{ "id": 205856, "title": "Xgboost nan outputs", "url": "/competitions/rfcx-species-audio-detection/discussion/205856", "author_name": "", "post_date": "2020-12-22T07:18:13.986282700Z", "votes": 5, "comment_count": 4, "views": 0, "content": "

I tried to use xgboost in the competition but i think god did not like it .

\n

I tried to predict outputs in probability as i needed them in probability not normal output everywhere there was nan , i used predict function i got all the answers as 0

\n

Note there was no nan in y_train

\n

Please help 🤒

", "messages": [ { "id": "1122087", "postDate": "12/22/2020 07:18:13", "content": "

I tried to use xgboost in the competition but i think god did not like it .

\n

I tried to predict outputs in probability as i needed them in probability not normal output everywhere there was nan , i used predict function i got all the answers as 0

\n

Note there was no nan in y_train

\n

Please help 🤒

", "rawMarkdown": "I tried to use xgboost in the competition but i think god did not like it .\n\nI tried to predict outputs in probability as i needed them in probability not normal output everywhere there was nan , i used predict function i got all the answers as 0\n\nNote there was no nan in y_train \n\nPlease help 🤒", "votes": null }, { "id": "1123835", "postDate": "12/23/2020 14:33:27", "content": "

Hello there,

\n

you can try to use predict_proba instead of predict. predict_proba predicts the probability of each sample, predict simply returns 0 or 1 as the classified label, not the probabilities.

", "rawMarkdown": "Hello there,\n\nyou can try to use `predict_proba` instead of `predict`. `predict_proba` predicts the probability of each sample, `predict` simply returns 0 or 1 as the classified label, not the probabilities.", "votes": null }, { "id": "1123851", "postDate": "12/23/2020 14:42:46", "content": "

Sorry but you completely miss understood it I have used predict_proba all the values are coming nan when I used predict all the values come 0

", "rawMarkdown": "Sorry but you completely miss understood it I have used `predict_proba ` all the values are coming nan when I used `predict ` all the values come 0", "votes": null }, { "id": "1123867", "postDate": "12/23/2020 14:49:19", "content": "

Can you post the code here that causes this behaviour or publish your notebook? Otherwise it's hard to tell what could cause this problem.

", "rawMarkdown": "Can you post the code here that causes this behaviour or publish your notebook? Otherwise it's hard to tell what could cause this problem.", "votes": null }, { "id": "1123971", "postDate": "12/23/2020 16:08:01", "content": "

here is the code :-
\n'''
\nimport xgboost
\ndef get_model(Y):
\n return xgboost.XGBClassifier(n_estimators=1000,
\n max_depth=4,
\n learning_rate=0.05,
\n verbosity=0,
\n objective=\"binary:logistic\",
\n subsample=0.95,
\n colsample_bytree=0.95,
\n random_state=2021,
\n n_jobs=2,
\n scale_pos_weight = np.sum(Y==0) / np.sum(Y==1),
\n )

\n

'''
\n'''
\nfrom sklearn.model_selection import KFold
\nimport pickle
\nt=Y.T
\nmodels=[]
\naverages=[]
\ndef extract(indexes,i):
\n global X
\n global train

\n
s_X=[train[ind] for ind in indexes]\ns_Y=[int(Y[ind][i]) for ind in indexes]\nreturn (s_X,s_Y)\n
\n

kf = KFold(n_splits=3)

\n

for s in range(0,24):
\n temp=[0,None]# accuracy , model
\n ac=[]# scores
\n g=0
\n for ind_train,ind_test in kf.split(train):
\n train2=extract(ind_train,s)
\n test2=extract(ind_test,s)
\n X_train,y_train=train2
\n X_test,y_test=test2
\n try:
\n g=list(y_test).index(np.nan)
\n print('y')
\n except:
\n g=1

\n
    print(X_train,X_test)\n    model.fit(np.array(X_train),np.array(y_train))\n    score=model.score(np.array(X_test),np.array(y_test))\n    ac.append(score)\n    if score>temp[0]:\n        temp[0]=score\n        temp[1]=model\n    print(g)\n    g+=1\n\n\npickle.dump(temp[1],open(str(s)+'xgboost.pkl','wb'))\n\nprint('------------------------------------')\nprint('MaxScore :',' ',s,temp[0])\nprint('------------------------------------')\nprint('Average :',' ',s,sum(ac)/len(ac))\naverages.append(sum(ac)/len(ac))\nmodels.append(temp[1])\n
\n

print('average score:',sum(averages)/len(averages))
\n'''

\n

Sorry , you will some problem reading it

", "rawMarkdown": "here is the code :-\n'''\nimport xgboost\ndef get_model(Y):\n return xgboost.XGBClassifier(n_estimators=1000,\n max_depth=4,\n learning_rate=0.05,\n verbosity=0,\n objective=\"binary:logistic\",\n subsample=0.95,\n colsample_bytree=0.95,\n random_state=2021,\n n_jobs=2,\n scale_pos_weight = np.sum(Y==0) / np.sum(Y==1),\n ) \n \n \n'''\n'''\nfrom sklearn.model_selection import KFold\nimport pickle\nt=Y.T\nmodels=[]\naverages=[]\ndef extract(indexes,i):\n global X\n global train\n \n s_X=[train[ind] for ind in indexes]\n s_Y=[int(Y[ind][i]) for ind in indexes]\n return (s_X,s_Y)\nkf = KFold(n_splits=3)\n\nfor s in range(0,24):\n temp=[0,None]# accuracy , model\n ac=[]# scores\n g=0\n for ind_train,ind_test in kf.split(train):\n train2=extract(ind_train,s)\n test2=extract(ind_test,s)\n X_train,y_train=train2\n X_test,y_test=test2\n try:\n g=list(y_test).index(np.nan)\n print('y')\n except:\n g=1\n \n print(X_train,X_test)\n model.fit(np.array(X_train),np.array(y_train))\n score=model.score(np.array(X_test),np.array(y_test))\n ac.append(score)\n if score>temp[0]:\n temp[0]=score\n temp[1]=model\n print(g)\n g+=1\n \n \n pickle.dump(temp[1],open(str(s)+'xgboost.pkl','wb'))\n \n print('------------------------------------')\n print('MaxScore :',' ',s,temp[0])\n print('------------------------------------')\n print('Average :',' ',s,sum(ac)/len(ac))\n averages.append(sum(ac)/len(ac))\n models.append(temp[1])\n \nprint('average score:',sum(averages)/len(averages))\n'''\n\nSorry , you will some problem reading it", "votes": null } ], "comments": [ { "id": 1123835, "author_name": "jonas0", "author_url": "", "post_date": "12/23/2020 14:33:27", "content": "

Hello there,

\n

you can try to use predict_proba instead of predict. predict_proba predicts the probability of each sample, predict simply returns 0 or 1 as the classified label, not the probabilities.

", "votes": null, "replies": [ { "id": 1123851, "author_name": "swaralipibose", "author_url": "", "post_date": "12/23/2020 14:42:46", "content": "

Sorry but you completely miss understood it I have used predict_proba all the values are coming nan when I used predict all the values come 0

", "votes": null, "replies": [] }, { "id": 1123867, "author_name": "jonas0", "author_url": "", "post_date": "12/23/2020 14:49:19", "content": "

Can you post the code here that causes this behaviour or publish your notebook? Otherwise it's hard to tell what could cause this problem.

", "votes": null, "replies": [] }, { "id": 1123971, "author_name": "swaralipibose", "author_url": "", "post_date": "12/23/2020 16:08:01", "content": "

here is the code :-
\n'''
\nimport xgboost
\ndef get_model(Y):
\n return xgboost.XGBClassifier(n_estimators=1000,
\n max_depth=4,
\n learning_rate=0.05,
\n verbosity=0,
\n objective=\"binary:logistic\",
\n subsample=0.95,
\n colsample_bytree=0.95,
\n random_state=2021,
\n n_jobs=2,
\n scale_pos_weight = np.sum(Y==0) / np.sum(Y==1),
\n )

\n

'''
\n'''
\nfrom sklearn.model_selection import KFold
\nimport pickle
\nt=Y.T
\nmodels=[]
\naverages=[]
\ndef extract(indexes,i):
\n global X
\n global train

\n
s_X=[train[ind] for ind in indexes]\ns_Y=[int(Y[ind][i]) for ind in indexes]\nreturn (s_X,s_Y)\n
\n

kf = KFold(n_splits=3)

\n

for s in range(0,24):
\n temp=[0,None]# accuracy , model
\n ac=[]# scores
\n g=0
\n for ind_train,ind_test in kf.split(train):
\n train2=extract(ind_train,s)
\n test2=extract(ind_test,s)
\n X_train,y_train=train2
\n X_test,y_test=test2
\n try:
\n g=list(y_test).index(np.nan)
\n print('y')
\n except:
\n g=1

\n
    print(X_train,X_test)\n    model.fit(np.array(X_train),np.array(y_train))\n    score=model.score(np.array(X_test),np.array(y_test))\n    ac.append(score)\n    if score>temp[0]:\n        temp[0]=score\n        temp[1]=model\n    print(g)\n    g+=1\n\n\npickle.dump(temp[1],open(str(s)+'xgboost.pkl','wb'))\n\nprint('------------------------------------')\nprint('MaxScore :',' ',s,temp[0])\nprint('------------------------------------')\nprint('Average :',' ',s,sum(ac)/len(ac))\naverages.append(sum(ac)/len(ac))\nmodels.append(temp[1])\n
\n

print('average score:',sum(averages)/len(averages))
\n'''

\n

Sorry , you will some problem reading it

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1122087": "I tried to use xgboost in the competition but i think god did not like it .\n\nI tried to predict outputs in probability as i needed them in probability not normal output everywhere there was nan , i used predict function i got all the answers as 0\n\nNote there was no nan in y_train \n\nPlease help 🤒", "1123835": "Hello there,\n\nyou can try to use `predict_proba` instead of `predict`. `predict_proba` predicts the probability of each sample, `predict` simply returns 0 or 1 as the classified label, not the probabilities.", "1123851": "Sorry but you completely miss understood it I have used `predict_proba ` all the values are coming nan when I used `predict ` all the values come 0", "1123867": "Can you post the code here that causes this behaviour or publish your notebook? Otherwise it's hard to tell what could cause this problem.", "1123971": "here is the code :-\n'''\nimport xgboost\ndef get_model(Y):\n return xgboost.XGBClassifier(n_estimators=1000,\n max_depth=4,\n learning_rate=0.05,\n verbosity=0,\n objective=\"binary:logistic\",\n subsample=0.95,\n colsample_bytree=0.95,\n random_state=2021,\n n_jobs=2,\n scale_pos_weight = np.sum(Y==0) / np.sum(Y==1),\n ) \n \n \n'''\n'''\nfrom sklearn.model_selection import KFold\nimport pickle\nt=Y.T\nmodels=[]\naverages=[]\ndef extract(indexes,i):\n global X\n global train\n \n s_X=[train[ind] for ind in indexes]\n s_Y=[int(Y[ind][i]) for ind in indexes]\n return (s_X,s_Y)\nkf = KFold(n_splits=3)\n\nfor s in range(0,24):\n temp=[0,None]# accuracy , model\n ac=[]# scores\n g=0\n for ind_train,ind_test in kf.split(train):\n train2=extract(ind_train,s)\n test2=extract(ind_test,s)\n X_train,y_train=train2\n X_test,y_test=test2\n try:\n g=list(y_test).index(np.nan)\n print('y')\n except:\n g=1\n \n print(X_train,X_test)\n model.fit(np.array(X_train),np.array(y_train))\n score=model.score(np.array(X_test),np.array(y_test))\n ac.append(score)\n if score>temp[0]:\n temp[0]=score\n temp[1]=model\n print(g)\n g+=1\n \n \n pickle.dump(temp[1],open(str(s)+'xgboost.pkl','wb'))\n \n print('------------------------------------')\n print('MaxScore :',' ',s,temp[0])\n print('------------------------------------')\n print('Average :',' ',s,sum(ac)/len(ac))\n averages.append(sum(ac)/len(ac))\n models.append(temp[1])\n \nprint('average score:',sum(averages)/len(averages))\n'''\n\nSorry , you will some problem reading it" }, "source": "meta" }