CODEWITHSUNDEEP
pythontensorflowkerasdeep-learningembedding
jartymcfly
jartymcfly

Reputation: 2005

I got Assertion Error when fitting a model in Keras after concatenating Embedded layers

I am new with Embedded Layers and I have a problem with that. I was trying to fit a neural network model, but first I encoded the categorical features of my dataset with an Embedding layer. Here is my code:

import numpy as np 
def preproc(X_train, categorical_vars, other_cols) : 

    input_list_train = []
    for c in categorical_vars :

        jjj = np.asarray(X_train[c].tolist())
        jjj = pd.factorize( jjj )[0]
        input_list_train.append( np.asarray(jjj)  )
        """
        raw_vals = np.unique(X_train[c])
        val_map = {}
        for i in range(len(raw_vals)):
            val_map[raw_vals[i]] = i       
        input_list_train.append(X_train[c].map(val_map).values)
        """
    #the rest of the columns
    input_list_train.append(X_train[other_cols].values)
    return input_list_train 

X_train = preproc(X_train, categorical_columns, numeric_columns)
X_validation = preproc(X_validation, categorical_columns, numeric_columns)
X_test = preproc(X_test, categorical_columns, numeric_columns)


from keras.layers import *
from keras.models import *

models = []

for categorical_var in categorical_columns:
    model = Sequential()
    model.reset_states( )
    no_of_unique_cat  = train_df[categorical_var].nunique()
    embedding_size = min(np.ceil((no_of_unique_cat)/2), 50 )
    embedding_size = int(embedding_size)
    model.add(  Embedding( no_of_unique_cat+1, embedding_size, input_length = 1 ) )
    model.add(Reshape(target_shape=(embedding_size,)))
    models.append( model )


model_rest = Sequential()
model_rest.add(Dense(  64 , input_dim = (train_df.shape[1] - len(categorical_columns)) ))
model_rest.reset_states( )
models.append(model_rest)

layer_nodes = len(X_train) + 1
dropout_hidden_layers = 0.5
output_nodes = 1

full_model = Sequential()
full_model.add(Concatenate(models))


full_model.add(Dense(units = layer_nodes, kernel_initializer = "uniform"))
full_model.add(Activation('relu'))
full_model.add(Dropout(dropout_hidden_layers))


full_model.add(Dense(units = layer_nodes, kernel_initializer = "uniform"))
full_model.add(Activation('relu'))
full_model.add(Dropout(dropout_hidden_layers))



full_model.add(Dense(units = output_nodes, kernel_initializer = "uniform",
                     activation = "sigmoid"))


full_model.compile(optimizer = "adam", loss = "binary_crossentropy", metrics = ["accuracy"])


history  =  full_model.fit(X_train, y_train, epochs =  200, batch_size = 20)

I got the next error:


Traceback (most recent call last):

  File "<ipython-input-28-d5c2b04c2cc3>", line 51, in <module>
    history  =  full_model.fit(X_train, y_train, epochs =  200, batch_size = 20)

  File "/home/javier/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 1154, in fit
    batch_size=batch_size)

  File "/home/javier/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 504, in _standardize_user_data
    self._set_inputs(x)

  File "/home/javier/anaconda3/lib/python3.7/site-packages/keras/engine/training.py", line 414, in _set_inputs
    assert len(inputs) == 1

AssertionError

Which could be the problem?

Upvotes: 0

Views: 360

Answers (1)

Marco Cerliani
Marco Cerliani

Reputation: 22021

for this kind of model, I suggest you use the keras functional structure. here your model adapted

models = []
inps = []

for categorical_var in categorical_columns:
    inp = Input((1,))
    no_of_unique_cat  = df[categorical_var].max()
    embedding_size = int(min(np.ceil((no_of_unique_cat)/2), 50 ))
    x = Embedding( no_of_unique_cat+1, embedding_size )(inp)
    x = Flatten()(x)
    inps.append( inp )
    models.append( x )


inp = Input(((df.shape[1] - len(categorical_columns)),))
x = Dense( 64 )(inp)
inps.append( inp )
models.append( x )

dropout_hidden_layers = 0.5
output_nodes = 1

x = Concatenate()(models)
x = Dense(128, kernel_initializer = "uniform")(x)
x = Activation('relu')(x)
x = Dropout(dropout_hidden_layers)(x)

x = Dense(64, kernel_initializer = "uniform")(x)
x = Activation('relu')(x)
x = Dropout(dropout_hidden_layers)(x)

out = Dense(units = output_nodes, kernel_initializer = "uniform", activation = "sigmoid")(x)

full_model = Model(inps, out)
full_model.compile(optimizer = "adam", loss = "binary_crossentropy", metrics = ["accuracy"])
full_model.summary()

I also provide u a dummy example available here: https://colab.research.google.com/drive/1LuyC_MosbU9wqvU9azjzBpG3c2oUDwMU?usp=sharing

Upvotes: 1

Related Questions