def cache_different_limits(limits):
prediction_texts = PredictionTexts([1, 2, 3, 4, 5, 7, 8])
ngram_predictions = NgramPredictor(4).batch_predict(prediction_texts)
perplexities = []
for limit in limits:
perplexity = predictor_perplexity(
interpolated_with_cache_probability(
0.22,
UnigramCachePredictor(limit=limit).batch_predict(prediction_texts),
ngram_predictions
)
)
perplexities.append(perplexity)
print limit, perplexity
plt.ylabel("Perplejidad")
plt.xlabel("Limite de Cache")
plt.plot(limits, perplexities, ".-")
plt.show()
def interpolate(cache_lambda, cache_probs, probs):
if len(cache_probs) == 1:
cache_probs = [cache_probs[0] / len(probs)] * len(probs)
elif len(probs) == 1:
probs = [probs[0] / len(cache_probs)] * len(cache_probs)
return interpolated_with_cache_probability(cache_lambda, cache_probs, probs)
