Part 1: Treasure hunt

You recently won a all-paid trip to a lush tropical island. While you were wandering around, you found an ancient treasure map pointing to a great treasure, which had a few secret messages written using 1s and 0s. Having just taken this course, you instantly recognize that it is a sequence of onehot encoded vectors. You have also been lucky to find the word to index mapping to know which word refers to which ID.

Now you need to decrypt the secret message and find out what this map is saying. You have been provided with a treasure_map which is a number of sentences by number of words by onehot vector length matrix. You have also been provided with the index2word Python dictionary that maps an ID to a word.

This exercise is part of the course

Machine Translation with Keras

View Course

Exercise instructions

Get the word IDs for the onehot encoded vectors in the treasure_map (onehot vector dimension is the last dimension).
Get the sequence length (i.e. number of time steps) from treasure_map and assign to seq_len.
Get the word ID in the i-th sentence at t-th position.
Append the String word (i.e. not the word ID) corresponding to wid to the list words.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Get the word IDs from the treasure map
word_ids = ____.____(____, axis=____)
# Get the sequence length from the treasure map
seq_len = treasure_map.shape[____]

for i in range(treasure_map.shape[0]):
	words = []
	for t in range(seq_len):
      	# Get the word ID for the i-th sentence and t-th position
	    wid = word_ids[i, ____]
	    if wid != 0:
          	# Append the word corresponding to wid
	        words.append(____[____])
	print("Instruction ", i+1, ": ", ' '.join(words))

Edit and Run Code