Optimize Code with ActiveClassification

Considering two programs implementing the same function, construct a classifier able to predict which program is the fastest for a given input. The classifier has access to features of the program input (matrix sizes here), and is obtained using ActiveClassification, a function to learn from a "teacher" (a benchmarking function here) instead of a labeled dataset.

Consider two implementations of the same function.

In[1]:=1

✖

In[2]:=2

✖

In[3]:=3

✖

Out[3]=3

In[4]:=4

✖

Out[4]=4

Create an input sampler and benchmarking function.

In[5]:=5

✖

In[6]:=6

✖

In[7]:=7

✖

Out[7]=7

In[8]:=8

✖

Out[8]=8

Create a logistic classifier that imitates the benchmarking function.

In[9]:=9

✖

Out[9]=9

Extract the classifier and analyze its performances.

In[10]:=10

✖

Out[10]=10

In[11]:=11

✖

Out[11]=11

Use the classifier to create an optimized function.

In[12]:=12

✖

In[13]:=13

✖

Out[13]=13

Obtain a simple function from the classifier to simplify the optimized function.

In[14]:=14

✖

Out[14]=14

In[15]:=15

✖