LDDMM: PLS regression#

Uses data from LDDMM: how to estimate a deterministic atlas?.

TODO: add purpose of notebook

[1]:

from pathlib import Path

import numpy as np
from sklearn.cross_decomposition import PLSRegression
from sklearn.model_selection import train_test_split

import polpo.lddmm as plddmm
import polpo.preprocessing.pd as ppd
from polpo.model_eval import (
    R2Score,
    collect_eval_results,
)
from polpo.preprocessing.load.pregnancy.jacobs import TabularDataLoader
from polpo.preprocessing.np import FlattenButFirst
from polpo.sklearn.adapter import EvaluatedModel

No CUDA runtime is found, using CUDA_HOME='/usr'

[2]:

DATA_DIR = Path("results") / "atlas_example"
ATLAS_DIR = DATA_DIR / "atlas"

[3]:

cp = plddmm.io.load_cp(ATLAS_DIR)
momenta = plddmm.io.load_momenta(ATLAS_DIR)

momenta = FlattenButFirst()(momenta)

cp.shape, momenta.shape

[3]:

((24, 3), (26, 72))

[4]:

data = TabularDataLoader(subject_subset=["01"], index_by_session=True)()

INFO: Data has already been downloaded... using cached file ('/home/luisfpereira/.herbrain/data/maternal/maternal_brain_project_pilot/rawdata/28Baby_Hormones.csv').

[5]:

data_ = ppd.ColumnToDict("gestWeek")(data)

[6]:

x = np.array(list(data_.values()))[:, None]
y = momenta

x.shape, y.shape

[6]:

((26, 1), (26, 72))

[7]:

model = EvaluatedModel(
    PLSRegression(),
    R2Score(),
)

[8]:

X_train, X_test, y_train, y_test = train_test_split(
    y,
    x,
    train_size=0.8,
    shuffle=False,
)


model.fit(X_train, y_train);

[9]:

eval_res_train = collect_eval_results(model, unnest=True, outer_key="obj_regr")

print(list(eval_res_train.keys()))

['obj_regr']

[10]:

eval_res_train

[10]:

{'obj_regr': {'r2': array([0.71588517])}}

[11]:

model.predict_eval(X_test, y_test);

[12]:

eval_res_test = collect_eval_results(
    model, unnest=True, outer_key="obj_regr", train=False
)

print(list(eval_res_test.keys()))

['obj_regr']

[13]:

eval_res_test

[13]:

{'obj_regr': {'r2': array([-2.32858353])}}