def _compute_metrics(model, testset):
all_utterances = list(flatten(testset.values(),
set)) # Get all unique responses
context_to_weighted_responses = _get_context_to_weighted_responses(
model, testset, all_utterances)
test_set_size = len(all_utterances)
metrics = {
'mean_ap':
compute_retrieval_metric_mean(compute_average_precision,
testset,
context_to_weighted_responses,
top_count=test_set_size),
'mean_recall@10':
compute_retrieval_metric_mean(compute_recall_k,
testset,
context_to_weighted_responses,
top_count=10),
'mean_recall@25%':
compute_retrieval_metric_mean(compute_recall_k,
testset,
context_to_weighted_responses,
top_count=test_set_size // 4)
}
print('Test set size = %i' % test_set_size)
for metric_name, metric_value in iteritems(metrics):
print('%s = %s' % (metric_name, metric_value))
def _compute_metrics(model, testset):
all_utterances = list(flatten(testset.values(), set)) # Get all unique responses
context_to_weighted_responses = _get_context_to_weighted_responses(model, testset, all_utterances)
test_set_size = len(all_utterances)
metrics = {
'mean_ap':
compute_retrieval_metric_mean(
compute_average_precision, testset, context_to_weighted_responses, top_count=test_set_size),
'mean_recall@10':
compute_retrieval_metric_mean(compute_recall_k, testset, context_to_weighted_responses, top_count=10),
'mean_recall@25%':
compute_retrieval_metric_mean(
compute_recall_k, testset, context_to_weighted_responses, top_count=test_set_size / 4)
}
print 'Test set size = %i' % test_set_size
for metric_name, metric_value in metrics.iteritems():
print '%s = %s' % (metric_name, metric_value)
def _get_ngrams(self, tokenized_line):
ngrams = [nltk.ngrams(tokenized_line, i) for i in xrange(1, self._max_ngram_len + 1)]
return flatten(ngrams, constructor=set)
def _get_ngrams(self, tokenized_line):
ngrams = [nltk.ngrams(tokenized_line, i) for i in xrange(1, self._max_ngram_len + 1)]
return flatten(ngrams, constructor=set)