Need Help for developing Word2Vector Explanation in Tensrflow

[prakritidev]

I'm not able to understand the generate_batch function in the word2vector. If someone knows how is this function working please let me know or can explain that function in the script

Thanks

[mamonraab]

i will try to get some free time to explain these for you but right now i am so busy

[anujgupta82]

@prakritidev : Try this piece of code. If you understand skip-gram this can be understood. I feel examples in tensorflow are badly documented. I added comments to connect it with the process of skip-gram. Hope it helps.

num_skips=2 # no of words to be picked from the window
skip_window=1 #define how much will we see on one side of the word
batch_size = 16 

data_index = 0 #global circular counter over the data

def generate_batch(batch_size, num_skips, skip_window):
    # skip window is the amount of words we're looking at from each side of a given word
    # creates a single batch
    
    global data_index

    # num_skips => # of times we select a random word within the span? so no of picks should be integer
    assert batch_size % num_skips == 0
    assert num_skips <= 2 * skip_window # maximum no of samples picked is size of span

    batch = np.ndarray(shape=(batch_size), dtype=np.int32)
    labels = np.ndarray(shape=(batch_size, 1), dtype=np.int32)
    
    # e.g if skip_window = 2 then span = 5
    # span is the length of the whole frame we are considering for a single word (left + word + right)
    # skip_window is the length of one side
    
    span = 2 * skip_window + 1 # [ span defines the whole window, which is 2 * skip_window + the word itself ]
    
    # queue which add and pop at the end
    buffer = collections.deque(maxlen=span)
    
    #print "span = %d" %span
    
    #get words starting from index 0 to span
    for _ in range(span):
        #print "_ = %d" %_
        #print "data_index = %d" %data_index
        buffer.append(data[data_index]) # fill the buffer with elements in window
        data_index = (data_index + 1) % len(data)  #this is just to circle at the end of text corpus


    # num_skips => # of times we select a random word within the span
    # batch_size (8) and num_skips (2) (4 times)
    # batch_size (8) and num_skips (1) (8 times)
    
    #denotes the number of (input, output) pairs generated from the single window: [skip_window target skip_window]. 
    #So num_skips restrict the number of context words we would use as output words.
    
    #since num_skips = # of elements picked in each window, 
    # of windows = batch_size // num_skips
    # we iterate - for each window (i)
    #                   for each pick in given window
    #                            fit the pick in the batch
    
    # to fit the pick in the batch : jth element in ith pick = i * num_skips + j
    
    # from each window, we pick #num_skips elemnts, so to make a batch, how many windows we need ?
    num_of_windows = batch_size // num_skips 
    
    for i in range(num_of_windows):
        target = skip_window  # target label at the center of the buffer
        targets_to_avoid = [ skip_window ] # we only need to know the words around a given word, not the word itself
        
        for j in range(num_skips):
            while target in targets_to_avoid:
                # find a target word that is not the word itself
                # while loop will keep repeating until the algorithm find a suitable target word
                target = random.randint(0, span - 1)
                
            # add selected target to avoid_list for next time
            targets_to_avoid.append(target)
            
            # e.g. i=0, j=0 => 0; i=0,j=1 => 1; i=1,j=0 => 2
            batch[i * num_skips + j] = buffer[skip_window] # [skip_window] => middle element
            labels[i * num_skips + j, 0] = buffer[target]
            
        #populate the buffer with elements of next window - which is one elemnt on right
        buffer.append(data[data_index])
        data_index = (data_index + 1) % len(data)
        
    return batch, labels

Do let me know if u face any issue

[TorosFanny]

I'm still confused, but much more clear