Precomputed Features

Incorporating Precomputed Features into GLM Design

In some cases, your data may contain features that are not directly computable through the available bases in NeMoS, for instance a Principal Component Analysis (PCA) of high-dimensional signals. The IdentityEval basis allows you to incorporate these precomputed features into the model design, and combine them with other predictors through basis composition.

For example, let’s assume that you want to compute a Principal Component Analysis (PCA) of some signals, and use the first 2 Principal Components as GLM features.

Currently, NeMoS doesn’t provide a PCA basis, but you can compute the Principal Components in sklearn.

import numpy as np
from sklearn.decomposition import PCA

n_samples = 100
n_signals = 10

# generate some random signals
high_dim_singals = np.random.randn(n_samples, n_signals)

# generate some counts
counts = np.random.poisson(size=n_samples)

# compute the first 2 pcs
pcs = PCA(2).fit_transform(high_dim_singals)

Now, let’s see how to use the IdentityEval basis to model jointly the PCs and a spike history filter.

import nemos as nmo

# create a composite basis 
pc_basis = nmo.basis.IdentityEval()
history_basis = nmo.basis.RaisedCosineLogConv(3, window_size=10)
composite_basis = pc_basis + history_basis

# create the model design
X = composite_basis.compute_features(pcs, counts)

print(f"Design matrix shape: {X.shape}")

# fit the glm
model = nmo.glm.GLM().fit(X, counts)

/home/docs/checkouts/readthedocs.org/user_builds/nemos/envs/stable/lib/python3.11/site-packages/nemos/basis/_basis_mixin.py:326: DeprecationWarning: `newshape` keyword argument is deprecated, use `shape=...` or pass shape positionally instead. (deprecated in NumPy 2.1)
  return np.reshape(conv, newshape=(conv.shape[0], -1))

Design matrix shape: (100, 5)