使用具有多个输入的网络进行Hyperas网格搜索

Question

使用具有多个输入的网络进行Hyperas网格搜索

10

我目前在使用hyperas优化器时遇到了多输入网络的问题。

以下是我的实现方式：

def data():
    X_train, Y_train = next(train_generator())
    X_test, Y_test = next(test_generator())

    datagen = ImageDataGenerator()
    train_list = []
    for input in X_train:
        train_list.append(datagen.fit(input))

    return datagen, train_list, Y_train, X_test, Y_test

我正在使用数据生成器，因为所有的数据都无法容纳在内存中。

基于他们制作的数据示例，我写了这个：

def fws(datagen, X_train, Y_train, X_test, Y_test):
    # Input shape: (batch_size,40,45,3)
    # Output shape: (1,15,50)
    # Number of units in conv_feature_map = splitd
    filter_size = 8
    pooling_size = 28
    stride_step = 2
    pool_splits = ((splits - pooling_size)+1)/2
    temp_list = []
    sun_temp_list = []
    conv_featur_map = []
    pool_feature_map = []
    print "Printing shapes"

    list_of_input = [Input(shape = (window_height,total_frames_with_deltas,3)) for i in range(splits)]

    # Convolution
    shared_conv = Conv2D(filters = 150, kernel_size = (filter_size,45), activation='relu')
    for i in range(splits):
        conv_featur_map.append(shared_conv(list_of_input[i]))

    # Pooling
    input = Concatenate()(conv_featur_map)
    input = Reshape((splits,-1))(input)
    pooled = MaxPooling1D(pool_size = pooling_size, strides = stride_step)(input)

    # fc
    dense1 = Dense(units = 1000, activation = 'relu',    name = "dense_1")(pooled)
    dense2 = Dense(units = 1000, activation = 'relu',    name = "dense_2")(dense1)
    dense3 = Dense(units = 50 , activation = 'softmax', name = "dense_3")(dense2)

    model = Model(inputs = list_of_input , outputs = dense3)
    sgd = keras.optimizers.SGD(lr = {{uniform(0, 1)}}, decay = {{uniform(0, 1)}}, momentum = {{uniform(0, 1)}}, nesterov = True)
    model.compile(loss="categorical_crossentropy", optimizer=sgd , metrics = [metrics.categorical_accuracy])

    hist_current = model.fit_generator(datagen.flow(X_train, Y_train),
                        steps_per_epoch=32,
                        epochs = 1000,
                        verbose = 1,
                        validation_data = (X_test, Y_test),
                        validation_steps=32,
                        pickle_safe = True,
                        workers = 4)

    score, acc = model.evaluate(X_test, Y_test, verbose=0)

    return {'loss': -acc, 'status': STATUS_OK, 'model': model}

这个网络的特殊之处在于它接受多个输入。我原本可以让它只接受一个输入并使用 Lambda 层来分割，但由于分割过程相当繁琐，所以我决定将其分割后存储，并分割后进行馈送，从而创建 33 个输入。否则该网络就是非常标准的。

（网络可视化）

if __name__ == '__main__':

    datagen, X_train, Y_train, X_test, Y_test = data()

    best_run, best_model = optim.minimize(model=fws,
                                          data=data,
                                          algo=tpe.suggest,
                                          max_evals=5,
                                          trials=Trials())

    print("Evalutation of best performing model:")
    print(best_model.evaluate(X_test, Y_test))

这是我开始优化的地方，也是我收到错误消息的地方:

Traceback (most recent call last):
  File "keras_cnn_phoneme_original_fit_generator_hyperas.py", line 211, in <module>
    trials=Trials())
  File "/usr/local/lib/python2.7/dist-packages/hyperas/optim.py", line 43, in minimize
    notebook_name=notebook_name, verbose=verbose)
  File "/usr/local/lib/python2.7/dist-packages/hyperas/optim.py", line 63, in base_minimizer
    model_str = get_hyperopt_model_string(model, data,functions,notebook_name, verbose, stack)
  File "/usr/local/lib/python2.7/dist-packages/hyperas/optim.py", line 130, in get_hyperopt_model_string
    imports = extract_imports(cleaned_source, verbose)
  File "/usr/local/lib/python2.7/dist-packages/hyperas/utils.py", line 44, in extract_imports
    import_parser.visit(tree)
  File "/usr/lib/python2.7/ast.py", line 241, in visit
    return visitor(node)
  File "/usr/lib/python2.7/ast.py", line 249, in generic_visit
    self.visit(item)
  File "/usr/lib/python2.7/ast.py", line 241, in visit
    return visitor(node)
  File "/usr/local/lib/python2.7/dist-packages/hyperas/utils.py", line 14, in visit_Import
    if (self._import_asnames(node.names)!=''):
  File "/usr/local/lib/python2.7/dist-packages/hyperas/utils.py", line 36, in _import_asnames
    return ''.join(asname)
TypeError: sequence item 0: expected string, NoneType found

我不确定应该如何解释这个错误。这是实现错误还是库中的错误？我不知道...

一个最小可行示例：

import numpy as np
import re
from keras.utils import np_utils
from keras import metrics
import keras
from keras.models import Sequential
from keras.optimizers import SGD
import scipy
from keras.layers.core import Dense, Activation, Lambda, Reshape,Flatten
from keras.layers import Conv1D,Conv2D,MaxPooling2D, MaxPooling1D, Reshape
#from keras.utils.visualize_util import plot
from keras.utils import np_utils
from keras.models import Model
from keras.layers import Input, Dense
from keras.layers import Dropout
from keras import backend as K
from keras.layers.merge import Concatenate
from keras.models import load_model
from keras.utils import plot_model
from keras.preprocessing.image import ImageDataGenerator
import math
import random
from keras.callbacks import ModelCheckpoint
import tensorflow as tf
from hyperopt import Trials, STATUS_OK, tpe
from hyperas import optim
from hyperas.distributions import uniform

def train_generator():
    while True:
        train_input = np.random.randint(100,size=(1,33,8,45,3))
        train_input_list = np.split(train_input,33,axis=1)

        for i in range(len(train_input_list)):
            train_input_list[i] = train_input_list[i].reshape(1,8,45,3)

        train_output = np.random.randint(100,size=(1,3,50))
        yield (train_input_list, train_output)

def test_generator():
    while True:
        test_input = np.random.randint(100,size=(1,33,8,45,3))
        test_input_list = np.split(test_input,33,axis=1)

        for i in range(len(test_input_list)):
            test_input_list[i] = test_input_list[i].reshape(1,8,45,3)

        test_output = np.random.randint(100,size=(1,3,50))

        yield (test_input_list, test_output)

def data():
    X_train, Y_train = next(train_generator())
    X_test, Y_test = next(test_generator())

    datagen = ImageDataGenerator()
    train_list = []
    for input in X_train:
        train_list.append(datagen.fit(input))

    return datagen, train_list, Y_train, X_test, Y_test

def fws(datagen, X_train, Y_train, X_test, Y_test):
    # Input shape: (batch_size,40,45,3)
    # Output shape: (1,15,50)
    # Number of units in conv_feature_map = splitd
    filter_size = 8
    pooling_size = 28
    stride_step = 2
    pool_splits = ((splits - pooling_size)+1)/2
    temp_list = []
    sun_temp_list = []
    conv_featur_map = []
    pool_feature_map = []
    print "Printing shapes"

    list_of_input = [Input(shape = (8,45,3)) for i in range(33)]

    # Convolution
    shared_conv = Conv2D(filters = 150, kernel_size = (filter_size,45), activation='relu')
    for i in range(splits):
        conv_featur_map.append(shared_conv(list_of_input[i]))

    # Pooling
    input = Concatenate()(conv_featur_map)
    input = Reshape((splits,-1))(input)
    pooled = MaxPooling1D(pool_size = pooling_size, strides = stride_step)(input)

    #reshape = Reshape((3,-1))(pooled)

    #fc
    dense1 = Dense(units = 1000, activation = 'relu',    name = "dense_1")(pooled)
    dense2 = Dense(units = 1000, activation = 'relu',    name = "dense_2")(dense1)
    dense3 = Dense(units = 50 , activation = 'softmax', name = "dense_3")(dense2)


    model = Model(inputs = list_of_input , outputs = dense3)
    sgd = keras.optimizers.SGD(lr = {{uniform(0, 1)}}, decay = {{uniform(0, 1)}}, momentum = {{uniform(0, 1)}}, nesterov = True)
    model.compile(loss="categorical_crossentropy", optimizer=sgd , metrics = [metrics.categorical_accuracy])

    hist_current = model.fit_generator(datagen.flow(X_train, Y_train),
                        steps_per_epoch=32,
                        epochs = 1000,
                        verbose = 1,
                        validation_data = (X_test, Y_test),
                        validation_steps=32,
                        pickle_safe = True,
                        workers = 4)

    score, acc = model.evaluate(X_test, Y_test, verbose=0)

    return {'loss': -acc, 'status': STATUS_OK, 'model': model}

if __name__ == '__main__':

    datagen, X_train, Y_train, X_test, Y_test = data()

    best_run, best_model = optim.minimize(model=fws,
                                          data=data,
                                          algo=tpe.suggest,
                                          max_evals=5,
                                          trials=Trials())

    print("Evalutation of best performing model:")
    print(best_model.evaluate(X_test, Y_test))

- J.Down

2

你的最小工作示例有多个问题。splits未定义，许多变量被赋值但从未使用过。你能再检查一下吗？ - Vivek Kumar

错误信息“TypeError: sequence item 0: expected string, NoneType found”听起来像是您的输入序列的第一个元素没有被正确读取。第一步可能是检查您的数据（缺失值？），并检查您如何读取数据。 - StatsSorceress

1个回答

网页内容由stack overflow 提供, 点击上面的

可以查看英文原文，
原文链接

- spadarian · Answer 1

我认为你的问题与 data 函数有关。

这里：

datagen, X_train, Y_train, X_test, Y_test = data()

X_train 对应于 train_list，它是通过以下方式生成的：

datagen = ImageDataGenerator()
train_list = []
for input in X_train:
    train_list.append(datagen.fit(input))

所以train_list不是一个数组。它只是一个充满了datagen.fit返回值（即None）的列表。