为Keras编写自定义数据生成器 [英] Write custom Data Generator for Keras

问题描述

我将每个数据点都存储在.npy文件中，并带有shape=(1024,7,8).我想通过类似于ImageDataGenerator的方式将它们加载到Keras模型中，所以我编写并尝试了不同的自定义生成器，但是它们都不起作用，这是我从

I have each datapoint stored in a .npy file, with shape=(1024,7,8). I want to load them to a Keras model by a manner similar to ImageDataGenerator, so I wrote and tried different custom generators but none of them work, here is one I adapted from this

def find(dirpath, prefix=None, suffix=None, recursive=True):
    """Function to find recursively all files with specific prefix and suffix in a directory
    Return a list of paths
    """
    l = []
    if not prefix:
        prefix = ''
    if not suffix:
        suffix = ''
    for (folders, subfolders, files) in os.walk(dirpath):
        for filename in [f for f in files if f.startswith(prefix) and f.endswith(suffix)]:
            l.append(os.path.join(folders, filename))
        if not recursive:
            break
    l
    return l

def generate_data(directory, batch_size):
    i = 0
    file_list = find(directory)
    while True:
        array_batch = []
        for b in range(batch_size):
            if i == len(file_list):
                i = 0
                random.shuffle(file_list)
            sample = file_list[i]
            i += 1

            array = np.load(sample)
            array_batch.append(array)

        yield array_batch

我发现缺少标签，因此无法使用fit_generator将其放入模型中.假设我可以将标签存储在numpy数组中，如何将其添加到此生成器中?

I found this lacks of the label, so it won't be fit into the model using fit_generator . How can I add the label into this generator, given that I can store them in a numpy array?

推荐答案

from tensorflow.python.keras.utils import Sequence
import numpy as np   

class Mygenerator(Sequence):
    def __init__(self, x_set, y_set, batch_size):
        self.x, self.y = x_set, y_set
        self.batch_size = batch_size

    def __len__(self):
        return int(np.ceil(len(self.x) / float(self.batch_size)))

    def __getitem__(self, idx):
        batch_x = self.x[idx * self.batch_size:(idx + 1) * self.batch_size]
        batch_y = self.y[idx * self.batch_size:(idx + 1) * self.batch_size]

        # read your data here using the batch lists, batch_x and batch_y
        x = [my_readfunction(filename) for filename in batch_x] 
        y = [my_readfunction(filename) for filename in batch_y]
        return np.array(x), np.array(y)

这篇关于为Keras编写自定义数据生成器的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！