About the average: The two common scenarios are "micro" and "macro" average (I think "macro" is typically the default in scikit-learn) -- you calculated the macro average in your example.
To further explain the difference betw. macro and micro, let's consider a simple 2-class scenario and calculate the precision a) macro-average precision: (PRE1 + PRE2) / 2 b) micro-average precision: (TP1+TP2)/(TP1+TP2+FP1+FP2) Hope that helps. Best, Sebastian > On Jun 17, 2015, at 10:49 AM, Herbert Schulz <hrbrt....@gmail.com> wrote: > > Ok i think i have it, thanks everyone for the help! > > But there is an another problem. > > How are you calculating the avg? > > example: > > ----------- k-NN ----------- > > precision recall f1-score support > > 1.0 0.50 0.43 0.46 129 > 2.0 0.31 0.40 0.35 88 > 3.0 0.45 0.36 0.40 107 > 4.0 0.06 0.03 0.04 33 > 5.0 0.42 0.58 0.49 76 > > avg / total 0.40 0.40 0.40 433 > > so: (0.5+0.31+0.45+0.06+0.42) / 5 = 0.348 ~ 0.35 like i calculated it in > my avg part. Are you using some weights? > > Class: 1 > sensitivity:0.43 > specificity: 0.81 > ballanced accuracy: 0.62 > precision 0.50 > . > . > . > . > > Class: 5 > sensitivity:0.58 > specificity: 0.83 > ballanced accuracy: 0.70 > precision 0.42 > > avg total: > sensitivity: 0.36 > specificity: 0.85 > avg ballance: 0.60 > avg precision: 0.35 > > > > > > > > On 17 June 2015 at 16:06, Herbert Schulz <hrbrt....@gmail.com> wrote: > I actually computed it like this, maybe I did something in my TP,FP,FN,TN > calculation wrong? > > > c1,c2,c3,c4,c5=[1,0,0,0,0],[2,0,0,0,0],[3,0,0,0,0],[4,0,0,0,0],[5,0,0,0,0] > alle=[c1,c2,c3,c4,c5] > > > #as i mentioned 1 vs all, so the first element in the array is just the class > #[1,0,0,0,0] == class 1, then in the order: TP,FP,FN,TN > #maybe here is something wring: > > for i in alle: > pred=predicted > > for k in range(len(predicted)): > > if float(i[0]) == y_test[k]: > if float(i[0]) == pred[k]: > i[1]+=1 > else: > i[2]+=1 > > elif pred[k] == float(i[0]): > i[3]+=1 > elif pred[k] !=float(i[0]) and y_test[k] !=float(i[0]): > i[4]+=1 > > #specs looks like this: [1, 71, 51, 103, 208] > > sens=specs[1]/float(specs[1]+specs[3]) > > > > > if I'm calculatig > > sens=specs[1]/float(specs[1]+specs[2]) im getting also the recall like in the > matrix. > > On 17 June 2015 at 15:42, Andreas Mueller <t3k...@gmail.com> wrote: > Sensitivity is recall: > https://en.wikipedia.org/wiki/Sensitivity_and_specificity > > Recall is TP / (TP + FN) and precision is TP / (TP + FP). > > What did you compute? > > > On 06/17/2015 09:32 AM, Herbert Schulz wrote: >> Yeah i know, thats why I'm asking. i thought precision is not the same like >> recall/sensitivity. >> >> recall == sensitivity!? >> >> But in this matrix, the precision is my calculated sensitivity, or is the >> precision in this case the sensitivity? >> >> On 17 June 2015 at 15:29, Andreas Mueller <t3k...@gmail.com> wrote: >> Yeah that is the rounding of using %2f in the classification report. >> >> >> On 06/17/2015 09:20 AM, Joel Nothman wrote: >>> To me, those numbers appear identical at 2 decimal places. >>> >>> On 17 June 2015 at 23:04, Herbert Schulz <hrbrt....@gmail.com> wrote: >>> Hello everyone, >>> >>> i wrote a function to calculate the sensitivity,specificity, ballance >>> accuracy and accuracy from a confusion matrix. >>> >>> >>> Now i have a Problem, I'm getting different values when I'm comparing my >>> Values with those from the metrics.classification_report function. >>> The general problem ist, my predicted sensitivity is in the classification >>> report the precision value. I'm computing every sensitivity with the one >>> vs all approach. So e.g. Class 1 == true, class 2,3,4,5 are the rest (not >>> true). >>> >>> I did this only to get the specificity, and to compare if i computed >>> everything right. >>> >>> >>> >>> ----------- ensemble ----------- >>> >>> precision recall f1-score support >>> >>> 1.0 0.56 0.68 0.61 129 >>> 2.0 0.28 0.15 0.20 78 >>> 3.0 0.45 0.47 0.46 116 >>> 4.0 0.29 0.05 0.09 40 >>> 5.0 0.44 0.66 0.53 70 >>> >>> avg / total 0.43 0.47 0.43 433 >>> >>> >>> Class: 1 >>> sensitivity:0.556962025316 >>> specificity: 0.850909090909 >>> ballanced accuracy: 0.703935558113 >>> >>> Class: 2 >>> sensitivity:0.279069767442 >>> specificity: 0.830769230769 >>> ballanced accuracy: 0.554919499106 >>> >>> Class: 3 >>> sensitivity:0.446280991736 >>> specificity: 0.801282051282 >>> ballanced accuracy: 0.623781521509 >>> >>> Class: 4 >>> sensitivity:0.285714285714 >>> specificity: 0.910798122066 >>> ballanced accuracy: 0.59825620389 >>> >>> Class: 5 >>> sensitivity:0.442307692308 >>> specificity: 0.927051671733 >>> ballanced accuracy: 0.68467968202 >>> >>> >>> >>> >>> ------------------------------------------------------------------------------ >>> >>> _______________________________________________ >>> Scikit-learn-general mailing list >>> Scikit-learn-general@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >>> >>> >>> >>> >>> ------------------------------------------------------------------------------ >>> >>> >>> >>> _______________________________________________ >>> Scikit-learn-general mailing list >>> >>> Scikit-learn-general@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> >> >> ------------------------------------------------------------------------------ >> >> _______________________________________________ >> Scikit-learn-general mailing list >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general >> >> >> >> >> ------------------------------------------------------------------------------ >> >> >> >> _______________________________________________ >> Scikit-learn-general mailing list >> >> Scikit-learn-general@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > > ------------------------------------------------------------------------------ > > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > > > > ------------------------------------------------------------------------------ > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general ------------------------------------------------------------------------------ _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
