Please explain collections.defaultdict(lambda: 1)

[metaperl.com]

I'm reading http://norvig.com/spell-correct.html

and do not understand the expression listed in the subject which is
part of this function:

def train(features):
model = collections.defaultdict(lambda: 1)
for f in features:
model[f] += 1
return model
Per http://docs.python.org/lib/defaultdict-examples.html

It seems that there is a default factory which initializes each key to
1. So by the end of train(), each member of the dictionary model will
have value >= 1

But why wouldnt he set the value to zero and then increment it each
time a "feature" (actually a word) is encountered? It seems that each
model value would be 1 more than it should be.

[Paul McGuire]

On Nov 6, 8:54 am, "metaperl.com" <[e-mail]:

I'm readinghttp://norvig.com/spell-correct.html

and do not understand the expression listed in the subject which is
part of this function:

def train(features):
model = collections.defaultdict(lambda: 1)
for f in features:
model[f] += 1
return model

Perhttp://docs.python.org/lib/defaultdict-examples.html

It seems that there is a default factory which initializes each key to
1. So by the end of train(), each member of the dictionary model will
have value >= 1

But why wouldnt he set the value to zero and then increment it each
time a "feature" (actually a word) is encountered? It seems that each
model value would be 1 more than it should be.

The explanation is a little further down on that same page, on the
discussion of "novel" words and avoiding the probablity of them being
0 just because they have not yet been seen in the training text.

-- Paul

[Duncan Booth]

"metaperl.com" <me******@gmail.comwrote:

Per http://docs.python.org/lib/defaultdict-examples.html

It seems that there is a default factory which initializes each key to
1. So by the end of train(), each member of the dictionary model will
have value >= 1

But why wouldnt he set the value to zero and then increment it each
time a "feature" (actually a word) is encountered? It seems that each
model value would be 1 more than it should be.

The author explains his reasoning in the article: he wants to treat novel
words (i.e. those which did not appear in the training corpus) as having
been seen once.