446 2
12.576363887 1
13.101409676 2
14.007385153 2
...
88.474534193 1
88.964630795 1
93.821826978 2
94.204925384 0
94.970708699 2
96.332807592 2
99.352586223 2
dtype: float64, shape: (60,)
The object tsd contains all the timestamps of the tsgroup with
the associated value being the index of the unit in the TsGroup.
The method to_tsgroup converts the Tsd object back to the original TsGroup.
back_to_tsgroup = tsd.to_tsgroup()
print(back_to_tsgroup)
Index rate
------- -------
0 0.10449
1 0.20898
2 0.31347
Parameterizing a raster#
The method to_tsd makes it easier to display a raster plot.
TsGroup object can be plotted with plt.plot(tsgroup.to_tsd(), 'o').
Timestamps can be mapped to any values passed directly to the method
or by giving the name of a specific metadata name of the TsGroup.
tsgroup['label'] = np.arange(3)*np.pi
print(tsgroup)
Index rate label
------- ------- -------
0 0.10449 0
1 0.20898 3.14159
2 0.31347 6.28319
Show code cell source
plt.figure()
plt.subplot(2,2,1)
plt.plot(tsgroup.to_tsd(), '|')
plt.title("tsgroup.to_tsd()")
plt.xlabel("Time (s)")
plt.subplot(2,2,2)
plt.plot(tsgroup.to_tsd([10,20,30]), '|')
plt.title("togroup.to_tsd([10,20,30])")
plt.xlabel("Time (s)")
plt.subplot(2,2,3)
plt.plot(tsgroup.to_tsd("label"), '|')
plt.title("togroup.to_tsd('label')")
plt.xlabel("Time (s)")
plt.tight_layout()
plt.show()
Special slicing : TsdFrame#
For users that are familiar with pandas, TsdFrame is the closest object to a DataFrame.
but there are distinctive behavior when slicing the object. TsdFrame behaves primarily like a numpy array. This section
lists all the possible ways of slicing TsdFrame.
1. If not column labels are passed#
tsdframe = nap.TsdFrame(t=np.arange(4), d=np.random.randn(4,3))
print(tsdframe)
Time (s) 0 1 2
---------- -------- -------- --------
0 0.67235 0.49356 1.07342
1 -0.31564 0.89974 -1.66425
2 -0.5559 0.02566 -0.28388
3 0.64425 -3.31971 -0.1555
dtype: float64, shape: (4, 3)
Slicing should be done like numpy array :
tsdframe[0]
array([0.67235197, 0.49355787, 1.07341533])
tsdframe[:, 1]
Time (s)
---------- ---------
0 0.493558
1 0.899737
2 0.025657
3 -3.31971
dtype: float64, shape: (4,)
tsdframe[:, [0, 2]]
Time (s) 0 2
---------- -------- --------
0 0.67235 1.07342
1 -0.31564 -1.66425
2 -0.5559 -0.28388
3 0.64425 -0.1555
dtype: float64, shape: (4, 2)
2. If column labels are passed as integers#
The typical case is channel mapping. The order of the columns on disk are different from the order of the columns on the recording device it corresponds to.
tsdframe = nap.TsdFrame(t=np.arange(4), d=np.random.randn(4,4), columns = [3, 2, 0, 1])
print(tsdframe)
Time (s) 3 2 0 1
---------- -------- -------- -------- --------
0 1.4782 -1.06293 0.88022 1.58468
1 1.31308 1.94066 -0.21492 0.994
2 -0.35018 1.40097 0.55563 -0.88606
3 1.01834 -1.61953 -0.40403 -0.44337
dtype: float64, shape: (4, 4)
In this case, indexing like numpy still has priority which can led to confusing behavior :
tsdframe[:, [0, 2]]
Time (s) 3 0
---------- -------- --------
0 1.4782 0.88022
1 1.31308 -0.21492
2 -0.35018 0.55563
3 1.01834 -0.40403
dtype: float64, shape: (4, 2)
Note how this corresponds to column labels 3 and 0.
To slice using column labels only, the TsdFrame object has the loc method similar to Pandas :
tsdframe.loc[[0, 2]]
Time (s) 0 2
---------- -------- --------
0 0.88022 -1.06293
1 -0.21492 1.94066
2 0.55563 1.40097
3 -0.40403 -1.61953
dtype: float64, shape: (4, 2)
In this case, this corresponds to columns labelled 0 and 2.
3. If column labels are passed as strings#
Similar to Pandas, it is possible to label columns using strings.
tsdframe = nap.TsdFrame(t=np.arange(4), d=np.random.randn(4,3), columns = ["kiwi", "banana", "tomato"])
print(tsdframe)
Time (s) kiwi banana tomato
---------- -------- -------- --------
0 0.12016 0.93234 0.89293
1 -1.05345 0.15376 1.08414
2 0.20945 0.35254 0.22174
3 -0.21919 -0.23143 -1.19852
dtype: float64, shape: (4, 3)
When the column labels are all strings, it is possible to use either direct bracket indexing or using the loc method:
print(tsdframe['kiwi'])
print(tsdframe.loc['kiwi'])
Time (s)
---------- ---------
0 0.120164
1 -1.05345
2 0.209447
3 -0.219186
dtype: float64, shape: (4,)
Time (s)
---------- ---------
0 0.120164
1 -1.05345
2 0.209447
3 -0.219186
dtype: float64, shape: (4,)
4. If column labels are mixed type#
It is possible to mix types in column names.
tsdframe = nap.TsdFrame(t=np.arange(4), d=np.random.randn(4,3), columns = ["kiwi", 0, np.pi])
print(tsdframe)
Time (s) kiwi 0 3.141592653589793
---------- -------- -------- -------------------
0 -0.5738 -0.12272 -0.06678
1 1.2474 1.19624 -0.88244
2 1.0075 -1.10065 0.4872
3 0.87666 0.62203 -0.06102
dtype: float64, shape: (4, 3)
Direct bracket indexing only works if the column label is a string.
print(tsdframe['kiwi'])
Time (s)
---------- ---------
0 -0.573799
1 1.2474
2 1.0075
3 0.876664
dtype: float64, shape: (4,)
To slice with mixed types, it is best to use the loc method :
print(tsdframe.loc[['kiwi', np.pi]])
Time (s) kiwi 3.141592653589793
---------- -------- -------------------
0 -0.5738 -0.06678
1 1.2474 -0.88244
2 1.0075 0.4872
3 0.87666 -0.06102
dtype: float64, shape: (4, 2)
In general, it is probably a bad idea to mix types when labelling columns.
Interval sets methods#
Interaction between epochs#
epoch1 = nap.IntervalSet(start=0, end=10) # no time units passed. Default is us.
epoch2 = nap.IntervalSet(start=[5, 30], end=[20, 45])
print(epoch1, "\n")
print(epoch2, "\n")
index start end
0 0 10
shape: (1, 2), time unit: sec.
index start end
0 5 20
1 30 45
shape: (2, 2), time unit: sec.
union#
epoch = epoch1.union(epoch2)
print(epoch)
index start end
0 0 20
1 30 45
shape: (2, 2), time unit: sec.
intersect#
epoch = epoch1.intersect(epoch2)
print(epoch)
index start end
0 5 10
shape: (1, 2), time unit: sec.
set_diff#
epoch = epoch1.set_diff(epoch2)
print(epoch)
index start end
0 0 5
shape: (1, 2), time unit: sec.
split#
Useful for chunking time series, the split method splits an IntervalSet in a new
IntervalSet based on the interval_size argument.
epoch = nap.IntervalSet(start=0, end=100)
print(epoch.split(10, time_units="s"))
index start end
0 0 10
1 10 20
2 20 30
3 30 40
4 40 50
5 50 60
6 60 70
7 70 80
8 80 90
9 90 100
shape: (10, 2), time unit: sec.
Drop intervals#
epoch = nap.IntervalSet(start=[5, 30], end=[6, 45])
print(epoch)
index start end
0 5 6
1 30 45
shape: (2, 2), time unit: sec.
drop_short_intervals#
print(
epoch.drop_short_intervals(threshold=5)
)
index start end
0 30 45
shape: (1, 2), time unit: sec.
drop_long_intervals#
print(
epoch.drop_long_intervals(threshold=5)
)
index start end
0 5 6
shape: (1, 2), time unit: sec.
merge_close_intervals#
Show code cell source
epoch = nap.IntervalSet(start=[1, 7], end=[6, 45])
print(epoch)
index start end
0 1 6
1 7 45
shape: (2, 2), time unit: sec.
If two intervals are closer than the threshold argument, they are merged.
print(
epoch.merge_close_intervals(threshold=2.0)
)
index start end
0 1 45
shape: (1, 2), time unit: sec.