1. An additive mixture comprising
A) at least one terpolymer consisting of ethylene, propene and at least one ethylenically unsaturated ester, wherein the terpolymer
i) contains from 6.0 to 12.0 mol % of structural units derived from at least one ethylenically unsaturated ester, wherein the at least one ethylenically unsaturated ester is selected from the group consisting of ethylenically unsaturated esters having a C1- to C3-alkyl radical,
ii) contains from 0.5 to 4.0 methyl groups derived from propene per 100 aliphatic carbon atoms,
iii) has fewer than 8.0 methyl groups stemming from chain ends per 100 CH2 groups, and
B) from 0.5 to 20 parts by weight, based on A), of at least one copolymer of ethylene and an olefinically unsaturated compound whose content of olefinically unsaturated compounds is at least 2 mol % higher than the content of ethylenically unsaturated ester in the terpolymer defined under A).
2. The additive mixture as claimed in claim 1, in which the ethylenically unsaturated ester of component A) is the vinyl ester of a carboxylic acid having from 1 to 4 carbon atoms.
3. The additive mixture as claimed in claim 1, in which the ethylenically unsaturated ester is vinyl acetate.
4. The additive mixture of claim 1, in which the sum G of molar content of unsaturated ester i) and the number of methyl groups derived from propene per 100 aliphatic carbon atoms of the polymer ii)
G=mol % of unsaturated ester+propene-CH3
is between 8.0 and 14.0.
5. The additive mixture of claim 1, in which component A) further comprises from 0.3 to 5.0% by weight of at least one structural unit derived from a moderator comprising carbonyl groups.
6. The additive mixture of claim 1, in which the content of olefinically unsaturated compounds in copolymer B1) is at least three mol % higher than that of the ethylenically unsaturated ester in the terpolymer A).
7. A process for preparing the terpolymer A) of claim 1, by reacting a mixture consisting of ethylene, propene and at least one vinyl ester under elevated pressure and elevated temperature in the presence of a free radical-forming initiator, and in which the molecular weight of the terpolymer A) is adjusted by a moderator comprising carbonyl groups.
8. A process for improving the flowability of a fuel oil by adding to the fuel oil the additive mixture of claim 1.
9. A composition comprising the additive mixture of claim 1 and at least one oil-soluble polar nitrogen compound.
10. A composition comprising the additive mixture of claim 1 and at least one alkylphenol-aldehyde resin.
11. A composition comprising the additive mixture of claim 1 and at least one olefin polymer.
12. A composition comprising the additive mixture of claim 1 and at least one polyoxyalkylene compound.
13. A fuel oil composition comprising a middle distillate and the additive mixture of claim 1.
14. A composition comprising the additive mixture of claim 1 and at least one comb polymer.
The claims below are in addition to those above.
All refrences to claim(s) which appear below refer to the numbering after this setence.
1. A computer-implemented method for generating a monotonic classifier, the method comprising:
mapping a plurality of feature vectors to a plurality of ordered classes, wherein each feature vector maps to a corresponding class;
for each class other than a lowest-ordered class, training, via a processor, a binary classifier to distinguish between the class and a preceding class, wherein training a binary classifier comprises:
determining a data set impurity value for a data set of feature vectors, wherein the data set impurity value equals the sum, over all classes, of the product of (i) the number of feature vectors in the data set having a particular class divided by the number of feature vectors in the data set and (ii) the logarithm of the number of feature vectors in the data set having the particular class divided by the number of feature vectors in the data set,
determining a decision rule for separating the data set into a plurality of subsets that maximizes the decrease in impurity, wherein the decrease in impurity equals the data set impurity value for the data set minus, for each subset, the product of a number of feature vectors in the subset and a data set impurity value for the subset divided by the number of feature vectors in the data set, and
repeating the determining a data set impurity and the determining a decision rule on one or more subsets until the data set impurity value for each subset is minimized; and
ordering, via the processor, the binary classifiers as a monotonic classifier based on the class corresponding to each binary classifier, wherein a number of classes in the plurality of ordered classes is one more than a number of binary classifiers.
2. The method of claim 1 wherein ordering the binary classifiers comprises ordering the binary classifier as a balanced binary decision tree.
3. The method of claim 1 wherein ordering the binary classifiers comprises ordering the binary classifiers as an unbalanced binary decision tree.
4. The method of claim 1 wherein ordering the binary classifiers comprises ordering the binary classifiers as a class-ordered linear chain of binary classifiers.
5. The method of claim 1 wherein mapping a plurality of feature vectors to a plurality of ordered classes comprises:
ordering the plurality of classes in a linear order; and
for each feature vector, assigning the feature vector to a particular class, wherein the feature vector is not greater than any feature vector corresponding to a class that is greater than the particular class.
6. The method of claim 1 wherein the ordered classes comprise one or more of the following:
personal data;
a credit score;
a housing price; and
a test score.
7. A system for generating a monotonic classifier, the system comprising:
a processor; and
a memory,
wherein the memory comprises one or more programming instructions for performing a method for generating a monotonic classifier, the method comprising:
mapping a plurality of feature vectors to a plurality of ordered classes, wherein each feature vector maps to a corresponding class,
for each class other than a lowest-ordered class, training a binary classifier to distinguish between the class and a preceding class, wherein training a binary classifier comprises:
determining a data set impurity value for a data set of feature vectors, wherein the data set impurity value equals the sum, over all classes, of the product of (i) the number of feature vectors in the data set having a particular class divided by the number of feature vectors in the data set and (ii) the logarithm of the number of feature vectors in the data set having the particular class divided by the number of feature vectors in the data set,
determining a decision rule for separating the data set into a plurality of subsets that maximizes the decrease in impurity, wherein the decrease in impurity equals the data set impurity value for the data set minus, for each subset, the product of a number of feature vectors in the subset and a data set impurity value for the subset divided by the number of feature vectors in the data set, and
repeating the determining a data set impurity and the determining a decision rule on one or more subsets until the data set impurity value for each subset is minimized, and
ordering the binary classifiers as a monotonic classifier based on the class corresponding to each binary classifier, wherein a number of classes in the plurality of ordered classes is one more than a number of binary classifiers.
8. The system of claim 7 wherein ordering the binary classifiers comprises one or more programming instructions for ordering the binary classifier as a balanced binary decision tree.
9. The system of claim 7 wherein ordering the binary classifiers comprises one or more programming instructions for ordering the binary classifiers as an unbalanced binary decision tree.
10. The system of claim 7 wherein ordering the binary classifiers comprises one or more programming instructions for ordering the binary classifiers as a class-ordered linear chain of binary classifiers.
11. The system of claim 7 wherein mapping a plurality of feature vectors to a plurality of ordered classes comprises one or more programming instructions for:
ordering the plurality of classes in a linear order; and
for each feature vector, assigning the feature vector to a particular class, wherein the feature vector is not greater than any feature vector corresponding to a class that is greater than the particular class.
12. The system of claim 7 wherein the ordered classes comprise one or more of the following:
personal data;
a credit score;
a housing price; and
a test score.
13. A computer-implemented method of assigning a credit-worthiness score to an applicant, the method comprising:
mapping a plurality of feature vectors to a plurality of ordered credit-worthiness classes, wherein each feature vector maps to a corresponding credit-worthiness class;
for each credit-worthiness class other than a lowest-ordered credit-worthiness class, training, via a processor, a binary classifier to distinguish between the credit-worthiness class and a preceding credit-worthiness class, wherein training a binary classifier comprises:
determining a data set impurity value for a data set of feature vectors, wherein the data set impurity value equals the sum, over all credit-worthiness classes, of the product of (i) the number of feature vectors in the data set having a particular credit-worthiness class divided by the number of feature vectors in the data set and (ii) the logarithm of the number of feature vectors in the data set having the particular credit-worthiness class divided by the number of feature vectors in the data set,
determining a decision rule for separating the data set into a plurality of subsets that maximizes the decrease in impurity, wherein the decrease in impurity equals the data set impurity value for the data set minus, for each subset, the product of a number of feature vectors in the subset and a data set impurity value for the subset divided by the number of feature vectors in the data set, and
repeating the determining a data set impurity and the determining a decision rule on one or more subsets until the data set impurity value for each subset is minimized;
ordering, via the processor, the binary classifiers as a monotonic classifier based on the credit-worthiness class corresponding to each binary classifier, wherein a number of credit-worthiness classes in the plurality of ordered credit-worthiness classes is one more than a number of binary classifiers;
receiving applicant data pertaining to an applicant;
organizing, via the processor, the applicant data into an applicant feature vector pertaining to the applicant; and
assigning a credit-worthiness class to the applicant based on the monotonic classifier and the applicant feature vector.
14. The method of claim 13 wherein mapping a plurality of feature vectors to a plurality of ordered credit-worthiness classes comprises:
ordering the plurality of credit-worthiness classes in a linear order; and
for each feature vector, assigning the feature vector to a particular credit-worthiness class, wherein the feature vector is not greater than any feature vector corresponding to a credit-worthiness class that is greater than the particular credit-worthiness class.
15. The method of claim 13 wherein the applicant data comprises one or more of the following:
personal income;
years of employment;
years of home ownership; and
outstanding debt.