Commit 15c1882c authored by Rudolf Chrispens's avatar Rudolf Chrispens
Browse files

fixed vocab creation

parent f2825fe6

caption-lib/lstm/Readers/coco_reader.py

+58 −12
Original line number Diff line number Diff line
@@ -9,7 +9,7 @@ import json as jsonlib 
import re



class coco_reader:

    def load_data(self, data_path=helper.inputPath, filename=_config.path.input_train):

    def load_data_encoded(self, data_path=helper.inputPath, filename=_config.path.input_train):

        start_t = time.time()

        data = {}

        formatted_data = {}
@@ -62,6 +62,56 @@ class coco_reader: 
        print("#>\tLoad data elapsed time: %.2f" % (end_t - start_t))

        return formatted_data



    def load_data_decoded(self, data_path=helper.inputPath, filename=_config.path.input_train):

        start_t = time.time()

        data = {}

        formatted_data = {}



        features = []

        captions = []

        image_idxs = []



        with open(data_path + filename + '.pickle', 'rb') as handle:

            data = pickle.load(handle)



        for i, data_from_image in enumerate(data):

            for image_captions_dict in data_from_image['captions']:

                #5 times

                #Image Append

                features.append(data_from_image['vector'])

                #Caption Append

                startword = ['<START>']

                words = re.findall(r"\w+|[^\w\s]", image_captions_dict['caption'])

                words.append('<END>')

                words = self.add_padding(startword + words)

                captions.append(np.array(words))

                #Image ID Append

                image_idxs.append(np.array(data_from_image['id']))



            if(_config.reader.verbose and i < 3):

                print('\n#> Reader printing ' + filename + ' data:')

                print('#>\tcaptions:', data_from_image['captions'][0]['caption'])

                print('\tcaptions:', data_from_image['captions'][1]['caption'])

                print('\tcaptions:', data_from_image['captions'][2]['caption'])

                print('\tcaptions:', data_from_image['captions'][3]['caption'])

                print('\tcaptions:', data_from_image['captions'][4]['caption'])

                print('#>\timage_id:', data_from_image['id'])

                print('#>\tvector:', data_from_image['vector'])



        formatted_data['captions'] = np.array(captions)

        formatted_data['image_idxs'] = np.array(image_idxs)

        formatted_data['features'] = np.array(features)



        if(_config.reader.verbose):

            print("#> Data shape:")

            print("#>\tfeature set count complete:", formatted_data['features'].shape)

            print("#>\tcaption set count complete:", formatted_data['captions'].shape)

            print("#>\timageId set count complete:", formatted_data['image_idxs'].shape)



        end_t = time.time()

        print("#>\tLoad data elapsed time: %.2f" % (end_t - start_t))

        return formatted_data



    def add_padding(self, sequence):

        new_seq = sequence

        seq_length = len(sequence)
@@ -136,16 +186,17 @@ class coco_reader: 
            print("Elapse time: %.2f" % (end_t - start_t))

        return data



    """

    def create_vocab_from_filesystem(self, data_path: str=helper.inputPath, filename: str=_config.path.input_train):

        data = self.load_data_masa(data_path, filename)

        return self._create_vocab_from_data(data, filename)

        return self.create_vocab_from_data(data, filename)

    """



    def _create_vocab_from_data(self, data, save_filename_prefix):

    def create_vocab_from_data(self, data, save_filename_prefix):

        counter: int = 0

        vocab_id_word = {}

        for x in data:

            for i in x['captions']:

                for c in re.findall(r"\w+|[^\w\s]", i['caption']):

        for i in data['captions']:

            for c in i:

                if c not in vocab_id_word.values():

                    vocab_id_word[counter] = c

                    counter = counter + 1
@@ -154,12 +205,7 @@ class coco_reader: 
                        vocab_id_word[counter] = '<UNKN>'

                        counter = counter + 1

                        break

        vocab_id_word[counter] = '<END>'

        counter = counter + 1

        vocab_id_word[counter] = '<START>'

        counter = counter + 1

        vocab_id_word[counter] = '<NULL>'

        counter = counter + 1



        print('#>\tfinal vocab size: ', len(vocab_id_word))



        vocab_word_id = {v: k for k, v in vocab_id_word.items()}

caption-lib/lstm/lstm_main.py

+4 −3
Original line number Diff line number Diff line
@@ -29,15 +29,16 @@ def main(self, parameter_list): 


    # load train dataset

    current_reader = _reader.coco_reader()

    train_data = current_reader.load_data()

    word_to_idx = current_reader.create_vocab_from_filesystem(filename=_config.path.input_train)

    train_data_decoded = current_reader.load_data_decoded()

    word_to_idx = current_reader.create_vocab_from_data(data=train_data_decoded, save_filename_prefix=_config.path.input_train)

    train_data = current_reader.load_data_encoded()



    #test1 = current_reader.get_stored_vocab_id_word(filename_prefix=_config.path.read_train)

    #test2 = current_reader.get_stored_vocab_word_id(filename_prefix=_config.path.read_train)





    # load val dataset to print out bleu scores every epoch

    val_data = current_reader.load_data(filename=_config.path.input_validate)

    val_data = current_reader.load_data_encoded(filename=_config.path.input_validate)



    model = _model.coco_model(  word_to_idx,

                                current_reader,