Dear community , I need your help , I'm training my model in order to classify sleep stages , after extracting features from my signal I collected the features(X) in a DataFrame with shape(335,48) , and y (labels) in shape of (335,)
this is my code :
def get_base_model(): inp = Input(shape=(335,48)) img_1 = Convolution1D(16, kernel_size=5, activation=activations.relu, padding="valid")(inp) img_1 = Convolution1D(16, kernel_size=5, activation=activations.relu, padding="valid")(img_1) img_1 = MaxPool1D(pool_size=2)(img_1) img_1 = SpatialDropout1D(rate=0.01)(img_1) img_1 = Convolution1D(32, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = Convolution1D(32, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = MaxPool1D(pool_size=2)(img_1) img_1 = SpatialDropout1D(rate=0.01)(img_1) img_1 = Convolution1D(32, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = Convolution1D(32, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = MaxPool1D(pool_size=2)(img_1) img_1 = SpatialDropout1D(rate=0.01)(img_1) img_1 = Convolution1D(256, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = Convolution1D(256, kernel_size=3, activation=activations.relu, padding="valid")(img_1) img_1 = GlobalMaxPool1D()(img_1) img_1 = Dropout(rate=0.01)(img_1) dense_1 = Dropout(0.01)(Dense(64, activation=activations.relu, name="dense_1")(img_1)) base_model = models.Model(inputs=inp, outputs=dense_1) opt = optimizers.Adam(0.001) base_model.compile(optimizer=opt, loss=losses.sparse_categorical_crossentropy, metrics=['acc']) model.summary() return base_model model=get_base_model() test_loss, test_acc = model.evaluate(Xtest, ytest, verbose=0) model.fit(X,y) print('\nTest accuracy:', test_acc)
I got the error : Input 0 is incompatible with layer model_16: expected shape=(None, 335, 48), found shape=(None, 48)
you can have in this picture an idea about my data shape :
So you need to code your network to get the input of size (none, 1, 48). Each feature is of dimension 1x48, while 'none' would take up the size of the number of sample points (335 in your case). Hence, your input would be of shape 335x1x48. So, modify the input layer of your network to expect input of size 1x48 instead of 335x48, and provide input as 335x1x48.
Aakarsh Malhotra Thank you for answering , actually the problem is solved but I receive a low accuracy how can I increase it ?
Wassim Diai If your data set is balanced(number of examples for each class is the same) try to tune the hyper-parameters. Otherwise first you need to solve the imbalance data set problem.
Maryam Edalati thank you for answering , data is already balanced , I have 67 sample for each class 67*5=335 .
To improve accuracy, you need to identify which part of the ML pipeline can give you significant improvements if worked on. So try to use actual values instead of learnt values on different stages and see how the accuracy changes
Shadman Sakib Arnob Thank you for answering , can you please give more details . thank you again
With just 67 samples for each class and each sample consisting of 48 features I would recommend any classic data mining algorithm like RandomForest or SVM. You can even try the KNN. The number of samples is probably too low for your really deep neural network. It would work better if you significantly increase the number of samples. You may also try data augmentation but I am not sure if it would be enough.
Pawel Kasprowski thank you for answering I tried svm , I got 95% training accuracy but 60% test accuracy is it a problem ?
- Badges
- Science topic