How to implement generator function for one hot encoding

Question

I implemented a generator function to yield one hot encoded vectors but the generator is actually throwing errors

I went for generator function to yield one hot encoded vectors because the latter would be used as inputs of deep learning lstm model. I am doing this to avoid excessive load and memory failures when trying to create one hot encoding out on data sets of very large size. However, i am not getting errors with the generator function. I need help to figure out where i am going wrong.

Code before:

X = np.zeros((len(sequences), seq_length, vocab_size), dtype=np.bool)
y = np.zeros((len(sequences), vocab_size), dtype=np.bool)
for i, sentence in enumerate(sequences):
    for t, word in enumerate(sentence):
        X[i, t, vocab[word]] = 1
    y[i, vocab[next_words[i]]] = 1

Here,

sequences = sentences generated from data set
seq_length = length of each sentence(this is constant)
vocab_size = number of unique words in dictionary

My program when run on the large data set produces,

sequences = 44073315
seq_length = 30
vocab_size = 124958

So, when the above code is directly used on the latter inputs, it gives beloe error.

Traceback (most recent call last):
  File "1.py", line 206, in <module>
    X = np.zeros((len(sequences), seq_length, vocab_size), dtype=np.bool)
MemoryError
(my_env) [rjagannath1@login ~]$

So, i tried creating a generator function(for testing) as below -

def gen(batch_size, no_of_sequences, seq_length, vocab_size):
    bs = batch_size
    ns = no_of_sequences
    X = np.zeros((batch_size, seq_length, vocab_size), dtype=np.bool)
    y = np.zeros((batch_size, vocab_size), dtype=np.bool)
    while(ns > bs):
        for i, sentence in enumerate(sequences):
            for t, word in enumerate(sentence):
                X[i, t, vocab[word]] = 1
            y[i, vocab[next_words[i]]] = 1
        print(X.shape())
        print(y.shape())
        yield(X, y)
        ns = ns - bs 

for item in gen(1000, 44073315, 30, 124958):
    print(item) 

But i get the below error -

File "path_of_file", line 247, in gen
    X[i, t, vocab[word]] = 1

IndexError: index 1000 is out of bounds for axis 0 with size 1000

What mistake am i doing in the generator function?

Answer 1

Modify as follows in your generator:

batch_i = 0
while(ns > bs):
    s = batch_i*batch_size
    e = (batch_i+1)*batch_size
    for i, sentence in enumerate(sequences[s:e]):

Basically, you want to run over windows of size batch_size so you are making a running slice through sequences which appears to be your entire dataset.

you also have to increment batch_i, place that just after yield, so add batch_i+=1