Develop Reference

How to give a sequece of numpy array as input to CNN

How can I reshape a sequence of arrays of shape (90,30,1662)? Meaning 90 arrays with 30 frames each and 1662 keypoints for each frames.And 90 array meaning 30 videos of numpy arrays for a word with 30 frames per video.
x_train, x_test, y_train, y_test=train_test_split(x, y, test_size=0.05)
x_train.shape ---->(85, 30, 1662)
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense
from tensorflow.keras.callbacks import TensorBoard
model = Sequential()
model.add(LSTM(64, return_sequences=True, activation='relu', input_shape=(30,1662)))
model.add(LSTM(128, return_sequences=True, activation='relu'))
model.add(LSTM(64, return_sequences=False, activation='relu'))
model.add(Dense(64, activation='relu'))
model.add(Dense(32, activation='relu'))
model.add(Dense(actions.shape[0], activation='softmax'))
How can I add CNN before the LSTM?

Reference: https://machinelearningmastery.com/cnn-long-short-term-memory-networks/
You can add CNN model first(with input shape=(30,90,1622)), and use LTSM model to encapsulate CNN model.
It will look like this:
cnn = Sequential()
cnn.add(Conv2D(your output size, (your filter size,your filter size),
activation='relu', padding='same', input_shape=(30,90,1622)))
cnn.add(MaxPooling2D(pool_size=(2, 2)))
cnn.add(Flatten())
model = Sequential()
model.add(TimeDistributed(cnn, ...)) # convert to LTSM type
model.add(LSTM(..))
model.add(Dense(...))

How to build a 1D CNN

I am trying to use a CNN for classification. My training data is shown in the picture below and has 9923 pieces of data with each piece containing 1k numeric values.
My current model has only around 10 percent accuracy and I am wondering if anyone knows if I am doing something wrong.
model = Sequential()
model.add(Conv1D(64,3, activation ='relu', input_shape= (1000, 1)))
model.add(MaxPooling1D(2))
model.add(Conv1D(64,3, activation ='relu'))
model.add(MaxPooling1D(pool_size=(2)))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dense(28, activation='softmax'))
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(X, Y, epochs = 30, validation_split = 0.1)

Validation data performing worse than training data in keras

I am training a CNN on some text data. The sentences are padded and embedded and fed to a CNN. The model architecture is:
model = Sequential()
model.add(Embedding(max_features, embedding_dims, input_length=maxlen))
model.add(Conv1D(128, 5, activation='relu'))
model.add(GlobalMaxPooling1D())
model.add(Dense(50, activation = 'relu'))
model.add(BatchNormalization())
model.add(Dense(50, activation = 'relu'))
model.add(BatchNormalization())
model.add(Dense(25, activation = 'relu'))
#model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
Any help would be appreciated.

You model is over-fitting so the best practice is:
add layers and preferably that goes in the power of 2
instead of
model.add(Dense(50, activation = 'relu'))
use
model.add(Dense(64, activation = 'relu'))
and go with 512 128 64 32 16
add some dropout layers preferably after two layers.
train on bigger data.

You can try removing BatchNormalization and adding more Convolutional and Pooling Layer that may increase your accuracy.
You can also check out this -:
https://forums.fast.ai/t/batch-normalization-with-a-large-batch-size-breaks-validation-accuracy/7940

Saving the specific layer from within a sequential Keras model

I am building an auto-encoder and training the model so the targeted output is the same as the input.
I am using a sequential Keras model. When I use model.predict I would like it to export the array from a specific layer (Dense256) not the output.
This is my current model:
model = Sequential()
model.add(Dense(4096, input_dim = x.shape[1], activation = 'relu'))
model.add(Dense(2048, activation='relu'))
model.add(Dense(1024, activation='relu'))
model.add(Dense(512, activation='relu'))
model.add(Dense(256, activation='relu'))
model.add(Dense(512, activation='relu'))
model.add(Dense(1024, activation='relu'))
model.add(Dense(2048, activation='relu'))
model.add(Dense(4096, activation='relu'))
model.add(Dense(x.shape[1], activation ='sigmoid'))
model.compile(loss = 'mean_squared_error', optimizer = 'adam')
history = model.fit(data_train,data_train,
verbose=1,
epochs=10,
batch_size=256,
shuffle=True,
validation_data=(data_test, data_test))

After training, create a new model (model2) from your trained model (model) ending in your desired layer.
You can do so either with layer name:
(In model.summary(), your dense's layer 'name' with 256 neurons is dense_5)
from keras.models import Model
model2= Model(model.input,model.get_layer('dense_5').output)
Or with layer order:
(your dense layer with 256 neurons is fifth in model.summary())
from keras.models import Model
model2= Model(model.input,model.layers[4].output)
Then you can use predict
preds=model2.predict(x)

layer.get_weights() returns the weights of a layer as a numpy array which can then be saved, for example with np.save.
To set the weights from a numpy array, layer.set_weights(weights) can be used.
You can access your layer either by name (model.get_layer(LAYER_NAME) or by its number (model.layers[LAYER_INDEX]).

How to make RNN with same input as CNN?

I have a working cnn with text
my X_train's shape is (39971, 10000) , y_train: (39971, 4)
max_words = 10000
model = Sequential()
model.add(Dense(512, input_shape=(max_words,), activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(256, activation='sigmoid'))
model.add(Dropout(0.5))
model.add(Dense(4, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
this works so far.
And this his is how I'm trying to make the RNN:
model = Sequential()
model.add(Embedding(max_words, 128))
model.add(LSTM(64, return_sequences=True, dropout=0.5, recurrent_dropout=0.5))
model.add(Dense(4, activation='softmax'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
I can compile the model but when I try to run model.fit with my training data i get an error:
ValueError: Error when checking target: expected dense_42 to have 3 dimensions, but got array with shape (39971, 4)
What does this mean? How could I fix this?
As this suggest I might need to add input_shape but I'm not sure with what values.

The LSTM layer is currently returning the full sequence, so the output of the network has shape (batch_size, nb_timesteps, 4). I presume you want to use the last LSTM output only. If that's the case, set return_sequences=False.