def test_ranking_short():
parsed = None
file = "data/cacmShort.txt"
for path in ["./", "../"]:
try:
parsed = myParser.buildDocCollectionSimple(path + file, ".T")
break
except FileNotFoundError:
pass
assert parsed
indexer = indexerSimple.IndexerSimple(parsed.docs)
requete = "home computer microphotographi"
models = [weighter.c1, weighter.c2, weighter.c3, weighter.c4, weighter.c5]
models = [clas(indexer) for clas in models]
models = [vectoriel.Vectoriel(indexer, weight, False) for weight in models]
models.append(jelinekMercer.JelinekMercer(indexer, 1))
models.append(okapiBM25.OkapiBM25(indexer, 1.2, .75))
rankings = [m.getRanking(requete) for m in models]
# 5 docs ont un score non nul qqsoit modèle
for ranking in rankings:
assert len(ranking) == 5
# ordre des résultats
for ranking in rankings[:-1]:
assert [x[0] for x in ranking] == ["7", "6", "4", "2", "10"]
assert [x[0] for x in rankings[-1]] == ["6", "7", "4", "10", "2"]
def testVeryShort():
docs = [
"the new home has been saled on top forecasts",
"the home sales rise in july",
"there is an increase in home sales in july",
"july encounter a new home sales rise"
]
parsed = myParser.loadCollection(docs)
indexer = indexerSimple.IndexerSimple(parsed.docs)
for d in indexer.ind, indexer.inv, indexer.ind_n, indexer.inv_n:
for e in d.items():
print(e)
print("\n")
def testShort():
parsed = None
parsed2 = None
file = "data/cacmShort.txt"
for path in ["./", "../"]:
try:
parsed = myParser.buildDocCollectionSimple(path + file)
parsed2 = myParser.buildDocumentCollectionRegex(path + file)
break
except FileNotFoundError:
pass
assert parsed and parsed2
# équivalence des deux méthodes de parsing
for d1, d2 in zip(sorted(parsed.docs.values(), key=lambda x: x.I),
sorted(parsed2.docs.values(), key=lambda x: x.I)):
assert d1.I == d2.I
assert d1.T == d2.T
indexer = indexerSimple.IndexerSimple(parsed2.docs)
assert 'algebra' in indexer.ind['1']
assert len(indexer.ind['2']) == 6
assert sum(indexer.ind['11'].values()) == 8
assert 'algebra' in indexer.ind_n['1']
assert abs(sum(indexer.ind_n['2'].values()) - 1) < 1e-4
assert indexer.inv['matrix'] == {'3': 1}
assert len(indexer.inv['comput']) == 5
assert indexer.inv_n['matrix'] == {'3': .2}
assert len(indexer.inv_n['comput']) == 5
tf_idf = indexer.create_tf_idf()
#tfidf à la même structure que ind
assert tf_idf.keys() == indexer.ind.keys()
for i_doc in tf_idf.keys():
assert tf_idf[i_doc].keys() == indexer.ind[i_doc].keys()
#contenu de tfidf
assert abs(tf_idf['4']['programm'] - 0.875) < 1e-3
def testLong():
print("test long")
print("lecture")
parsed = None
file = "data/cisi/cisi.txt"
for path in ["./", "../"]:
try:
parsed = myParser.buildDocCollectionSimple(path + file, '.W')
break
except FileNotFoundError:
pass
assert parsed
print("création index")
indexer = indexerSimple.IndexerSimple(parsed.docs)
assert len(indexer.ind) == 2459
print("création tfidf")
tf_idf = indexer.create_tf_idf()
assert len(tf_idf) == 2459
def main():
test_ranking_veryshort()
test_ranking_short()
test_ranking_long()
parsed = None
# exécution d'un modèle simple
file = "data/cacm/cacm.txt"
for path in ["./", "../"]:
try:
parsed = myParser.buildDocCollectionSimple(path + file,
".T",
balise2=".X")
break
except FileNotFoundError:
pass
assert parsed
indexer = indexerSimple.IndexerSimple(parsed.docs)
test_full_pagerank(indexer)
test_alea_pagerank(indexer)
def test_ranking_veryshort():
docs = [
"the new home has home been saled on top forecasts",
"the home sales rise in july", "there is an increase in sales in july",
"july encounter a new home sales rise"
]
parsed = myParser.loadCollection(docs)
indexer = indexerSimple.IndexerSimple(parsed.docs)
weights_doc = []
scores = []
for clas in [
weighter.c1, weighter.c2, weighter.c3, weighter.c4, weighter.c5
]:
w = clas(indexer)
weights_doc.append(w.getWeightsForDoc(0))
v = vectoriel.Vectoriel(indexer, w, False)
scores.append(v.getScores("home sales top"))
#documents dans le même ordre qqsoit weighter
list0 = sorted(scores[0].keys(), key=lambda x: scores[0][x])
for s in scores[1:]:
assert list0 == sorted(s.keys(), key=lambda x: s[x])
def test_ranking_long():
parsed = None
file = "data/cacm/cacm.txt"
for path in ["./", "../"]:
try:
parsed = myParser.buildDocCollectionSimple(path + file, ".T")
break
except FileNotFoundError:
pass
assert parsed
indexer = indexerSimple.IndexerSimple(parsed.docs)
requete = "home computer microphotographi"
models = [weighter.c1, weighter.c2, weighter.c3, weighter.c4, weighter.c5]
models = [clas(indexer) for clas in models]
models = [vectoriel.Vectoriel(indexer, weight, False) for weight in models]
models.append(jelinekMercer.JelinekMercer(indexer, .2))
models.append(okapiBM25.OkapiBM25(indexer, 1.2, .75))
rankings = [m.getRanking(requete) for m in models]
#modèle O rang 0 à un score de 2
assert rankings[0][0][1] == 2
#modèle O rang 9 à un score de 1
assert rankings[0][9][1] == 1
#modèle 1
assert rankings[1][0][1] == 2
assert rankings[1][9][1] == 1
#meilleur docs
assert rankings[0][0][0] == "80"
assert rankings[1][0][0] == "80"
assert rankings[2][0][0] == "3646"
assert rankings[3][0][0] == "3646"
assert rankings[4][0][0] == "80"
assert rankings[5][0][0] == "866"
assert rankings[6][0][0] == "3156"
def testLong():
parsedQuery = None
parsedText = None
file = "data/cisi/cisi"
for path in ["./", "../"]:
try:
parsedQuery = queryParser.parse(path + file)
parsedText = myParser.buildDocCollectionSimple(
path + file + ".txt", ".W")
break
except FileNotFoundError:
pass
assert parsedQuery
assert parsedText
assert len(parsedQuery.queries) == 112
assert len(parsedQuery.queries["1"].pertient_list_id) == 46
#nombre de requêtes ayant au moins un document pertinent
assert sum(
len(q.pertient_list_id) > 0
for q in parsedQuery.queries.values()) == 76
print("calcul indexer")
indexer = indexerSimple.IndexerSimple(parsedText.docs)
models = [weighter.c1, weighter.c2, weighter.c3, weighter.c4, weighter.c5]
models = [clas(indexer) for clas in models]
models = [vectoriel.Vectoriel(indexer, weight, False) for weight in models]
jelinek = jelinekMercer.JelinekMercer(indexer)
models.append(jelinek)
okapi = okapiBM25.OkapiBM25(indexer)
models.append(okapi)
data_fit = [q.T for q in parsedQuery.queries.values()]
labels = [q.pertient_list_id for q in parsedQuery.queries.values()]
print("fit")
# jelinek.fit(np.linspace(0, 2, 2), data_fit, labels)
# okapi.fit((np.linspace(0, 2, 2), np.linspace(0, 2, 2)), data_fit, labels)
# train test
print(len(data_fit))
n = 100
jelinek.fit(np.linspace(.2, .7, 3), data_fit[:n], labels[:n])
okapi.fit((np.linspace(0, 2, 2), np.linspace(0, 2, 2)), data_fit[:n],
labels[:n])
for i in range(len(models)):
models.append(pagerank.PagerankMarcheAlea(indexer, models[i]))
models[-2].fit(np.linspace(.2, .7, 3), data_fit[:n], labels[:n])
print("précisions")
for m in models:
pred = [m.getRanking(d) for d in data_fit[n:]]
avgPrec = 0
for p, l in zip(pred, labels[n:]):
avgPrec += m.avgPrec(p, l)
print(m, avgPrec / len(pred))
def eval():
parsedQuery = None
parsedText = None
file = "data/cisi/cisi"
for path in ["./", "../"]:
try:
parsedQuery = queryParser.parse(path + file)
parsedText = myParser.buildDocCollectionSimple(path + file + ".txt", ".W")
break
except FileNotFoundError:
pass
indexer = indexerSimple.IndexerSimple(parsedText.docs)
models = [weighter.c1, weighter.c2, weighter.c3, weighter.c4, weighter.c5]
models = [clas(indexer) for clas in models]
models = [vectoriel.Vectoriel(indexer, weight, False) for weight in models]
jelinek = jelinekMercer.JelinekMercer(indexer)
models.append(jelinek)
okapi = okapiBM25.OkapiBM25(indexer)
models.append(okapi)
data_fit = [q.T for q in parsedQuery.queries.values()]
labels = [q.pertient_list_id for q in parsedQuery.queries.values()]
print("fit")
# jelinek.fit(np.linspace(0, 2, 2), data_fit, labels)
# okapi.fit((np.linspace(0, 2, 2), np.linspace(0, 2, 2)), data_fit, labels)
for i in range(len(models)):
models.append(pagerank.PagerankMarcheAlea(indexer, models[i]))
k = 9
metrics = [
averagePrecision.AveragePrecision(),
precisionAtK.PrecisionAtK(k),
fMesureK.FMesureK(1, k),
rappelAtK.RappelAtK(k),
NDCG.NDCG(k),
reciprocalRank.ReciprocalRank()]
perf = []
print(models)
print(metrics)
for i, model in enumerate(models):
print(i,"/", len(models))
perf.append([])
pred = [model.getRanking(data_fit[k]) for k in range(len(data_fit))]
for metric in metrics:
score, std = metric.eval_list_query(pred, labels)
perf[-1].append(score)
print([round(x, 4) for x in perf[-1]])
import matplotlib.pyplot as plt
plt.imshow(perf)
plt.colorbar()
plt.xlabel("Metrique")
plt.ylabel("Modèle")
plt.figtext(0,0,"Metriques : 0 : averagePrecision,1 : precisionAtK,2 : fMesureK,3 : rappelAtK,4 : NDCG,5 : reciprocalRank;Modèles : 0-4: Vectoriel, 5 : jelinekMercer,6 : okapiBM25, 7-12 : avec pagerank")
plt.show()