Aug
15
comment When are precision and recall inversely related?
not exactly limit, but change the number of results depending on some threshold - but, basically, right. And, yes, as for the binary classification problems you've mentioned you, really, deal with plain accuracy here, i.e. precision will be the same as recall. Because the denominator is the same for both measures - the total number of cases to classify.
Aug
14
revised When are precision and recall inversely related?
added 11 characters in body
Aug
14
answered When are precision and recall inversely related?
Aug
3
comment What are the main differences between CLISP, ECL, and SBCL?
@mark Well, SBCL should work fine on Windows (thread support was fully merged some time ago). I'm not sure though, as I'm not using it there. Speaking about CLISP, can you run Hunchentoot in multi-threaded mode in it on Windows? That would be a real test of its maturity there. I'd say that now SBCL is at least as much portable. Also, Viaweb was written a decade ago, a lot of things has changed since then...
Aug
3
comment Parameters for calculating accuracy of part of speech tagger
if you count for each tag, then yes, precision/recall would be different, but if you count an aggregate metric it will be the same, because you have N points you need to predict and if you get M right than your precision is M/N - and for recall you have N suggestions and you get the same M right so it's also M/N
Aug
3
comment Parameters for calculating accuracy of part of speech tagger
You're right: you need to "manually" prepare the word-tag mapping. This is usually done by separating a part of the corpus (at least 20%) as test data and not using it for training. Also note, that in the case of POS tagging precision and recall are the same, so you can speak of just accuracy.
Aug
2
comment What are the main differences between CLISP, ECL, and SBCL?
@RainerJoswig to some extent you're right, although, there are objective facts listed here, like the availability/lack of support, or concurrency capabilities
Aug
2
answered What are the main differences between CLISP, ECL, and SBCL?
Jun
22
revised The purpose of Lisp syntax to model AST
added 3 characters in body
Apr
26
awarded Yearling
Apr
5
comment CL-FAD - redundancy in defsystem?
you're correct about the redundancy
Mar
5
comment socket descriptor closes after assignment
How does the RoutingManager class look like? Is it possible, that when you do myConnection->newConnections.pop_back(), the connection instance is finalized and closed?
Mar
5
comment socket descriptor closes after assignment
that's what accept is for
Feb
17
awarded Caucus
Feb
15
comment Array with initial content that is determined by function
Right, the results should be used as :initial-contents from the code in the question
Feb
15
answered Array with initial content that is determined by function
Feb
11
comment How does generative language model work in the natural language processing?
Yes, you generate a new random number for each sample. As for the first question: you've got it backwards. In this case you'll get yes is the generated number is below 0.67 and no otherwise. If you'll have 3 words: yes, no, maybe with equal probabilities, you'll need to take yes if the generated number is from 0 to 1/3, no if it's from 1/3 to 2/3, and maybe if it's above 2/3. You can think of it as placing you words on a [0,1] section, each word taking the size proportional to the probability.
Feb
11
revised How does generative language model work in the natural language processing?
added 25 characters in body
Feb
11
comment How does generative language model work in the natural language processing?
I see, I've updated the answer with one variant of how to do that.
Feb
11
revised How does generative language model work in the natural language processing?
added 548 characters in body
1 2 3 4 5