I try to reproduce the cnn/pcnn with tensorflow. In test phase, I treat a sentence as a special bag, and draw Precision-Recall curve. Surprisingly, single CNN/PCNN get better performance than CNN/PCNN+ONE/ATT。
Another question：the performance of various models look very similar，while in papers, the performance of various models look very different?
I try to reproduce the cnn/pcnn with tensorflow. In test phase, I treat a sentence as a special bag, and draw Precision-Recall curve. Surprisingly, single CNN/PCNN get better performance than CNN/PCNN+ONE/ATT。
Another question：the performance of various models look very similar，while in papers, the performance of various models look very different?