3.20. Experiments

This notebook documents the use of the Experiment class for running many experiments, and comparing the results.

import conx as cx

Using TensorFlow backend.
Conx, version 3.6.0

First, we create an experiment object:

exp = cx.Experiment("XOR-Test1")

Experiment takes a function, and any number of variations of values.

The function takes whatever argumnets you wish, creates a network, trains it, and returns the network and a category describing the run:

def function(optimizer, activation, **options):
    category = "%s-%s" % (optimizer, activation)
    print("category %s running..." % category)
    net = cx.Network("XOR", 2, 2, 1, activation=activation)
    net.compile(error="mse", optimizer=optimizer)
    net.dataset.append_by_function(2, (0, 4), "binary", lambda i,v: [int(sum(v) == len(v))])
    net.train(report_rate=10000, verbose=0, plot=False, **options)
    return category, net

Then we run a number of trials (perhaps just 1). Notice that each argument should be given as a list. The total number of runs per trial is the product of the lengths of the arguments.

exp.run(function,
        trials=2,
        epochs=[1000],
        accuracy=[0.8],
        tolerance=[0.2],
        optimizer=["adam", "sgd"],
        activation=["sigmoid", "relu"],
        dir="/tmp/")

category adam-sigmoid running...
category sgd-sigmoid running...
category adam-relu running...
category sgd-relu running...
category adam-sigmoid running...
category sgd-sigmoid running...
category adam-relu running...
category sgd-relu running...

The results is a list of (category, network-name) pairs:

exp.results

[('adam-sigmoid', '/tmp/XOR-Test1-00001-00001'),
 ('sgd-sigmoid', '/tmp/XOR-Test1-00001-00002'),
 ('adam-relu', '/tmp/XOR-Test1-00001-00003'),
 ('sgd-relu', '/tmp/XOR-Test1-00001-00004'),
 ('adam-sigmoid', '/tmp/XOR-Test1-00002-00001'),
 ('sgd-sigmoid', '/tmp/XOR-Test1-00002-00002'),
 ('adam-relu', '/tmp/XOR-Test1-00002-00003'),
 ('sgd-relu', '/tmp/XOR-Test1-00002-00004')]

Often, you may wish to plot the results of learning. This may take some time, as the function will re-load each network:

exp.plot("loss")

Notice that each category has its own color.

There is also a generic apply method for calling a function with each of the (category, network-names). Ususally, you would probably want to re-load the network, and perform some operation in the function.

exp.apply(lambda category, exp_name: (category, exp_name))

[('adam-sigmoid', '/tmp/XOR-Test1-00001-00001'),
 ('sgd-sigmoid', '/tmp/XOR-Test1-00001-00002'),
 ('adam-relu', '/tmp/XOR-Test1-00001-00003'),
 ('sgd-relu', '/tmp/XOR-Test1-00001-00004'),
 ('adam-sigmoid', '/tmp/XOR-Test1-00002-00001'),
 ('sgd-sigmoid', '/tmp/XOR-Test1-00002-00002'),
 ('adam-relu', '/tmp/XOR-Test1-00002-00003'),
 ('sgd-relu', '/tmp/XOR-Test1-00002-00004')]