Feature/fpca_regression

[Ddelval]

Add the FPCA regression estimator

[codecov]

Codecov Report

Patch coverage: 99.01% and project coverage change: +0.11 🎉

Comparison is base (48d1fae) 85.51% compared to head (5ab79ca) 85.63%.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #466      +/-   ##
===========================================
+ Coverage    85.51%   85.63%   +0.11%     
===========================================
  Files          141      143       +2     
  Lines        11284    11384     +100     
===========================================
+ Hits          9650     9749      +99     
- Misses        1634     1635       +1     

Impacted Files	Coverage Δ
skfda/tests/test_fpca_regression.py	98.14% <98.14%> (ø)
...da/misc/operators/_linear_differential_operator.py	95.91% <100.00%> (+0.17%)	⬆️
skfda/ml/regression/_fpca_regression.py	100.00% <100.00%> (ø)
skfda/representation/basis/_custom_basis.py	86.25% <100.00%> (+0.53%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[vnmabus]

I think this functionality is implemented in the package fda.usc. We should have a test verifying that we obtain the same results.

[Ddelval]

I think this functionality is implemented in the package fda.usc. We should have a test verifying that we obtain the same results.

I have (finally) added this test. However, while going back and forth between python and R, I stumbled upon a python packet that provides a simple way to execute R code from within python. I am sceptical regarding whether or not this could be useful for the tests. On the one hand, it would ensure that the tests are testing against the correct values while forcing us to include the R source used to obtain the test reference data. On the other hand, it would make the tests considerably slower and require modifications to the GitHub actions.

I think the timing tradeoff is too significant (I measured a slowdown of around 6x). Nevertheless, I thought I might just mention it just in case it comes in handy at some point. The following snippet generates the testing data for the test I just added.

from rpy2 import robjects

Rresult = robjects.r('''
library("fda.usc")
data(tecator)
# Fit the regression model with the first 129 observations
x=tecator$absorp.fdata[1:129,]
y=tecator$y$Fat[1:129]
res2=fregre.pc(x,y,l=1:10)

# Predict the response for the remaining observations
n = length(tecator$y$Fat)
xnew=tecator$absorp.fdata[130:n,]
result = predict(res2, xnew)
names(result) = NULL

# Output the predicted values
paste(
    round(result,8),
    collapse = ", "
)
''')
r_predictions = np.array(Rresult[0].split(", "), dtype=np.float64)

[vnmabus]

Remember to also add the class to the API reference documentation.

[vnmabus]

Were you able to compare results with refund too?

[Ddelval]

This is the issue that I opened: refunders/refund#102.
It doesn't seem like we'll be getting any solutions from their side anytime soon.

[vnmabus]

Why is this specialized for FDataBasis?

[Ddelval]

Leftover from the previous implementation. Removed now.