jessicanada/cnn_spectrogram_algorithm

A method to classify spectrograms from raw EEG data using a convolutional neural network

cnn_spectrogram_algorithm's Issues

Changes to SpectrogramClassificationAlgorithm.ipynb

1. Add these two lines.

# train the model, run on test data and get output dat
NEWPATH = set_patient(0)
arch=resnet34                    <--- ADD THIS LINE
data = get_the_data(arch)          <--- ADD THIS INPUT TO FUNCTION
dat = train_the_model(arch,data)

2. Add input arch to def get_the_data(arch):

3. Use:

def train_the_model(arch,data):
    #train the model
    learn = ConvLearner.pretrained(arch,data,precompute=True)
    lr = 1e-2
    learn.fit(lr,1)
    learn.precompute = False
    learn.fit(1e-3,3,cycle_len=1)
    learn.unfreeze()
    lr = np.array([1e-4,1e-3,1e-2])
    cb = [EarlyStopping(learn,save_path='best_mod',patience = 6)]
    learn.fit(lr,6,cycle_len=1,cycle_mult=2,callbacks=cb)
    torch.save(learn.model.state_dict(),'test_saved_model.pkl')
    
    #get output predictions and probabilities
    log_preds_test = learn.predict(is_test=True)
    preds_test = np.argmax(log_preds_test,axis=1)
    probs_test = np.exp(log_preds_test[:,1])
    
    #make test: a dataframe of test image names, predictions, and probabilities
    test_names = np.empty_like(data.test_ds.fnames)
    for i in range(len(data.test_ds.fnames)):
        test_names[i] = data.test_ds.fnames[i]
        #temp = data.test_ds.fnames[i]
        #matchobj = re.search('.*im.*',temp)
        #test_names[i] = matchobj.group()
    test = pd.DataFrame(data = test_names,columns = ['image_number'])
    test['prediction'] = preds_test
    test['probability'] = probs_test
    
    return test