问题
I am working on a multilabel classification problem with KERAS. When i execute the code like this i get the following error:
ValueError: Error when checking target: expected activation_19 to have 2 dimensions, but got array with shape (32, 6, 6)
This is because of my lists full of "0" and "1" in the labels dictionary, which dont fit to keras.utils.to_categorical in return statement, as i learned recently. softmax cant handle more than one "1" as well.
I guess I first need a Label_Encoder and afterwards One_Hot_Encoding for labels, to avoid multiple "1" in labels, which dont go together with softmax.
I hope someone can give me a hint how to preprocess or transform labels data, to get the code fixed. I will appreciate a lot. Even a code snippet would be awesome.
csv looks like this:
Filename label1 label2 label3 label4 ... ID
abc1.jpg 1 0 0 1 ... id-1
def2.jpg 0 1 0 1 ... id-2
ghi3.jpg 0 0 0 1 ... id-3
...
current data to feed looks like this:
partition: {'train': ['id-1','id-2','id-3',...], 'validation': ['id-7','id-14','id-21',...]}
labels: {'id-0': [1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
'id-1': [0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
'id-2': [0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
...}
My images are converted to arrays and saved in single npy files in seperate folder as you can see in DataGenerator. id-1.npy, id-2.npy...
import numpy as np
import keras
from keras.layers import *
from keras.models import Sequential
class DataGenerator(keras.utils.Sequence):
'Generates data for Keras'
def __init__(self, list_IDs, labels, batch_size=32, dim=(224,224), n_channels=3,
n_classes=21, shuffle=True):
'Initialization'
self.dim = dim
self.batch_size = batch_size
self.labels = labels
self.list_IDs = list_IDs
self.n_channels = n_channels
self.n_classes = n_classes
self.shuffle = shuffle
self.on_epoch_end()
def __len__(self):
'Denotes the number of batches per epoch'
return int(np.floor(len(self.list_IDs) / self.batch_size))
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]
# Find list of IDs
list_IDs_temp = [self.list_IDs[k] for k in indexes]
# Generate data
X, y = self.__data_generation(list_IDs_temp)
return X, y
def on_epoch_end(self):
'Updates indexes after each epoch'
self.indexes = np.arange(len(self.list_IDs))
if self.shuffle == True:
np.random.shuffle(self.indexes)
def __data_generation(self, list_IDs_temp):
'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)
# Initialization
X = np.empty((self.batch_size, *self.dim, self.n_channels))
y = np.empty((self.batch_size, self.n_classes), dtype=int)
# Generate data
for i, ID in enumerate(list_IDs_temp):
# Store sample
X[i,] = np.load('Folder with npy files/' + ID + '.npy')
# Store class
y[i] = self.labels[ID]
return X, keras.utils.to_categorical(y, num_classes=self.n_classes)
-----------------------
# Parameters
params = {'dim': (224, 224),
'batch_size': 32,
'n_classes': 21,
'n_channels': 3,
'shuffle': True}
# Datasets
partition = partition
labels = labels
# Generators
training_generator = DataGenerator(partition['train'], labels, **params)
validation_generator = DataGenerator(partition['validation'], labels, **params)
# Design model
model = Sequential()
model.add(Conv2D(32, (3,3), input_shape=(224, 224, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
...
model.add(Flatten())
model.add(Dense(64))
model.add(Activation('relu'))
model.add(Dense(21))
model.add(Activation('softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
# Train model on dataset
model.fit_generator(generator=training_generator,
validation_data=validation_generator)
回答1:
Since you already have the labels as a vector of 21 elements of 0 and 1, you shouldn't use keras.utils.to_categorical in the function __data_generation(self, list_IDs_temp). Just return X and y.
回答2:
Ok i have a solution but i'm not sure that's the best .. :
from sklearn import preprocessing #for LAbelEncoder
labels_list = [x[1] for x in labels.items()] #get the list of all sequences
def convert(list):
res = int("".join(map(str, list)))
return res
label_int = [convert(i) for i in labels_list] #Convert each sequence to int
print(label_int) #E.g : [1,2,3] become 123
le = preprocessing.LabelEncoder()
le.fit(label_int)
labels = le.classes_ #Encode each int to only get the uniques
print(labels)
d = dict([(y,x) for x,y in enumerate(labels)]) #map each unique sequence to an label like 0, 1, 2, 3 ...
print(d)
labels_encoded = [d[i] for i in label_int] #get all the sequence and encode them with label obtained
print(labels_encoded)
labels_encoded = to_categorical(labels_encoded) #encode to_cagetorical
print(labels_encoded)
This is not really clean i think, but it's working
You need to change your last Dense layer to have a number of neurones equal to the lenght of the labels_encoded sequences.
For the predictions, you will have the dict "d" that map the predicted value to your orginal sequence style.
Tell me if you need clarifications !
For a few test sequences, it's gives you that :
labels = {'id-0': [1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1],
'id-1': [0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
'id-2': [0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1],
'id-3': [1, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1],
'id-4': [0, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1]}
[100100001100000001011, 10100001100000000001, 100001100010000001, 100100001100000001011, 10100001100000000001]
[100001100010000001 10100001100000000001 100100001100000001011]
{100001100010000001: 0, 10100001100000000001: 1, 100100001100000001011: 2}
[2, 1, 0, 2, 1]
[[0. 0. 1.]
[0. 1. 0.]
[1. 0. 0.]
[0. 0. 1.]
[0. 1. 0.]]
EDIT after clarification :
Ok i read a little more about the subject, once more the problem of softmax is that it will try to maximize a class while minize the others.
So i would sugest to keep your arrays of 21 ones's and zeros's but instead of using Softmax, use Sigmoid (to predict a probability between 0 and 1 for each class) with binary_crossentropy.
And use a treshold for your predictions :
preds = model.predict(X_test)
preds[preds>=0.5] = 1
preds[preds<0.5] = 0
Keep me in touch of the results !
来源：https://stackoverflow.com/questions/60322652/how-to-feed-datagenerator-for-keras-multilabel-issue