SQuAD-task

Machine Reading Comprehension task

Model: DCN Encoder + Answer Pointer Approach

Hyper parameter values (final) and Results

• Hyper-Parameters: batch_size = 40, embed_size = 100, lstm_units = 100, dropout = 0.07, learning_rate = 0.001, Optimiser = AdamOptimizer
• Results: Train-Data : Exact-Match = 57.433 %, F1 = 71.077 %
  Dev-Data : Exact-Match = 59.0909 %, F1 = 71.467 %