How can I statistically compare these lines?
I've been struggling with this for weeks, but I know its a basic question and just don't know the correct terms to google.
I basically , I have several ‘lines’ on a line chart, each representing what is known as a ‘spectral profile’ for different plant species.
What I am hoping to output, is some sort of quantitative value, that can outline how ‘unique’ these profiles are from each other.
I've been told PCA but that isn't making any fucking sense.
Unique in what way? Shape or the dependent variable?
>>8855125
Dependent - I'm trying to numerically explain that the grey one is much more different than the yellow, etc...
>>8855111
It looks like a hat.
Did you know this?
How do you compare hats?
Normalize one of the functions and represent your data with respect to that normalized curve.
Then there's clear representation of contrast between the curves.
But in this case you should really pick a wavelength (show units please) and compare the reflectance at that wavelength for all species.
PCA is fine but you generally want higher dimensionality than this dataset for good PCA.
imo you should do what >>8855146 said
so you have lines for datasets 0 through 10, just normalize every line to be data[i]-data[0] and call em deviation datasets.
you could also do root-mean-square deviation to get a scalar value for each series. lets say you have data = [[data0],[data1],...,[data10]], pythonically RMSD would be:
rmsd = []
for series in data:
rmsd.append(average((series - data[0])**2.)**0.5)
or any other 'deviation' feature you can think of
>>8855111
the interaction term in a two-way ANOVA might be relevant
>>8855111
Float some centroids, there isn't many lines so two or three might do
like so
>>8855111
You can compare the derivatives of the graphs after you normalize the curves and then take the integral difference of the curves divided by the interval you're analyzing
there you'd get a quantitative measure for difference
>>8855111
Learn to type you degenerate autist
>>8855111
take the mean of the set of spectral profiles.
find the L2 norm of this mean
take the variance too, find the L2 norm of that
take the L2 norm of all spectral profiles individually
plot the results and compare to a gaussian distribution
you're welcome OP
>>8856963
Thank you too!
Is there any websites that will autistically walk me through some of these? I'm getting to the point where i'm about to bribe some cheap Indian to do it for me. Argh...