Making new timeseries channels from running averages

FreyaS

Hi,

I've run a data reduction scheme (3D trace elements) on my laser ablation data. From this I got trace elements (TE) in ppm, which I then used to calculate the TE/Ca ratio. I did this for all the different TE and they now exist as channels. One of these is Ba135/Ca. My plan was to make selections based on values over a certain threshold but given how much variation there is, I see that will mean me losing a lot of data. So I instead planned to make a running average for the Ba135/Ca. However after playing around a lot with the python control, I'm just getting long lists of errors so something is clearly going wrong. Is there a way (either via the interface or python control), to make a running average from the data in one channel, and then import it into a new channel, so it can be viewed in time series?

Thanks!

Freya

Bence

Hi Freya,

You could do this via the Calculator (in the Channels menu). I would create a new channel of smoothed values. To do that, in the Calculator create a new channel called "Ba/Ca_smoothed". The expression part would be:

convolve(Ba135_Ca, ones(5,"d") / 5.0,  mode="valid")

This will apply a five point smooth. If you wanted to change the number of points smoothed over, change the "5" in the ones() call, and in the divisor "/ 5.0". For example, to create a 10 point smooth, use this:

convolve(Ba135_Ca, ones(10,"d") / 10.0,  mode="valid")

Note that in the Calculator, you have to remove "/" characters and replace them with an underscore, hence changing "Ba135/Ca" to "Ba135_Ca" above.

If you have any troubles with that, please let me know.

Kind regards,
Bence

FreyaS

Hi Bence, thanks that works great! Is ther eany documentation for the different mathematical formulations that work in the Calculator tool, or is it just following python nomenclature?

Cheers,

Freya

Bence

Hi Freya,

You basically have access to anything in Numpy (v1.21.0), but you also have access to anything in iolite's main python interface (i.e. anything starting with data. in the python interface).

If you have anything in particular you need, please let us know.

Kind regards,
Bence

FreyaS

Hi Bence,

Just following up calculations in the time series - supposing you wanted to work out rate of change along a time series? I can't figure out that there is a way to call it up in the Calculations tool as the 'time' variable does not appear to be defined in a way I can call up. I also tried writing it in the python script, but then I was struggling to call up my time series channel that I wanted to calculate the rate of change for.

All the best,

Freya

Bence

Hi Freya,
You can't access the time arrays via the calculator at this point, but we'll look at adding it in the future. Instead, you can do this via a script (e.g. in the Python Workspace).

You can use diff() to work out rates of change on a channels values.

To get time values from a channel, use the time() function, like so:

# Get time series object:
ts = data.timeSeries('Pb206')
d = ts.data() # the measurements
t = ts.time() # the time stamp of each measurement
d_diff = np.diff(d)
t_diff = np.diff(t)

The last line would give you the time difference in seconds between each measurement, if you wanted to normalise to this (if there are significant differences between measurement intervals, this might be handy).

Please let me know how it goes.

Kind regards,
Bence

FreyaS

Hi Bence,

Thanks for the help, I've gotten as far as calculating rate of chance for the times series of interest. Is it possible to now reopen the calculated rate of change in the Time Series window as a new channel that be looked at and worked on?

Best regards,

Freya

import numpy as np
ts = data.timeSeries('Ca43_CPS')
d = ts.data()
t = ts.time()
d_diff = np.diff(d)
t_diff = np.diff(t)

rate_of_change = d_diff / t_diff

Joe

For anyone else that ends up here, this conversation continued via email, but the answer to this question was to add the following lines:

rate_of_change = np.insert(rate_of_change, 0, 0)
data.createTimeSeries('Rate of change', data.Intermediate, None, rate_of_change)

When doing np.diff, you will lose 1 data point, but we want the diff to be on the same time axis as the other channels. So the first line just adds back a data point (value of 0 at index 0). The second line creates a time series object in iolite with the name 'Rate of change' of type Intermediate. The 3rd argument, None, tells iolite to create it using the default time array.

A follow up question was how to adjust selections based on the rate of change. One way you could do that was with a script as follows:

import numpy as np
from scipy.ndimage import gaussian_filter1d
from iolite.QtCore import QDateTime

group_to_adjust = 'Otolith'
channel_to_check = 'Ca43'

ts = data.timeSeries(channel_to_check)
d = gaussian_filter1d(ts.data(), sigma=4)

t = ts.time()
d_diff = np.diff(d)
t_diff = np.diff(t)

rate_of_change = d_diff / t_diff
rate_of_change = np.insert(rate_of_change, 0, 0)
rate_ts = data.createTimeSeries('Rate of change', data.Intermediate, None, rate_of_change)

for sel in data.selectionGroup(group_to_adjust).selections():
     min_index = np.argmin(rate_ts.dataForSelection(sel))
     time_min = rate_ts.timeForSelection(sel)[min_index]
     sel.endTime = QDateTime.fromMSecsSinceEpoch(time_min*1000)