ptlasso 0.3.4

Pretrained Lasso: a two-step procedure for sparse linear models with grouped samples

ptlasso

Python implementation of the Pretrained Lasso — a two-step procedure for fitting sparse linear models when samples belong to distinct groups, leveraging shared structure across groups via pretraining.

Based on:

Craig, E., Pilanci, M., Le Menestrel, T., Narasimhan, B., Rivas, M. A., Gullaksen, S. E., ... & Tibshirani, R. (2025). Pretraining and the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology, qkaf050.

---

The idea

Standard group-specific Lasso models are fit independently per group, ignoring shared signal. The Pretrained Lasso fits in two steps:

Step 1 — Overall model. Fit a Lasso on all samples to capture shared structure:

$$\hat{\beta}^{\text{overall}} = \arg\min_\beta \frac{1}{2n}|y - X\beta|^2 + \lambda|\beta|_1$$

Step 2 — Group models. For each group $k$, fit a Lasso with an offset equal to $\alpha$ times the overall model's linear predictor:

$$\hat{\beta}^{(k)} = \arg\min_\beta \frac{1}{2n_k}|y^{(k)} - \underbrace{\alpha \cdot X^{(k)}\hat{\beta}^{\text{overall}}}_{\text{offset}} - X^{(k)}\beta|^2 + \lambda_k|\beta|_1$$

The parameter $\alpha \in [0, 1]$ controls the pretraining strength:

• $\alpha = 0$: pure group-specific models, no pretraining
• $\alpha = 1$: group models explain residuals from the overall model
• $\alpha \in (0, 1)$: group models are anchored to the overall fit

Final prediction for group $k$: $\hat{y}^{(k)} = \alpha \cdot X^{(k)}\hat{\beta}^{\text{overall}} + X^{(k)}\hat{\beta}^{(k)}$

Supports gaussian, binomial, and multinomial families.

---

Installation

pip install ptlasso

Requires Python ≥ 3.9 and adelie for the underlying Lasso solver, which supports fitting with offsets (unlike scikit-learn).

---

Quick start

import numpy as np
from ptlasso import PretrainedLasso, PretrainedLassoCV

rng = np.random.default_rng(42)
n, p, k = 300, 100, 3

X      = rng.standard_normal((n, p))
groups = rng.integers(0, k, size=n)
beta   = np.zeros(p)
beta[:5] = [2, -1.5, 1, -0.8, 0.5]
y      = X @ beta + 0.5 * rng.standard_normal(n)

# Fixed alpha
model = PretrainedLasso(alpha=0.5)
model.fit(X, y, groups)
print(model)
# PretrainedLasso(alpha=0.5, family='gaussian', overall_lambda='lambda.1se', ...)
#   family       : gaussian
#   n_features   : 100
#   n_groups     : 3
#   overall |Ŝ|  : |Ŝ| = 5 / 100  [0, 1, 2, 3, 4]
#   pretrain |Ŝ| : 0: |Ŝ|=5, 1: |Ŝ|=4, 2: |Ŝ|=6

y_pred = model.predict(X, groups)
print("R²:", model.score(X, y, groups))

# Cross-validate over alpha
cv = PretrainedLassoCV(alphas=[0.0, 0.25, 0.5, 0.75, 1.0])
cv.fit(X, y, groups)
print("Best alpha:", cv.alpha_)

---

Families

# Binary classification
model = PretrainedLasso(alpha=0.5, family="binomial")
model.fit(X, y_binary, groups)
probs = model.predict(X, groups)          # shape (n,), P(y=1)

# Multi-class classification (integer labels 0..K-1)
model = PretrainedLasso(alpha=0.5, family="multinomial")
model.fit(X, y_multiclass, groups)
probs = model.predict(X, groups)          # shape (n, K)

---

Feature names and group labels

Both fit() methods accept human-readable names. pandas DataFrames are supported natively — column names are picked up automatically.

import pandas as pd

X_df         = pd.DataFrame(X, columns=[f"gene_{i}" for i in range(p)])
group_labels = {0: "control", 1: "treated_A", 2: "treated_B"}

model = PretrainedLasso(alpha=0.5)
model.fit(X_df, y, groups, group_labels=group_labels)
# overall |Ŝ|  : |Ŝ| = 5 / 100  [gene_0, gene_1, gene_2, gene_3, gene_4]
# pretrain |Ŝ| : control: |Ŝ|=5, treated_A: |Ŝ|=4, treated_B: |Ŝ|=6

---

Inspecting the support

The support is the set of non-zero variables selected by the model, i.e., the features whose coefficients are not shrunk to zero by the L1 regularization and therefore actively contribute to the prediction.

from ptlasso import (
    get_overall_support,
    get_pretrain_support,
    get_pretrain_support_split,
    get_individual_support,
)

get_overall_support(model)                      # features from the overall model
get_pretrain_support(model)                     # union across pretrained group models
get_pretrain_support(model, common_only=True)   # features selected by >50% of groups
get_pretrain_support(model, groups=[0, 1])      # restrict to specific groups
get_individual_support(model)                   # features from no-pretraining baselines

common, indiv = get_pretrain_support_split(model)
# common : features from the overall model (stage 1)
# indiv  : additional features picked up by group models (stage 2)

---

Evaluating all sub-models at once

result = model.evaluate(X_test, y_test, groups_test)
# {"pretrain":   {"predictions": ..., "score": ...},
#  "individual": {"predictions": ..., "score": ...},
#  "overall":    {"predictions": ..., "score": ...}}

---

Retrieving coefficients

coefs = model.get_coef()                   # all sub-models
coefs["overall"]                           # {"coef": ndarray, "intercept": ndarray}
coefs["pretrain"]["control"]               # {"coef": ndarray, "intercept": ndarray}
coefs["individual"]["treated_A"]

model.get_coef(model="pretrain")           # just pretrain sub-dict

---

CV details

cv = PretrainedLassoCV(
    alphas=[0.0, 0.25, 0.5, 0.75, 1.0],
    cv=5,
    alphahat_choice="overall",   # or "mean" (unweighted mean of per-group CV errors)
    family="gaussian",
    overall_lambda="lambda.1se", # or "lambda.min"
    foldid=my_foldid,            # optional: custom integer fold assignments
)
cv.fit(X, y, groups)

cv.alpha_                        # globally best alpha
cv.varying_alphahat_             # {group: best_alpha} per group
cv.cv_results_                   # {alpha: mean CV loss}
cv.cv_results_se_                # {alpha: SE of CV loss}
cv.cv_results_per_group_         # {alpha: {group: mean CV loss}}
cv.cv_results_mean_              # {alpha: unweighted mean of per-group losses}
cv.cv_results_wtd_mean_          # {alpha: size-weighted mean of per-group losses}
cv.cv_results_individual_        # CV loss for individual (no-pretraining) baseline
cv.cv_results_overall_           # CV loss for overall model baseline
cv.best_estimator_               # PretrainedLasso fitted with alpha_
cv.all_estimators_               # {alpha: PretrainedLasso} for varying-alpha prediction

# Predict using each group's own best alpha
cv.predict(X, groups, alphatype="varying")
cv.evaluate(X, y, groups, alphatype="varying")

---

Plotting

from ptlasso import plot_cv, plot_paths

plot_cv(cv)           # CV loss curve over alpha with ±1 SE band
plot_paths(model)     # regularisation paths for all sub-models

---

Saving and loading models

PretrainedLasso and PretrainedLassoCV can be serialised with joblib:

import joblib

# Save
joblib.dump(model, "model.pkl")

# Load
model = joblib.load("model.pkl")

Note: serialised models are tied to the Python and library versions used to create them. For long-term storage, pin your dependencies or retrain from scratch after major version upgrades.

Avoid pickle directly — the underlying adelie solver stores C++ objects that are not natively picklable. ptlasso handles this transparently through joblib by converting solver state to plain numpy arrays before serialisation. Using pickle directly bypasses this and will raise a TypeError.

---

API reference

PretrainedLasso

Parameter	Default	Description
alpha	0.5	Pretraining strength $\in [0, 1]$
family	"gaussian"	"gaussian", "binomial", or "multinomial"
overall_lambda	"lambda.1se"	Lambda rule for stage-1 offset: "lambda.1se" or "lambda.min"
fit_intercept	True	Fit an intercept in all sub-models
lmda_path_size	100	Number of $\lambda$ values in the regularisation path
min_ratio	0.01	Ratio of smallest to largest $\lambda$
verbose	False	Show adelie progress bar

Methods:

• fit(X, y, groups, group_labels=None, feature_names=None)
• predict(X, groups, model="pretrain", lmda_idx=None) — model ∈ {"pretrain", "individual", "overall"}
• score(X, y, groups) — R² or accuracy
• evaluate(X, y, groups) — predict + score for all three sub-models
• get_coef(model="all", lmda_idx=None)

PretrainedLassoCV

Parameter	Default	Description
alphas	[0, 0.25, 0.5, 0.75, 1.0]	Candidate $\alpha$ values
cv	5	Number of CV folds
alphahat_choice	"overall"	"overall" or "mean" (unweighted per-group mean)
family	"gaussian"	Same as PretrainedLasso
overall_lambda	"lambda.1se"	Same as PretrainedLasso
fit_intercept	True
lmda_path_size	100
min_ratio	0.01
verbose	False
foldid	None	Integer array of fold assignments (overrides cv)

Same fit / predict / score / evaluate / get_coef interface as PretrainedLasso, plus:

Fitted attribute	Description
alpha_	Best $\alpha$ selected by CV
varying_alphahat_	{group: alpha} — per-group best $\alpha$
cv_results_	{alpha: mean CV loss}
cv_results_se_	{alpha: SE of CV loss}
cv_results_per_group_	{alpha: {group: mean CV loss}}
cv_results_mean_	{alpha: unweighted mean of per-group losses}
cv_results_wtd_mean_	{alpha: size-weighted mean of per-group losses}
cv_results_individual_	CV loss for individual baseline
cv_results_overall_	CV loss for overall baseline
best_estimator_	PretrainedLasso fitted with alpha_
all_estimators_	{alpha: PretrainedLasso} for each unique varying alpha

predict also accepts alphatype="varying" to route each group through its own best alpha.

---

Citation

@article{craig2025pretraining,
  title   = {Pretraining and the lasso},
  author  = {Craig, Erin and Pilanci, Mert and Le Menestrel, Thomas and Narasimhan, Balasubramanian and Rivas, Manuel A. and Gullaksen, Stein-Erik and Tibshirani, Robert},
  journal = {Journal of the Royal Statistical Society Series B: Statistical Methodology},
  pages   = {qkaf050},
  year    = {2025}
}

---

License

MIT