Basic Examples (5) 

Compute the LogSumExp of a list of numbers:

In[1]:=

Out[1]=

This is the same as:

In[2]:=

Out[2]=

However, LogSumExp can be computed in a numerically stable way using MachinePrecision even when larger numbers are involved:

In[3]:=

Out[3]=

The naive implementation loses precision and switches to arbitrary precision to compensate:

In[4]:=

Out[4]=

Compare with the exact result:

In[5]:=

Out[5]=

Scope (5) 

Compute the LogSumExp on symbolic data:

In[6]:=

Out[6]=

Reduce a matrix by applying LogSumExp to its rows:

In[7]:=

Out[7]=

In[8]:=

Out[8]=

Apply to columns:

In[9]:=

Out[9]=

Aggregate both rows and columns:

In[10]:=

Out[10]=

This is equivalent to:

In[11]:=

Out[11]=

Properties and Relations (3) 

The LogSumExp operation allows you to do addition outside of a Log in the same way that addition allows you to multiply outside of a Log:

In[12]:=

Out[13]=

A Log transform turns multiplication into addition:

In[14]:=

Out[14]=

And it turns addition into LogSumExp:

In[15]:=

Out[15]=