类型错误：__init__() 缺少 1 个必需的位置参数：“名称”

Question

目前我正在研究扬声器分类，我必须执行 LSTM。 我想在 keras 中制作自定义优化器。为此，我以自定义方式重新实现了 SMORMS3，我的意思是我为 LSTM 定义了类。部分代码： 我从https://gist.github.com/hbredin/89b2249504a62712441e7ffec7de9518获取代码

from keras.legacy import interfaces
from keras.optimizers import Optimizer
import keras.backend as K
class SMORMS3(Optimizer):
    def __init__(self, lr=0.001, epsilon=1e-16, decay=0.,
                 **kwargs):
        super(SMORMS3, self).__init__(**kwargs)
        with K.name_scope(self.__class__.__name__):
            self.lr = K.variable(lr, name='lr')
            self.decay = K.variable(decay, name='decay')
            self.iterations = K.variable(0, dtype='int64', name='iterations')
        self.epsilon = epsilon
        self.initial_decay = decay
    @interfaces.legacy_get_updates_support
    def get_updates(self, loss, params):
        grads = self.get_gradients(loss, params)
        shapes = [K.get_variable_shape(p) for p in params]
        ms = [K.zeros(shape) for shape in shapes]
        vs = [K.zeros(shape) for shape in shapes]
        mems = [K.zeros(shape) for shape in shapes]
        self.weights = [self.iterations] + ms + vs + mems
        self.updates = [K.update_add(self.iterations, 1)]
        lr = self.lr
        if self.initial_decay > 0:
            lr *= (1. / (1. + self.decay * K.cast(self.iterations,                                                      K.dtype(self.decay))))
        for p, g, m, v, mem in zip(params, grads, ms, vs, mems):
            r = 1. / (1. + mem)
            new_m = (1. - r) * m + r * g
            new_v = (1. - r) * v + r * K.square(g)
            denoise = K.square(new_m) / (new_v + self.epsilon)
            new_p = p - g * K.minimum(lr, denoise) / (K.sqrt(new_v) + self.epsilon)
            new_mem = 1. + mem * (1. - denoise)

            self.updates.append(K.update(m, new_m))
            self.updates.append(K.update(v, new_v))
            self.updates.append(K.update(mem, new_mem))

            # Apply constraints.
            if getattr(p, 'constraint', None) is not None:
                new_p = p.constraint(new_p)

            self.updates.append(K.update(p, new_p))
        return self.updates

    def get_config(self):
        config = {'lr': float(K.get_value(self.lr)),
                  'decay': float(K.get_value(self.decay)),
                  'epsilon': self.epsilon}
        base_config = super(SMORMS3, self).get_config()
        return dict(list(base_config.items()) + list(config.items()))

这是我的 lstm 代码的一部分：

    from keras.layers.core import Dense
from keras.models import Sequential
from keras.layers import Bidirectional, TimeDistributed, Dropout
from keras.layers import LSTM
import numpy as np
import keras

def train_bilstm():

    model = Sequential()

    model.add(Bidirectional(LSTM(128, return_sequences=True)))
    model.add(Dropout(0.3))
    model.add(Bidirectional(LSTM(128, return_sequences=True)))
    model.add(Dropout(0.3))
    model.add(TimeDistributed(Dense(32)))
    model.add(TimeDistributed(Dense(32)))
    model.add(TimeDistributed(Dense(1, activation='sigmoid')))

    model.build(input_shape=(None, 137, 35))

    model.compile(loss=keras.losses.binary_crossentropy, optimizer=SMORMS3(0.001, 1e-16, 0.), metrics=['accuracy'])
    model.summary()
    print("subhash")
    all_x, all_y = load_dataset()
    print(all_y.shape, np.sum(all_y))

    subsample_all_x = []
    subsample_all_y = []
    for index in range(all_y.shape[0]):
        class_positive = sum(all_y[index])
        if class_positive > 5:
            subsample_all_x.append(all_x[index][np.newaxis, :, :])
            subsample_all_y.append(all_y[index])

    all_x = np.vstack(subsample_all_x)
    all_y = np.vstack(subsample_all_y)
    print(all_y.shape, np.sum(all_y))

    all_y = all_y[:, :, np.newaxis]

    indices = np.random.permutation(all_x.shape[0])
    all_x_random = all_x[indices]
    all_y_random = all_y[indices]

    datasize = all_x_random.shape[0]
    train_size = int(datasize*0.97)
    train_x = all_x_random[0:train_size]
    valid_x = all_x_random[train_size:]

    train_y = all_y_random[0:train_size]
    valid_y = all_y_random[train_size:]
    print('train over')

    my = model.fit(x=train_x, y=train_y, batch_size=256, epochs=50,
              validation_data=(valid_x, valid_y), shuffle=True)
    model.save('model_hindi_2.h5')
    def save_model(model, json_model_file, h5_model_file):
        # serialize model to JSON
        model_json = model.to_json()
        with open(json_model_file, "w") as json_file:
            json_file.write(model_json)
        # serialize weights to HDF5
        model.save_weights(h5_model_file)
        print("Saved model to disk")

#     model_name = 'speech_seg1'
#     json_model_file = '/content/drive/My Drive/SRU/model_hindi_2'+'.json'
#     h5_model_file = '/content/drive/My Drive/SRU/model_hindi_2'+'.h5'
#     save_model(model, json_model_file, h5_model_file)

我遇到了一个错误：

TypeError                                 Traceback (most recent call last)
<ipython-input-133-e28aeaab5d03> in <module>
----> 1 train_bilstm()

<ipython-input-132-f20f9c93ac69> in train_bilstm()
     20     model.build(input_shape=(None, 137, 35))
     21 
---> 22     model.compile(loss=keras.losses.binary_crossentropy, optimizer=SMORMS3(), metrics=['accuracy'])
     23     model.summary()
     24     print("subhash")

    <ipython-input-127-f43b8a02704f> in __init__(self, lr, epsilon, decay, **kwargs)
         20     def __init__(self, lr=0.001, epsilon=1e-16, decay=0.,
         21                  **kwargs):
    ---> 22         super(SMORMS3, self).__init__(**kwargs)
         23         with K.name_scope(self.__class__.__name__):
         24             self.lr = K.variable(lr, name='lr')
    
    TypeError: __init__() missing 1 required positional argument: 'name'

Answer 1

当您创建优化器类的实例时，您必须为其命名。这在tf.keras.optimizers.Optimizer` 的文档中进行了解释。

在这种情况下，修改此行以包含名称：

super(SMORMS3, self).__init__(name="SMORMS3", **kwargs)

更灵活的选择是修改您的 __init__ 签名以包含带有默认参数的 name 变量，这将允许您根据需要覆盖名称，就像您可以为预定义优化器：

def __init__(self, lr=0.001, epsilon=1e-16, decay=0., name="SMORMS3"
                 **kwargs):
    super(SMORMS3, self).__init__(name=name, **kwargs)