# Conformal Predictive System

## Main.ConformalPredictiveSystem History

Hide minor edits - Show changes to output

July 31, 2017, at 06:29 PM
by - added universally consistent predictive distributions

Added lines 18-19:

There are universally consistent predictive distributions (Vovk, 2017).

Changed lines 22-23 from:

Conformal predictive systems can be applied for the purpose of [[conformal decision making | decision making]].

to:

Conformal predictive systems can be applied for the purpose of [[conformal decision making | decision making]]. Universally consistent predictive distributions can be used for making asymptotically efficient decisions.

Changed lines 25-26 from:

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line Compression Modelling Project (New Series), Working Paper 17, April 2017.

to:

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line Compression Modelling Project (New Series), Working Paper 17, April 2017.

* Vladimir Vovk (2017). [[http://www.alrw.net/articles/18.pdf | Universally consistent predictive distributions]]. On-line Compression Modelling Project (New Series), Working Paper 18, April 2017.

* Vladimir Vovk (2017). [[http://www.alrw.net/articles/18.pdf | Universally consistent predictive distributions]]. On-line Compression Modelling Project (New Series), Working Paper 18, April 2017.

July 31, 2017, at 05:04 PM
by - added another class of conformal predictive distributions and decision making

Changed lines 7-9 from:

where $\hat y$ is the prediction for the label of $x$ based on the training sequence $z_1,\ldots,z_n$.

The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

to:

where $\hat y$ is the prediction for the label of $x$ based on the training sequence $z_1,\ldots,z_n$ and $(x,y)$.

An even wider class is

\[

A((z_1,\ldots,z_n),(x,y)) := (y - \hat y)/\sigma_y,

\]

where $\sigma_y > 0$ is an estimate of the variability or difficulty of $y$ computed from the training sequence and $(x,y)$. (The methods for computing $\hat y$ and $\sigma_y$ are supposed invariant with respect to permutations of $z_1,\ldots,z_n$.) The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

An even wider class is

\[

A((z_1,\ldots,z_n),(x,y)) := (y - \hat y)/\sigma_y,

\]

where $\sigma_y > 0$ is an estimate of the variability or difficulty of $y$ computed from the training sequence and $(x,y)$. (The methods for computing $\hat y$ and $\sigma_y$ are supposed invariant with respect to permutations of $z_1,\ldots,z_n$.) The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

Added lines 17-20:

!!Conformal decision making

Conformal predictive systems can be applied for the purpose of [[conformal decision making | decision making]].

Changed line 1 from:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal predictor | conformal transducers]] that, for each training sequence and each test object, output [[p-~~values~~]] $p^y$ that are increasing as a function of the label $y$, assumed to be a real number. The function $y\mapsto p^y$ is then called a ''predictive distribution''.

to:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal predictor | conformal transducers]] that, for each training sequence and each test object, output [[p-value]]s $p^y$ that are increasing as a function of the label $y$, assumed to be a real number. The function $y\mapsto p^y$ is then called a ''predictive distribution''.

May 17, 2017, at 07:14 AM
by - added two remarks

Changed lines 1-12 from:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal predictor | conformal transducers]] that ~~are increasing as function of the label $y$~~, ~~assumed to be a real~~ number.

to:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal predictor | conformal transducers]] that, for each training sequence and each test object, output [[p-values]] $p^y$ that are increasing as a function of the label $y$, assumed to be a real number. The function $y\mapsto p^y$ is then called a ''predictive distribution''.

A wide class of [[conformal predictor | conformity measures]] that often lead to conformal predictive systems is

\[

A((z_1,\ldots,z_n),(x,y)) := y - \hat y,

\]

where $\hat y$ is the prediction for the label of $x$ based on the training sequence $z_1,\ldots,z_n$.

The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

The formal definition of conformal predictive systems takes account of the fact that, in the case of [[conformal predictor | smoothed conformal predictors]], $p^y$ also depends on the random number $\eta\in[0,1]$, and a fuller notation is $p^y(\eta)$. It is also required that $p^y(0)\to0$ as $y\to-\infty$ and $p^y(1)\to1$ as $y\to\infty$.

Notice that in the context of conformal predictive systems the p-values acquire properties of probabilities. Besides, they have some weak properties of [[conditionality | object conditionality]]: e.g., the ''central prediction regions'' $\{y\mid\epsilon/2\le p^y\le 1-\epsilon/2\}$ are not empty, except in very pathological cases.

A wide class of [[conformal predictor | conformity measures]] that often lead to conformal predictive systems is

\[

A((z_1,\ldots,z_n),(x,y)) := y - \hat y,

\]

where $\hat y$ is the prediction for the label of $x$ based on the training sequence $z_1,\ldots,z_n$.

The width $p^y(1)-p^y(0)$ of such conformal predictive distributions is typically equal to $1/(n+1)$, where $n$ is the length of the training sequence, except for at most $n$ values of $y$.

The formal definition of conformal predictive systems takes account of the fact that, in the case of [[conformal predictor | smoothed conformal predictors]], $p^y$ also depends on the random number $\eta\in[0,1]$, and a fuller notation is $p^y(\eta)$. It is also required that $p^y(0)\to0$ as $y\to-\infty$ and $p^y(1)\to1$ as $y\to\infty$.

Notice that in the context of conformal predictive systems the p-values acquire properties of probabilities. Besides, they have some weak properties of [[conditionality | object conditionality]]: e.g., the ''central prediction regions'' $\{y\mid\epsilon/2\le p^y\le 1-\epsilon/2\}$ are not empty, except in very pathological cases.

Changed line 1 from:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal ~~prediction~~ | conformal transducers]] that are increasing as function of the label $y$, assumed to be a real number.

to:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal predictor | conformal transducers]] that are increasing as function of the label $y$, assumed to be a real number.

Changed lines 4-5 from:

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line

Compression Modelling Project (New Series), Working Paper 17, April 2017.

Compression Modelling Project (New Series), Working Paper 17, April 2017.

to:

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line Compression Modelling Project (New Series), Working Paper 17, April 2017.

May 13, 2017, at 11:20 AM
by - created the page

Added lines 1-5:

Conformal predictive systems are introduced in the recent technical report Vovk et al. (2017). Essentially, these are [[conformal prediction | conformal transducers]] that are increasing as function of the label $y$, assumed to be a real number.

'''Bibliography'''

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line

Compression Modelling Project (New Series), Working Paper 17, April 2017.

'''Bibliography'''

* Vladimir Vovk, Jieli Shen, Valery Manokhin, and Min-ge Xie (2017). [[http://www.alrw.net/articles/17.pdf | Nonparametric predictive distributions based on conformal prediction]]. On-line

Compression Modelling Project (New Series), Working Paper 17, April 2017.