Eureqa User's Guide

Showing results for 
Search instead for 
Did you mean: 

Eureqa User's Guide

1. Entering data

You can paste, view, and edit your data in the far-left tab. Entering data is very similar to using an ordinary spreadsheet application. 

The Eureqa application uses a special layout:

  • Each column corresponds to a single variable of your data (e.g., time or oxygen concentration)
  • The first row, labeled desc, is for commenting and describing what the variable means or measures
  • The second row, labeled var, is the symbol given to each variable (e.g., x or CO2)
  • All remaining rows correspond to simultaneous data measurements and values

You can paste data into cells from many applications that contain spreadsheets such as Microsoft Excel, MATLAB's array editor, or any tab-separated-value text file.

Double-click on a cell to edit its value. You may also enter simple expressions into a cell to generate additional variables. For example, = x + sin(y) will fill the entire column with the numerical result of that expression on each row using the current variable symbols and values.

If your data is partitioned into discontinuous parts (e.g., two or more independent time series or experiments), they should be separated by a blank row. This tells the program not to smooth or differentiate across discontinuous data points.

2. Smoothing data

This is an optional step where the Eureqa application can automatically smooth your data. 

Smoothing data can greatly improve both the speed and the likelihood of finding accurate solutions with the Eureqa formula search. For a user to decide whether to smooth, they must have enough knowledge about their data to determine whether variables are, in fact, smooth signals combined with noise.

To smooth a variable in Eureqa:

  1. Select the row corresponding to the variable you'd like to smooth. This will show the data in the bottom plot.
  2. Select an independent variable in the Smooth vs. column to smooth against (e.g., time).
  3. Select a confidence weight in the Weight Smooth by column if you have a confidence strength variable entered.
  4. Check the Enable Smooth option to smooth.

Eureqa picks the best smooth using generalized cross-validation among cubic b-splines. If your data can benefit from more sophisticated pre-processing, you are encouraged to do so in another application of your choice and copy the result into Eureqa by hand.

3. Picking modeling task

Here you control what type of formula to search for, and how to search for it.

Search for a formula f( ) such that

Edit this expression to specify the type of relationship you want to model. For example, if you want to model the variable z as a function of and y, you would enter: 

z = f(x, y) 

To search for a differential equation, you can use the D(x,y) command. For example, to find an ordinary differential equation for y as a function of y, you could enter:

D(y,t) = f(y)

You may also enter more complex target forms. For example, entering z = f(x) + f(y) indicates that you would like to find a function that is evaluated on both x and y, then added together to model z.

Remaining options are described below...

Fitness metric

This specifies what type of error to measure when comparing and optimizing solutions. For example, you may wish to minimize squared error if your data has normally distributed noise, or logarithmic error if it contains many outliers.

The list below describes some of the fitness metrics available in Eureqa. All fitness metrics are normalized based on the target values in the dataset.

  • Mean Absolute Error (MAE): minimizes the mean of the absolute value of residual errors, mean(abs(error)). MAE assumes noise follows a double exponential distribution.
  • Mean Squared Error (MSE): minimizes the mean of the squared residual errors, mean(error^2). MSE assumes noise follows a normal distribution.
  • Root Mean Squared Error (RMSE): minimizes the square root of the MSE, sqrt(mean(error^2)). RMSE assumes noise follows a normal distribution.
  • Mean Logarithmic Error (MLE): minimizes the mean of the natural log of the residual errors, mean(log(1 + error)). MLE assumes noise follows a heavy-tailed distribution with large outliers.
  • Exponential Mean Logarithmic Error (EMLE): minimizes the exponential of the MLE, exp(mean(log(1 + error))). EMLE assumes noise follows a heavy-tailed distribution with large outliers.
  • Correlation Coefficient (R): maximizes normalized covariance. R finds patterns that explain the shape of the data quickly and does not require scale and offset.
  • Minimize the Difference: minimizes the signed difference between left- and right-hand sides of the target formula. Use this metric to create custom fitness functions. For example, (y - f(x))^4 = 0 would minimize the 4th-power error.
  • Akaike Information Criterion (AIC): minimizes natural log of the MSE and number of parameters. This is an entropy measure, used to explicitly minimize the number of free parameters of the model.
  • Bayesian Information Criterion (BIC): minimizes natural log of the MSE and number of parameters. This is an entropy measure, used to explicitly minimize the number of free parameters of the model.
  • Maximum Error (Maximum): minimizes the single highest error of the residuals. This metric is used to minimize the worst case error or to force an algorithm to model a small residual feature.
  • Median Error (Median): minimizes the single median error of the residuals. This metric is invariant to outliers, and is used to minimize the "typical case" error.
  • Implicit Derivative Error (Implicit): minimizes the difference between implicit derivatives derived from a model and estimated from the dataset. This metric is used to search for invariant relationships, e.g., f(x, y) = 0, where x and y are continuous and ordered by an independent variable such as time.

Order data points by

This is the variable Eureqa will use to plot your data against by default, and the order used for calculating derivatives if any are used.

Weight errors by

This is the variable each data point is weighted by in the fitness metric.

Using building blocks

Eureqa searches for formulas by combining mathematical building blocks (e.g., add, subtract, multiply, divide). You can limit the set of building blocks that the algorithm uses by checking and unchecking the various built-in operations.

The table below describes some of the Eureqa formula building blocks:






Allows solutions to use numeric constants.


x + y or add(x,y)



x - y or sub(x,y)



x * y or mul(x,y)



x / y or div(x,y)

y must be non-zero.

square root


Returns x^0.5 ,where x must be positive.



Returns e^x.



This is the natural logaritm (base e).



The angle is in radians.



The angle is in radians.



The angle is in radians.

absolute value


Returns the positive value of x.


x ^ y or pow(x,y)

x and y could be any expression.

power to constant


Provides a restricted form of the power building block. x can be any expression, c must be a constant.

time delay


Returns the value of expression x at c time units in the past. x can be any expression, c must be a positive constant.

time delay of variable


Provides a restricted form of the delay building block. Returns the value of variable v at c time units in the past. v must be a variable, c must be a positive constant.

simple moving average

sma(v,c) or sma_var(v,c)

Returns the average of the data points within the past c time units. v must be a variable, c must be a positive constant.

time integral


Returns the trapezoidal sum of the expression x, starting at 0 up to the current data point.

step function


Returns 1 if x is positive, zero otherwise.

sign function


Returns -1 if x is negative, +1 if x is positive, and 0 if x is zero.

logistic function


This is a common sigmoid squashing function. Returns 1/(1+ exp(-x)).

hill function


This is a common saturation function. Returns x^2/(1 + x^2). x must be non-zero.

gamma function


This is a continuous version of the factorial. It returns the fast approximation pow((x/e)*sqrt(x*sinh(1/x)),x)*sqrt(2*pi/x). x must be non-zero.

gaussian function


This is a bell-shaped squashing function. Returns exp(-x^2).



Returns the minimum (signed) result of x and y for the data point.



Returns the maximum (signed) result of x and y for the data point.



Returns the remainder of x/y.



Returns an integer of x rounded down toward -infinity.



Returns an integer of x rounded up toward +infinity.

less than


Returns 1 if x < y; returns 0 otherwise.

equal to


Returns 1 if x equals y numerically; returns 0 otherwise.

boolean and


Returns 1 if both x and y are greater than 0; returns 0 otherwise.

boolean or


Returns 1 if either x or y are greater than 0; returns otherwise.

boolean xor


Returns 1 if (x <= 0 and y > 0) or (x > 0 and y <= 0), 0 otherwise.

boolean not


Returns 0 if x is greater than 0, 1 otherwise.

inverse sine


x must be between -1 and +1.

inverse cosine


x must be between -1 and +1.

inverse tangent



inverse tangent (2argument)


Returns atan(y/x), respecting the quadrant and sign of the vector. x and y cannot both be zero.

hyperbolic sine



hyperbolic cosine



hyperbolic tangent


This is a common squashing function. Returns a value between -1 and +1.

inverse hyperbolic sine



inverse hyperbolic cosine



inverse hyperbolic tangent




Limiting the building blocks implies some expert knowledge from the user. For example, the user may know that a chemical reaction is unlikely to use trigonometric terms. 

Limiting the set of building blocks can greatly improve the speed and likelihood that Eureqa finds an exact solution. However, disabling too many building blocks could preclude the search from finding the exact solution if a necessary operation is disabled.

Using servers

Eureqa is designed to use multiple computers when searching for solutions. This option allows you to specify additional computers that are running the Eureqa Stand-alone Server. 

Eureqa servers running on the local network will be listed automatically. You may need to enter other servers manually by right-clicking on the server list and clicking the Add hostname... menu item.

4. Starting and monitoring the search

This view allows you to start, pause, and stop the formula search, as well as monitor various performance and progress statistics of the search. 

Click on the Start button to begin the search.

After starting, Eureqa will attempt to connect to and initialize the computers selected in the settings view. Important messages will be displayed in the Log messages and events text window.

The Progress and performance statistics window shows several important statistics such as the time duration of the search, the number of servers connected, and the performance speed of the search.

The main plot to the right shows the fitness metric of the best solution Eureqa was able to find since the search began. Long plateaus in progress may indicate that the search for simple explanations of the data has been exhausted.

5. Viewing and analyzing results

This view shows the best solutions Eureqa finds in real time. 

The best solutions are determined by two factors: their complexity (Size) and their accuracy (Error) on the validation data. Those listed in the List of current best solutions window have the highest accuracy for various complexities/sizes of solutions.

The Training Data is a subset of your data that is being used by the Eureqa algorithm to search for solutions. The Validation Data is a second subset that is only used by this window in order to select the best solutions to display/report to the user. 

  • Double-click on a formula in the list to select it in plain text for copying to the clipboard. 
  • Single-click on a formula to plot it against the data points and calculate some quick statistics of the fit.

Several tools are also available for further analysis of the selected solution. Capabilities include finding the global maximum, generating a report of solutions, and suggesting new experiments and data to collect.

Labels (1)
Image Sensor

Great post!

Sam Sam1
Computer Board



I am a student and I want to buy a license for Eureqa. I contacted with sales department, but they advised me to ask for help here in this section. 

Please advice me in this regard. 



Version history
Last update:
2 weeks ago
Updated by: