Add SHAP explainability analysis

.gitattributes

@@ -65,3 +65,9 @@ output_v2/figures/hinglish_to_hindi_to_english/Phase_hinglish_eval_full_roc.png
 output_v2/figures/hinglish_to_hindi_to_english/Phase_hinglish_eval_hindi_roc.png filter=lfs diff=lfs merge=lfs -text
 output_v2/figures/hinglish_to_hindi_to_english/Phase_hinglish_eval_hinglish_pr.png filter=lfs diff=lfs merge=lfs -text
 output_v2/figures/hinglish_to_hindi_to_english/Phase_hinglish_eval_hinglish_roc.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_english.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_full.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_hinglish.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/cross_model_comparison_full.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/cross_model_comparison_hindi.png filter=lfs diff=lfs merge=lfs -text
+output_v2/shap/cross_model_comparison_hinglish.png filter=lfs diff=lfs merge=lfs -text

output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_summary.csv

@@ -0,0 +1,5 @@
+model,eval_lang,top_hate_words,top_nonhate_words
+hinglish_to_hindi_to_english_v2,english,"nas(0.0408), fags(0.0308), sicko(0.0297), sabotage(0.0297), advocating(0.0223)","grow(-0.0357), barrel(-0.0341), homosexual(-0.0323), pak(-0.0301), join(-0.0241)"
+hinglish_to_hindi_to_english_v2,hindi,"वादा(0.0033), वैज्ञानिकों(0.0033), ऐ(0.0032), उतारा(0.0032), गला(0.0030)","जीतेगा(-0.0072), घोंटने(-0.0042), जिहादी(-0.0039), आपत्तिजनक(-0.0039), चमचो(-0.0024)"
+hinglish_to_hindi_to_english_v2,hinglish,"arey(0.0459), bahir(0.0409), punish(0.0396), papa(0.0389), interior(0.0373)","online(-0.0293), member(-0.0265), mam(-0.0230), messages(-0.0221), asha(-0.0204)"
+hinglish_to_hindi_to_english_v2,full,"blamed(0.0648), criticized(0.0473), syntax(0.0467), grown(0.0392), sine(0.0352)","underneath(-0.1046), smack(-0.0426), online(-0.0403), hole(-0.0354), clue(-0.0314)"

output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_english.csv

@@ -0,0 +1,51 @@
+word,mean_shap,direction,abs_shap
+nas,0.04084094986319542,hate,0.04084094986319542
+grow,-0.03568753972649574,non-hate,0.03568753972649574
+barrel,-0.03411834314465523,non-hate,0.03411834314465523
+homosexual,-0.032309435307979584,non-hate,0.032309435307979584
+fags,0.030772147700190544,hate,0.030772147700190544
+pak,-0.030134739354252815,non-hate,0.030134739354252815
+sicko,0.029741348698735237,hate,0.029741348698735237
+sabotage,0.029711255803704262,hate,0.029711255803704262
+join,-0.02413080632686615,non-hate,0.02413080632686615
+advocating,0.02234148420393467,hate,0.02234148420393467
+poppy,-0.02190038375556469,non-hate,0.02190038375556469
+alien,-0.021285340189933777,non-hate,0.021285340189933777
+cycle,0.0199173204600811,hate,0.0199173204600811
+illegal,0.019460361175636837,hate,0.019460361175636837
+crush,-0.01917188661172986,non-hate,0.01917188661172986
+tit,0.01897701807320118,hate,0.01897701807320118
+spoilt,0.017174284905195236,hate,0.017174284905195236
+accused,0.016549019142985344,hate,0.016549019142985344
+bigger,-0.016319772228598595,non-hate,0.016319772228598595
+fault,0.016076564012716215,hate,0.016076564012716215
+kad,0.01592048816382885,hate,0.01592048816382885
+social,-0.015574524737894535,non-hate,0.015574524737894535
+excuses,0.015325771644711494,hate,0.015325771644711494
+polite,-0.015105613352109989,non-hate,0.015105613352109989
+proverbial,0.014887955971062183,hate,0.014887955971062183
+caught,0.01482851430773735,hate,0.01482851430773735
+failed,0.014297339133918285,hate,0.014297339133918285
+revising,-0.014272860251367092,non-hate,0.014272860251367092
+boomer,-0.013116286136209965,non-hate,0.013116286136209965
+wear,-0.012863598763942719,non-hate,0.012863598763942719
+plot,-0.012858100235462189,non-hate,0.012858100235462189
+vandalize,0.01270614832174033,hate,0.01270614832174033
+unite,0.012388435192406178,hate,0.012388435192406178
+bitchy,0.012348399497568607,hate,0.012348399497568607
+perform,-0.012242045724633499,non-hate,0.012242045724633499
+pride,0.011979740811511874,hate,0.011979740811511874
+stop,0.011946490097146194,hate,0.011946490097146194
+btw,0.011738581582903862,hate,0.011738581582903862
+romania,0.011656457558274269,hate,0.011656457558274269
+boon,0.011625840328633785,hate,0.011625840328633785
+deny,0.011531622149050236,hate,0.011531622149050236
+trans,0.011500706896185875,hate,0.011500706896185875
+dear,-0.011394963517280607,non-hate,0.011394963517280607
+wwf,0.011149724945425987,hate,0.011149724945425987
+wikis,-0.011066155973821878,non-hate,0.011066155973821878
+bro,-0.011030015535652637,non-hate,0.011030015535652637
+rains,-0.010711784474551678,non-hate,0.010711784474551678
+attitude,0.01068491442129016,hate,0.01068491442129016
+hump,-0.01064093317836523,non-hate,0.01064093317836523
+medicare,-0.010626320727169514,non-hate,0.010626320727169514

output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_full.csv

@@ -0,0 +1,51 @@
+word,mean_shap,direction,abs_shap
+underneath,-0.10458830744028091,non-hate,0.10458830744028091
+blamed,0.06481243669986725,hate,0.06481243669986725
+criticized,0.047305628657341,hate,0.047305628657341
+syntax,0.046665407717227936,hate,0.046665407717227936
+smack,-0.042638324201107025,non-hate,0.042638324201107025
+online,-0.040302298963069916,non-hate,0.040302298963069916
+grown,0.039168521761894226,hate,0.039168521761894226
+hole,-0.035353163606487215,non-hate,0.035353163606487215
+sine,0.03520340472459793,hate,0.03520340472459793
+clue,-0.03141946718096733,non-hate,0.03141946718096733
+eliza,-0.025371160358190536,non-hate,0.025371160358190536
+fucked,0.02449769526720047,hate,0.02449769526720047
+ruling,0.02396106906235218,hate,0.02396106906235218
+expressive,0.023163892328739166,hate,0.023163892328739166
+advocating,0.02261391095817089,hate,0.02261391095817089
+manners,-0.022611092776060104,non-hate,0.022611092776060104
+madam,0.022388344630599022,hate,0.022388344630599022
+tweets,-0.0218605634290725,non-hate,0.0218605634290725
+pr,0.02108542900532484,hate,0.02108542900532484
+nawaz,0.019971828907728195,hate,0.019971828907728195
+fateful,0.01936914399266243,hate,0.01936914399266243
+congrats,0.019144877791404724,hate,0.019144877791404724
+flip,0.01885046996176243,hate,0.01885046996176243
+bitter,0.018671734258532524,hate,0.018671734258532524
+dubious,0.016950108110904694,hate,0.016950108110904694
+quran,-0.016676608473062515,non-hate,0.016676608473062515
+turkey,0.016496842727065086,hate,0.016496842727065086
+flop,0.015149658545851707,hate,0.015149658545851707
+punish,0.014967915951274335,hate,0.014967915951274335
+rosewood,-0.014851009473204613,non-hate,0.014851009473204613
+exclude,0.01474262960255146,hate,0.01474262960255146
+tha,0.014648808864876628,hate,0.014648808864876628
+troubles,0.014084530994296074,hate,0.014084530994296074
+attacks,0.013815064914524555,hate,0.013815064914524555
+angela,-0.013330272398889065,non-hate,0.013330272398889065
+bot,-0.01317788427695632,non-hate,0.01317788427695632
+isko,-0.012968616560101509,non-hate,0.012968616560101509
+cretin,-0.012747019529342651,non-hate,0.012747019529342651
+du,0.012732747942209244,hate,0.012732747942209244
+awful,0.012715496122837067,hate,0.012715496122837067
+ashamed,-0.012626993469893932,non-hate,0.012626993469893932
+ive,0.012543770484626293,hate,0.012543770484626293
+gand,0.012430441565811634,hate,0.012430441565811634
+wing,0.01240457221865654,hate,0.01240457221865654
+absolutely,0.012251339852809906,hate,0.012251339852809906
+easily,-0.012238632887601852,non-hate,0.012238632887601852
+ancestral,-0.011794610880315304,non-hate,0.011794610880315304
+bjp,0.011773284524679184,hate,0.011773284524679184
+lete,0.011766272597014904,hate,0.011766272597014904
+ace,0.01174088753759861,hate,0.01174088753759861

output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_hindi.csv

@@ -0,0 +1,51 @@
+word,mean_shap,direction,abs_shap
+जीतेगा,-0.007239299826323986,non-hate,0.007239299826323986
+घोंटने,-0.004180442541837692,non-hate,0.004180442541837692
+जिहादी,-0.003943843767046928,non-hate,0.003943843767046928
+आपत्तिजनक,-0.0038822039496153593,non-hate,0.0038822039496153593
+वादा,0.0033049401827156544,hate,0.0033049401827156544
+वैज्ञानिकों,0.003294713096693158,hate,0.003294713096693158
+ऐ,0.0032312911935150623,hate,0.0032312911935150623
+उतारा,0.0031968937255442142,hate,0.0031968937255442142
+गला,0.003006579587236047,hate,0.003006579587236047
+बेवक़ूफ़,0.002906812820583582,hate,0.002906812820583582
+पत्रकारिता,0.0027609646786004305,hate,0.0027609646786004305
+रद्दीके,0.0026735278079286218,hate,0.0026735278079286218
+शहरों,0.002631198847666383,hate,0.002631198847666383
+जब्त,0.002548573538661003,hate,0.002548573538661003
+बेहोश,0.0025251752231270075,hate,0.0025251752231270075
+भाव,0.002492323750630021,hate,0.002492323750630021
+टुडे,0.0024582596961408854,hate,0.0024582596961408854
+चमचो,-0.002437782706692815,non-hate,0.002437782706692815
+मुश्किलें,0.002396735828369856,hate,0.002396735828369856
+वर्षों,0.002379495883360505,hate,0.002379495883360505
+हालांकि,-0.0023559594992548227,non-hate,0.0023559594992548227
+लागत,0.0023375283926725388,hate,0.0023375283926725388
+सलाम,0.002286830684170127,hate,0.002286830684170127
+ग्राउंड,0.002279318403452635,hate,0.002279318403452635
+वर्तमान,0.0022609676234424114,hate,0.0022609676234424114
+भागे,-0.0022224474232643843,non-hate,0.0022224474232643843
+दौड़ेगी,-0.002178625902161002,non-hate,0.002178625902161002
+विभिन्न,0.0021422391291707754,hate,0.0021422391291707754
+लाठीडंडे,-0.0020988776814192533,non-hate,0.0020988776814192533
+भागवत,-0.002050387905910611,non-hate,0.002050387905910611
+पार,0.0020433420886547538,hate,0.0020433420886547538
+बुर,0.002026310656219721,hate,0.002026310656219721
+तमाशा,0.0020122036803513765,hate,0.0020122036803513765
+सजा,0.0019943982360928203,hate,0.0019943982360928203
+आदित्य,0.0019323243759572506,hate,0.0019323243759572506
+कहे,0.0018306594574823976,hate,0.0018306594574823976
+बबाकस,0.0018216012977063656,hate,0.0018216012977063656
+चेंज,0.0018008995102718472,hate,0.0018008995102718472
+रेपिस्ट,-0.0017585812602192163,non-hate,0.0017585812602192163
+गले,0.0017552308272570372,hate,0.0017552308272570372
+वाणिज्य,0.0017530020559206605,hate,0.0017530020559206605
+बयान,0.001748441020026803,hate,0.001748441020026803
+और,0.0017154300702309979,hate,0.0017154300702309979
+ज़िंदगी,0.0016978256171569228,hate,0.0016978256171569228
+सपा,0.001689926371909678,hate,0.001689926371909678
+उज्ज्वल,0.0016891625709831715,hate,0.0016891625709831715
+बढ़कर,0.001684919698163867,hate,0.001684919698163867
+सैलरी,0.0016782438615337014,hate,0.0016782438615337014
+सालाना,0.0016701605636626482,hate,0.0016701605636626482
+अलावा,0.0016669516408001073,hate,0.0016669516408001073

output_v2/shap/v2/hinglish_to_hindi_to_english_v2/shap_topwords_hinglish.csv

@@ -0,0 +1,51 @@
+word,mean_shap,direction,abs_shap
+arey,0.045903529971838,hate,0.045903529971838
+bahir,0.04093165323138237,hate,0.04093165323138237
+punish,0.03960862010717392,hate,0.03960862010717392
+papa,0.038925547152757645,hate,0.038925547152757645
+interior,0.03734187036752701,hate,0.03734187036752701
+race,0.03409697115421295,hate,0.03409697115421295
+chori,0.033206719905138016,hate,0.033206719905138016
+online,-0.02929110825061798,non-hate,0.02929110825061798
+kren,0.02906891517341137,hate,0.02906891517341137
+choro,0.02657279558479786,hate,0.02657279558479786
+member,-0.026523131877183914,non-hate,0.026523131877183914
+aad,0.026521164923906326,hate,0.026521164923906326
+highlight,0.025127805769443512,hate,0.025127805769443512
+bout,0.025043504312634468,hate,0.025043504312634468
+chalte,0.023972313851118088,hate,0.023972313851118088
+abduction,0.023200901225209236,hate,0.023200901225209236
+mam,-0.02300681360065937,non-hate,0.02300681360065937
+messages,-0.022140955552458763,non-hate,0.022140955552458763
+laya,0.02126455120742321,hate,0.02126455120742321
+srk,0.020947826730359036,hate,0.020947826730359036
+asha,-0.02039911597967148,non-hate,0.02039911597967148
+altaf,0.01923272293061018,hate,0.01923272293061018
+boss,-0.01831447333097458,non-hate,0.01831447333097458
+2018,0.017717858776450157,hate,0.017717858776450157
+etna,-0.017587222158908844,non-hate,0.017587222158908844
+geo,0.01754508726298809,hate,0.01754508726298809
+post,-0.016837802055912714,non-hate,0.016837802055912714
+sapna,0.016813437920063734,hate,0.016813437920063734
+mint,-0.015777967870235443,non-hate,0.015777967870235443
+bech,-0.015203659422695637,non-hate,0.015203659422695637
+condemn,0.014873333275318146,hate,0.014873333275318146
+she,0.01479670312255621,hate,0.01479670312255621
+dehli,-0.014155546203255653,non-hate,0.014155546203255653
+bangladeshi,0.01362423226237297,hate,0.01362423226237297
+penis,-0.013533218763768673,non-hate,0.013533218763768673
+dont,-0.01309139747172594,non-hate,0.01309139747172594
+banana,0.012709609232842922,hate,0.012709609232842922
+kah,0.012406427413225174,hate,0.012406427413225174
+sar,-0.012206245155539364,non-hate,0.012206245155539364
+deke,-0.012147174216806889,non-hate,0.012147174216806889
+bin,-0.012057552114129066,non-hate,0.012057552114129066
+ar,-0.011831492185592651,non-hate,0.011831492185592651
+your,-0.011781544308178127,non-hate,0.011781544308178127
+pr,0.011247894560386027,hate,0.011247894560386027
+prediction,-0.011225917376577854,non-hate,0.011225917376577854
+raj,-0.011192606762051582,non-hate,0.011192606762051582
+rajya,-0.010959833860397339,non-hate,0.010959833860397339
+ak,0.010884546441957355,hate,0.010884546441957355
+wohi,-0.010823655873537064,non-hate,0.010823655873537064
+72,-0.010761625715531409,non-hate,0.010761625715531409