QuaPy/Retrieval/experiments.py

246 lines
9.9 KiB
Python

import os.path
import pickle
from collections import defaultdict
from pathlib import Path
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
from sklearn.svm import LinearSVC
import quapy as qp
from Retrieval.commons import RetrievedSamples, load_sample
from method.non_aggregative import MaximumLikelihoodPrevalenceEstimation as Naive
from quapy.method.aggregative import ClassifyAndCount, EMQ, ACC, PCC, PACC, KDEyML
from quapy.data.base import LabelledCollection
from os.path import join
from tqdm import tqdm
from result_table.src.table import Table
"""
In this sixth experiment, we have a collection C of >6M documents.
We split C in two equally-sized pools TrPool, TePool
I have randomly split the collection in 50% train and 50% split. In each split we have approx. 3.25 million documents.
We have 5 categories we can evaluate over: Continent, Years_Category, Num_Site_Links, Relative Pageviews and Gender.
From the training set I have created smaller subsets for each category:
100K, 500K, 1M and FULL (3.25M)
For each category and subset, I have created a training set called: "classifier_training.json". This is the "base" training set for the classifier. In this set we have 500 documents per group in a category. (For example: Male 500, Female 500, Unknown 500). Let me know if you think we need more.
To "bias" the quantifier towards a query, I have executed the queries (97) on the different training sets and retrieved the 200 most relevant documents per group.
For example: (Male 200, Female 200, Unknown 200)
Sometimes this is infeasible, we should probably discuss this at some point.
You can find the results for every query in a file named:
"training_Query-[QID]Sample-200SPLIT.json"
Test:
To evaluate our approach, I have executed the queries on the test split. You can find the results for all 97 queries up till k=1000 in this file.
testRanking_Results.json
"""
def methods(classifier, class_name):
kde_param = {
'continent': 0.18,
'gender': 0.12,
'years_category':0.09
}
yield ('Naive', Naive())
yield ('NaiveQuery', Naive())
yield ('CC', ClassifyAndCount(classifier))
# yield ('PCC', PCC(classifier))
# yield ('ACC', ACC(classifier, val_split=5, n_jobs=-1))
yield ('PACC', PACC(classifier, val_split=5, n_jobs=-1))
# yield ('EMQ', EMQ(classifier, exact_train_prev=True))
# yield ('EMQ-Platt', EMQ(classifier, exact_train_prev=True, recalib='platt'))
# yield ('EMQh', EMQ(classifier, exact_train_prev=False))
# yield ('EMQ-BCTS', EMQ(classifier, exact_train_prev=True, recalib='bcts'))
# yield ('EMQ-TS', EMQ(classifier, exact_train_prev=False, recalib='ts'))
# yield ('EMQ-NBVS', EMQ(classifier, exact_train_prev=False, recalib='nbvs'))
# yield ('EMQ-VS', EMQ(classifier, exact_train_prev=False, recalib='vs'))
# yield ('KDE001', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.001))
# yield ('KDE005', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.005)) # <-- wow!
# yield ('KDE01', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.01))
# yield ('KDE02', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.02))
# yield ('KDE03', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.03))
# yield ('KDE-silver', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth='silverman'))
# yield ('KDE-scott', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth='scott'))
yield ('KDE-opt', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=kde_param[class_name]))
yield ('KDE01', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.01))
yield ('KDE02', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.02))
yield ('KDE03', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.03))
yield ('KDE04', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.04))
yield ('KDE05', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.05))
yield ('KDE07', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.07))
# yield ('KDE10', KDEyML(classifier, val_split=5, n_jobs=-1, bandwidth=0.10))
def train_classifier(train_path):
"""
Trains a classifier. To do so, it loads the training set, transforms it into a tfidf representation.
The classifier is Logistic Regression, with hyperparameters C (range [0.001, 0.01, ..., 1000]) and
class_weight (range {'balanced', None}) optimized via 5FCV.
:return: the tfidf-vectorizer and the classifier trained
"""
texts, labels = load_sample(train_path, class_name=class_name)
tfidf = TfidfVectorizer(sublinear_tf=True, min_df=3)
Xtr = tfidf.fit_transform(texts)
print(f'Xtr shape={Xtr.shape}')
print('training classifier...', end='')
classifier = LogisticRegression(max_iter=5000)
classifier = GridSearchCV(
classifier,
param_grid={'C': np.logspace(-4, 4, 9), 'class_weight': ['balanced', None]},
n_jobs=-1,
cv=5
)
classifier.fit(Xtr, labels)
classifier = classifier.best_estimator_
classifier_acc = classifier.best_score_
print(f'[done] best-params={classifier.best_params_} got {classifier_acc:.4f} score')
training = LabelledCollection(Xtr, labels)
print('training classes:', training.classes_)
print('training prevalence:', training.prevalence())
return tfidf, classifier
def reduceAtK(data: LabelledCollection, k):
# if k > len(data):
# print(f'[warning] {k=}>{len(data)=}')
X, y = data.Xy
X = X[:k]
y = y[:k]
return LabelledCollection(X, y, classes=data.classes_)
def benchmark_name(class_name, k):
scape_class_name = class_name.replace('_', '\_')
return f'{scape_class_name}@{k}'
def run_experiment():
results = {
'mae': {k: [] for k in Ks},
'mrae': {k: [] for k in Ks}
}
pbar = tqdm(experiment_prot(), total=experiment_prot.total())
for train, test in pbar:
Xtr, ytr, score_tr = train
Xte, yte, score_te = test
if HALF:
n = len(ytr) // 2
train_col = LabelledCollection(Xtr[:n], ytr[:n], classes=classifier_trained.classes_)
else:
train_col = LabelledCollection(Xtr, ytr, classes=classifier_trained.classes_)
if method_name not in ['Naive', 'NaiveQuery']:
quantifier.fit(train_col, val_split=train_col, fit_classifier=False)
elif method_name == 'Naive':
quantifier.fit(train_col)
test_col = LabelledCollection(Xte, yte, classes=classifier_trained.classes_)
for k in Ks:
test_k = reduceAtK(test_col, k)
if method_name == 'NaiveQuery':
train_k = reduceAtK(train_col, k)
quantifier.fit(train_k)
estim_prev = quantifier.quantify(test_k.instances)
mae = qp.error.mae(test_k.prevalence(), estim_prev)
mrae = qp.error.mrae(test_k.prevalence(), estim_prev, eps=(1. / (2 * k)))
results['mae'][k].append(mae)
results['mrae'][k].append(mrae)
pbar.set_description(f'{method_name}')
return results
data_home = 'data'
HALF=True
exp_posfix = '_half'
method_names = [name for name, *other in methods(None, 'continent')]
Ks = [5, 10, 25, 50, 75, 100, 250, 500, 750, 1000]
for class_name in ['gender', 'continent', 'years_category']: # 'relative_pageviews_category', 'num_sitelinks_category']:
tables_mae, tables_mrae = [], []
benchmarks = [benchmark_name(class_name, k) for k in Ks]
for data_size in ['10K', '50K', '100K', '500K', '1M', 'FULL']:
table_mae = Table(name=f'{class_name}-{data_size}-mae', benchmarks=benchmarks, methods=method_names)
table_mrae = Table(name=f'{class_name}-{data_size}-mrae', benchmarks=benchmarks, methods=method_names)
table_mae.format.mean_prec = 5
table_mae.format.remove_zero = True
table_mae.format.color_mode = 'global'
tables_mae.append(table_mae)
tables_mrae.append(table_mrae)
class_home = join(data_home, class_name, data_size)
# train_data_path = join(class_home, 'classifier_training.json')
# classifier_path = join('classifiers', data_size, f'classifier_{class_name}.pkl')
train_data_path = join(data_home, class_name, 'FULL', 'classifier_training.json') # <-------- fixed classifier
classifier_path = join('classifiers', 'FULL', f'classifier_{class_name}.pkl') # <------------ fixed classifier
test_rankings_path = join(data_home, 'testRanking_Results.json')
results_home = join('results'+exp_posfix, class_name, data_size)
tfidf, classifier_trained = qp.util.pickled_resource(classifier_path, train_classifier, train_data_path)
experiment_prot = RetrievedSamples(
class_home,
test_rankings_path,
vectorizer=tfidf,
class_name=class_name,
classes=classifier_trained.classes_
)
for method_name, quantifier in methods(classifier_trained, class_name):
results_path = join(results_home, method_name + '.pkl')
if os.path.exists(results_path):
print(f'Method {method_name=} already computed')
results = pickle.load(open(results_path, 'rb'))
else:
results = run_experiment()
os.makedirs(Path(results_path).parent, exist_ok=True)
pickle.dump(results, open(results_path, 'wb'), pickle.HIGHEST_PROTOCOL)
for k in Ks:
table_mae.add(benchmark=benchmark_name(class_name, k), method=method_name, v=results['mae'][k])
table_mrae.add(benchmark=benchmark_name(class_name, k), method=method_name, v=results['mrae'][k])
# Table.LatexPDF(f'./latex{exp_posfix}/{class_name}{exp_posfix}.pdf', tables=tables_mae+tables_mrae)
Table.LatexPDF(f'./latex{exp_posfix}/{class_name}{exp_posfix}.pdf', tables=tables_mrae)