6
$\begingroup$

Excuse me I'm not an English speaker and I have no idea whatsoever whether there is any standard naming scheme. And since I don't want to use names I have read in some books I will be overly descriptive.

Let's suppose I want to describe a dataset containing data for prediction of weather pollutants. Each row of this dataset consists of daily averages of some weather parameters for a particular (current) day, daily average of pollutants for the previous day, daily average of pollutants on the current day (this value will be predicted)

  1. What is the name of single row in this dataset;
  2. What is the name of single cell in single row;
  3. What is the name of part of this dataset that will be input of this model (in this case weather parameters, and pollutant for the previous day);
  4. What is the name of part of the dataset that model will predict based on the part in latter point;

I have some guesses but I'll post them when I get some answers. References to books/articles would be appreciated.

PS. If anyone knows proper Polish translations I would also be very interested :)

$\endgroup$

4 Answers 4

6
$\begingroup$

It seems like in your setup, your inputs (the data that you're using to model) and your outputs (what you'd like to predict) are both in the same table. In that case it's a bit complicated, as:

  1. A row is an input/output tuple (Example; Observation; Data point; Datum)
  2. A single cell is a either an input feature value (or attribute) or output value
  3. Input data or Training Set
  4. Outputs or Targets

Or mathematically, you'll often see:

  1. ${\bf{x}_i, y_i}$
  2. Either $x_{ij}$ or $y_i$ depending which column you select
  3. $\bf{X}$
  4. $\bf{y}$

It's worth looking at the wiki page on Cross-validation to see how to split a dataset up correctly.

$\endgroup$
4
$\begingroup$

Based on Andrew Ng's ml-class.org and Tom Mitchell's "Machine Learning" book I think they will be called

  1. Training example
  2. Feature value
  3. Training set
  4. Output/target variable

But naming will depend on the algorithm, I believe. Say, if you use Decision Trees then your training examples would become instances and your features would become attributes.

$\endgroup$
3
$\begingroup$

(1) data point, (2) feature value

I think that for regression:

(3) regressors, explanatory variables, input variables, predictor variables, (4) regressand, exogenous variable, response variable, measured variable

for classification:

(3) features, input features, input variable (4) class

$\endgroup$
1
  • $\begingroup$ I got different naming scheme for all apart from (2). $\endgroup$
    – jb.
    Feb 8, 2012 at 19:45
2
$\begingroup$

Answering more generally, as I'm not sure if your datasets or textbooks are always going to be restricted to weather data, and not duplicating the answers above

  1. observations, or cases
  2. I always refer to this as a vector ij
  3. independent variables (normally in an experimental or quasi-experimental context only)
  4. dependent variable

I feel for people in different disciplines. I wish we all referred to the same things with the same names.

$\endgroup$
2
  • $\begingroup$ I agree with 1, 3, and 4; 2 is widely called "data point" except among poseurs who call any finding a "data point" :-) $\endgroup$
    – rolando2
    Feb 23, 2012 at 0:18
  • $\begingroup$ @rolando2 I got stuck on 2, as I'm normally only worried about individual cells when I'm data cleaning, like I am doing today, and then I just do automated searching and recoding based on cell contents, not cell vector position. :) 8 hours of string cleaning and counting... /sigh $\endgroup$
    – Michelle
    Feb 23, 2012 at 0:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.