ogrisel

import numpy as np
import pytest
from sklearn.datasets import load_breast_cancer
from sklearn.utils import shuffle
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import roc_auc_score, roc_curve

ogrisel

@pytest.mark.parametrize("loss", ['huber', 'ls', 'lad', 'quantile'])
@pytest.mark.parametrize("use_sample_weight", [False, True])
def test_regressor_train_loss_convergence(loss, use_sample_weight):
    rng = np.random.RandomState(42)
    n_samples, n_features = 30, 5
    n_estimators = 300

    # Make random data (without duplicated samples) to make sure
    # it's possible to build an invertible (overfitting) mapping
    # from X to y that therefore should lead to a regression loss

ogrisel

(conda-forge-compilers) 0 [~/code/scikit-learn (master)]$ pip install -e . -v
Created temporary directory: /private/var/folders/69/7jxl92h50w10b4v998qt4tj00000gn/T/pip-ephem-wheel-cache-cn0u3xn5
Created temporary directory: /private/var/folders/69/7jxl92h50w10b4v998qt4tj00000gn/T/pip-req-tracker-7xtixh31
Created requirements tracker '/private/var/folders/69/7jxl92h50w10b4v998qt4tj00000gn/T/pip-req-tracker-7xtixh31'
Created temporary directory: /private/var/folders/69/7jxl92h50w10b4v998qt4tj00000gn/T/pip-install-q8mggn78
Obtaining file:///Users/ogrisel/code/scikit-learn
  Added file:///Users/ogrisel/code/scikit-learn to build tracker '/private/var/folders/69/7jxl92h50w10b4v998qt4tj00000gn/T/pip-req-tracker-7xtixh31'
    Running setup.py (path:/Users/ogrisel/code/scikit-learn/setup.py) egg_info for package from file:///Users/ogrisel/code/scikit-learn
    Running command python setup.py egg_info
    running egg_info

ogrisel

from time import time
from pprint import pprint

import numpy as np
import pandas as pd
from scipy.stats import expon, randint, uniform

from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OrdinalEncoder

ogrisel

ogrisel

Starting Microsoft Python language server.
##########Linting Output - flake8##########
Microsoft Python Language Server version 0.1.75.0
Initializing for /opt/venvs/py37/bin/python
Loading files from /home/ogrisel/code/scikit-learn
Parsing document file:///home/ogrisel/code/scikit-learn/setup.py
Parse complete for file:///home/ogrisel/code/scikit-learn/setup.py at version -1
Analysis queued for file:///home/ogrisel/code/scikit-learn/setup.py
Parsing document file:///home/ogrisel/code/scikit-learn/conftest.py
Parse complete for file:///home/ogrisel/code/scikit-learn/conftest.py at version -1

ogrisel

"""Empirical evaluation of the extended feature Gram matrix of a ReLU MLP

Here we try to estimate the spectrum of the H^\infty matrix as defined in:

Gradient Descent Provably Optimizes Over-parameterized Neural Networks (2018)
Simon S. Du, Xiyu Zhai, Barnabas Poczos, Aarti Singh
https://arxiv.org/abs/1810.02054

Theorem 4.1 relies on the assumption that H^\infty has a strictly positive
minimum eigenvalue. The following computes an estimate of this eigenvalue

ogrisel

from sklearn.datasets import make_blobs
from sklearn.cluster import KMeans
from sklearn.externals import joblib


m = joblib.Memory(cachedir='/tmp/joblib')
make_blobs = m.cache(make_blobs)
data, labels = make_blobs(n_samples=10**5, n_features=50, cluster_std=100,
                          centers=10, random_state=777)

ogrisel

from pickle import Pickler, load

try:
    from pickle import PickleBuffer
except ImportError:
    PickleBuffer = None
import copyreg
import os
import numpy as np
import time

ogrisel

from pickle import Pickler, _Pickler, Unpickler, _Unpickler, HIGHEST_PROTOCOL
import os
import time
import sys
import gc
from multiprocessing import get_context

PROTOCOL = HIGHEST_PROTOCOL

ctx = get_context('spawn')