TME RI TEXTUEL

from Index import Index
from ParserCACM import ParserCACM
from TextRepresenter import PorterStemmer
from Weighter import Binary, TF, TF_IDF, Log, Log_plus
from EvalMeasure import EvalIRModel
import numpy as np
import os.path
import time

# model_type = Random | Vectoriel | Okapi | Language | PageRank | Hits | MetaModel
fname = "data/cacm/cacm.txt"
query_file = "data/cacm/cacm.qry"
relevance_file = "data/cacm/cacm.rel"

Random

t1 = time.time()
type = "Random" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
<class 'IRmodel.RandomModel'>
Evaluation of our models ...



Model :  Random

AP:  0.00489365936907
Exec duration(s) :  37.8984589577

Vectoriel

t1 = time.time()
type = "Vectoriel" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
Creating models...Done!
<class 'IRmodel.Vectoriel'>
Evaluation of our models ...



Model :   Binary

AP:  0.123849729889


Model :   Text Frequency

AP:  0.13215026906


Model :   Text Frequency - Inverse Document Frequency

AP:  0.154529641961


Model :   Log

AP:  0.120869800911


Model :   Log +

AP:  0.143917859729
Exec duration(s) :  171.119966984

Okapi : k1=2, b=0.75

t1 = time.time()
type = "Okapi" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
L moy :  4925.14700285
<class 'IRmodel.Okapi'>
Evaluation of our models ...



Model :  Okapi

AP:  0.264364808856
Exec duration(s) :  540.410473108

Language : Lambda =0.2

t1 = time.time()
type = "Language" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
Init of Language model
<class 'IRmodel.LanguageModel'>
Evaluation of our models ...



Model :  Language Model

AP:  0.307058115209
Exec duration(s) :  580.641823053

PageRank : N=25, K=unlimited, d=0.85

t1 = time.time()
type = "PageRank" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
<class 'IRmodel.RankModel'>
Evaluation of our models ...



Model :  PageRank

AP:  0.133320317291
Exec duration(s) :  1294.32549

Hits : N=25, K=unlimited

type = "Hits" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()

MetaModel

t1 = time.time()
type = "MetaModel" 
eval_platform = EvalIRModel(fname,query_file,relevance_file,model_type=type)
models_recall, models_inter_prec, models_AP = eval_platform.eval_std()
print "Exec duration(s) : ",time.time()-t1

Indexing database...Done!
Grad theta : 0.2
Training achieved with Grad_theta <  0.1  !
Number of queries required : 4
<class 'IRmodel.MetaModel'>
Evaluation of our models ...



Model :  MetaModel

AP:  0.285386850473
Exec duration(s) :  2393.91500187

for idx,f in enumerate(eval_platform.models[0].listFeaturers.listFeaturers):    
    print "featurers :",str(f.model)[:-25],">"
    print "theta :",eval_platform.models[0].theta[idx],"\n"

featurers : <IRmodel.Vectoriel  >
theta : [ 0.01994691] 

featurers : <IRmodel.Vectoriel  >
theta : [ 0.88962339] 

featurers : <IRmodel.Vectoriel  >
theta : [ 0.06195642] 

featurers : <IRmodel.Vectoriel  >
theta : [ 0.20448804] 

featurers : <IRmodel.Okapi  >
theta : [ 1.82525156] 

featurers : <IRmodel.LanguageModel  >
theta : [ 1.11311955] 

featurers : <IRmodel.RankModel  >
theta : [ 0.74651456]