import gradio as gr from time import time from scipy import sparse from scipy import linalg from sklearn.datasets import make_regression from sklearn.linear_model import Lasso def load_dataset(): X, y = make_regression(n_samples=200, n_features=5000, random_state=0) # create a copy of X in sparse format X_sp = sparse.coo_matrix(X) return X,X_sp,y def compare_lasso_dense(): alpha_dense = 1 alpha_sparse = 0.1 sparse_lasso = Lasso(alpha= alpha_sparse, fit_intercept=False, max_iter=1000) dense_lasso = Lasso(alpha=alpha_dense, fit_intercept=False, max_iter=1000) t0 = time() sparse_lasso.fit(X_sp, y) # print(f"Sparse Lasso done in {(time() - t0):.3f}s") elapse1 = time() - t0 t1 = time() dense_lasso.fit(X, y) # print(f"Dense Lasso done in {(time() - t0):.3f}s") elapse2 = time() - t1 # compare the regression coefficients coeff_diff = linalg.norm(sparse_lasso.coef_ - dense_lasso.coef_) # print(f"Distance between coefficients : {coeff_diff:.2e}") return f"Sparse Lasso done in {(elapse1):.3f}s\t\n" + f"Dense Lasso done in {(elapse2):.3f}s\t\n" + f"Distance between coefficients : {coeff_diff:.2e}\t\n" def compare_lasso_sparse(): # make a copy of the previous data Xs = X.copy() # make Xs sparse by replacing the values lower than 2.5 with 0s Xs[Xs < 2.5] = 0.0 # create a copy of Xs in sparse format Xs_sp = sparse.coo_matrix(Xs) Xs_sp = Xs_sp.tocsc() # compute the proportion of non-zero coefficient in the data matrix print(f"Matrix density : {(Xs_sp.nnz / float(X.size) * 100):.3f}%") matrix_density = Xs_sp.nnz / float(X.size) * 100 alpha_dense = 1 alpha_sparse = 0.1 sparse_lasso = Lasso(alpha= alpha_sparse, fit_intercept=False, max_iter=1000) dense_lasso = Lasso(alpha=alpha_dense, fit_intercept=False, max_iter=1000) t0 = time() sparse_lasso.fit(Xs_sp, y) print(f"Sparse Lasso done in {(time() - t0):.3f}s") elapses1 = time() - t0 t1 = time() dense_lasso.fit(Xs, y) print(f"Dense Lasso done in {(time() - t1):.3f}s") elapses2 = time() - t1 # compare the regression coefficients coeff_diff = linalg.norm(sparse_lasso.coef_ - dense_lasso.coef_) print(f"Distance between coefficients : {coeff_diff:.2e}") return f"Matrix density : {(Xs_sp.nnz / float(X.size) * 100):.3f}%\t\n"+ f"Sparse Lasso done in {(elapses1):.3f}s\t\n" + f"Dense Lasso done in {(elapses2):.3f}s\t\n" + f"Distance between coefficients : {coeff_diff:.2e}\t\n" X,X_sp,y = load_dataset() # compare_lasso_dense(X,X_sp,y) # compare_lasso_sparse(X,X_sp,y) title = " Lasso on Dense and Sparse data " info = '''**Comparing the two Lasso implementations on Dense data** We create a linear regression problem that is suitable for the Lasso, that is to say, with more features than samples. We then store the data matrix in both dense (the usual) and sparse format, and train a Lasso on each. We compute the runtime of both and check that they learned the same model by computing the Euclidean norm of the difference between the coefficients they learned. Because the data is dense, we expect better runtime with a dense data format. ''' info2='''***Comparing the two Lasso implementations on Sparse data*** We make the previous problem sparse by replacing all small values with 0 and run the same comparisons as above. Because the data is now sparse, we expect the implementation that uses the sparse data format to be faster. ''' conclusion = '''**Conclusion** We show that linear_model.Lasso provides the same results for dense and sparse data and that in the case of sparse data the speed is improved**. ''' with gr.Blocks() as demo: gr.Markdown(f"# {title}") gr.Markdown(info) txt_3 = gr.Textbox(value="", label="Dense Lasso comparison") btn = gr.Button(value="Dense Lasso comparison") btn.click(compare_lasso_dense, outputs=[txt_3]) gr.Markdown(info2) txt_4 = gr.Textbox(value="", label="Sparse Lasso comparison") btn = gr.Button(value="Sparse Lasso comparison") btn.click(compare_lasso_sparse, outputs=[txt_4]) gr.Markdown(conclusion) if __name__ == "__main__": demo.launch()