def _run(parameters):
print(parameters)
dataset_params, label_params = parameters
label, params = label_params
dataset = logistic.PyDataset(**dataset_params)
dataset.load()
model = logistic.PygisticSGD(**params)
print('run model {}'.format(label))
res = model.fit(dataset)
print('done model {}'.format(label))
return label, res
import os
import logistic
import pickle
import numpy as np
directory = "results"
if not os.path.exists(directory):
os.makedirs(directory)
# create datasets
madelon = logistic.PyDataset("madelon", "/Users/jb/data/madelon.txt", 2000,
500, False)
rcv1 = logistic.PyDataset("rcv1", "/mlodata1/jb/data/rcv1_train.binary", 20242,
47236, True)
rcv1_test = logistic.PyDataset("rcv1_test",
"/mlodata1/jb/data/rcv1_test.binary", 677399,
47236, True)
epsilon = logistic.PyDataset("epsilon", "/mlodata1/jb/data/epsilon_normalized",
400000, 2000, True)
# """
# RCV1-test theory
# """
# if not rcv1_test.is_loaded():
# rcv1_test.load()
#
# num_samples = 677399
# num_features = 47236
# dataset = rcv1_test
# dataset_name = "RCV1"
# file = "rcv1-th.pickle"
#
# CREATE PARAMS FOR DATASET
#
dataset_params = {
"name": "rcv1_test",
"inputFile": "/mlodata1/jb/data/rcv1_test.binary",
"numSamples": 677399,
"numFeatures": 47236,
"is_sparse": True
}
num_features = dataset_params["numFeatures"]
num_samples = dataset_params["numSamples"]
dataset = logistic.PyDataset(**dataset_params)
#
# CREATE PARAMS FOR EXPERIMENT
#
common_params = {
"numEpochs": 20,
"lrType": "optimal",
"lr": 1.,
"tau": 10 * num_features,
"lambda_": 1. / num_samples,
"printPerEpoch": 20,
"weightingScheme": 'final',
}
from sklearn.linear_model import SGDClassifier
from sklearn.datasets import load_svmlight_file
import time
import numpy as np
repeat = 3
# #RVV1
# rcv1_test = logistic.PyDataset("rcv1_test", "/mlodata1/jb/data/rcv1_test.binary", 677399, 47236, True)
# svm_path = "/mlodata1/jb/data/rcv1_test.binary"
# dataset = rcv1_test
# num_samples = 677399
# print("RUN ON RCV1")
#epsilon
epsilon = logistic.PyDataset("epsilon", "/mlodata1/jb/data/epsilon_normalized",
400000, 2000, True)
svm_path = "/mlodata1/jb/data/epsilon_normalized"
dataset = epsilon
num_samples = 400000
print("RUN ON epsilon")
dataset.load()
res = np.zeros(repeat)
for i in range(repeat):
model = logistic.PygisticSGD(numEpochs=1,
lrType="bottou",
lr=.1,
tau=1.,
lambda_=1. / num_samples,
weightingScheme='final',