Conv1D for binary vector classification: can't get input dimensions correct

[levy....@gmail.com]

I am trying to use a keras 1D CNN to classify 300-dimensional vectors as either 1 or 0 based on a training set of around 2600 vectors. Each vector represents a document (based on gensim's doc2vec). Right now, here is the model:

def initialize_model(self, lr=0.01):
    """Initialize the neural network.
    @type self: Pydeology
    @type lr: float
        Learning rate.
    @rtype: None
    """

self.model = Sequential()

self.model.add(Conv1D(n_filters, kernel_size, padding='valid',

                      input_shape=(n_samples, input_dim, 1),

                      activation='relu', strides=strides))

self.model.add(MaxPooling1D(pool_size=2, strides=None, padding='valid'))

self.model.add(Dropout(0.2))

self.model.add(Conv1D(n_filters, kernel_size, padding='valid',

                      activation='relu', strides=strides))

self.model.add(MaxPooling1D(pool_size=2, strides=None, padding='valid'))

self.model.add(Dropout(0.2))

self.model.add(Dense(1, activation='sigmoid'))

self.model.compile(optimizer=Adam(lr), loss='binary_crossentropy',

                   metrics=['accuracy'])

print(self.model.summary())

Where self refers to a class I've created to contain the data arrays and the neural network itself. I then get the following error when I run the program:

ValueError: Input 0 is incompatible with layer conv1d_1: expected ndim=3, found ndim=4

What might I be doing wrong?

Thanks!

[kuangs...@gmail.com]

change the "input_shape=(n_samples, input_dim, 1)" to 
"input_shape=( input_dim, 1)"

在 2017年8月9日星期三 UTC+2下午5:14:39，Benjamin Levy写道：