AttributeError: 'Model' object has no attribute 'predict_classes'

• I’m up-to-date with the latest release:

  pip install -U talos
  

• I’ve confirmed that my Keras model works outside of Talos.

I have an error when I run scan. I am suspecting that it’s because my model is not Sequential but talos expects it to be Sequential. And it’s because my input is a list of numpy arrays, not a numpy array.

Here is the model code (it works and run until the last line because I saw it printed “Best val_r2_keras score”):

# build and train then return the history and the model
def build_deep_model(X_train, y_train, X_val, y_val, params):
    X_train_keras = wave_df_to_keras_inputs(X_train) # this function converts array of shape (?, 604) to a list of 4 numpy arrays of shape (?,200,1), (?,200,1), (?,200,1), and (?,4)
    X_val_keras = wave_df_to_keras_inputs(X_val)
    input_x = keras.Input(shape=(200,1), name='x')
    input_y = keras.Input(shape=(200,1), name='y')
    input_z = keras.Input(shape=(200,1), name='z')
    input_q = keras.Input(shape=(4,), name='q')
    def compute_wave_features(x, name=None):
        x = keras.layers.Conv1D(32, 5, strides=2, padding='same', activation='relu')(x)
        x = keras.layers.Conv1D(32, 5, strides=2, padding='same', activation='relu')(x)
        x = keras.layers.Conv1D(32, 5, strides=2, padding='same', activation='relu')(x)
        x = keras.layers.Conv1D(32, 5, strides=2, padding='valid', activation='relu')(x)
        x = keras.layers.Flatten()(x)
        x = keras.layers.Dense(64, activation='relu', name=name)(x)
        return x
    feature_x = compute_wave_features(input_x, 'feature_x')
    feature_y = compute_wave_features(input_y, 'feature_y')
    feature_z = compute_wave_features(input_z, 'feature_z')
    features = keras.layers.Concatenate()([feature_x, feature_y, feature_z, input_q])
    features = keras.layers.Dense(32, activation='relu')(features)
    readout = keras.layers.Dense(y_train.shape[1], activation='linear', name='readout')(features)
    deep_model = keras.models.Model(inputs=[input_x, input_y, input_z, input_q], outputs=readout)
    deep_model.compile('adam', loss='mse', metrics=['mae', r2_keras])
    deep_model.summary()

    callbacks = [
        keras.callbacks.EarlyStopping(patience=3)
    ]
    history = deep_model.fit(X_train_keras, y_train, epochs=500, validation_data=[X_val_keras, y_val], callbacks=callbacks, verbose=2)
    best_r2 = np.max(history.history['val_r2_keras'])
    epochs = history.epoch[-1] + 1
    print("Best val_r2_keras score:", best_r2, 'Epochs:', epochs)
    return history, deep_model

Here is the code that produces the error (no params yet):

params = {
    
}
h = ta.Scan(X_train.values, y_train.values, params, build_deep_model, val_split=0.2, dataset_name='em-wave-pos', experiment_no='1')

Here is the output:

__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
x (InputLayer)                  (None, 200, 1)       0                                            
__________________________________________________________________________________________________
y (InputLayer)                  (None, 200, 1)       0                                            
__________________________________________________________________________________________________
z (InputLayer)                  (None, 200, 1)       0                                            
__________________________________________________________________________________________________
conv1d_13 (Conv1D)              (None, 100, 32)      192         x[0][0]                          
__________________________________________________________________________________________________
conv1d_17 (Conv1D)              (None, 100, 32)      192         y[0][0]                          
__________________________________________________________________________________________________
conv1d_21 (Conv1D)              (None, 100, 32)      192         z[0][0]                          
__________________________________________________________________________________________________
conv1d_14 (Conv1D)              (None, 50, 32)       5152        conv1d_13[0][0]                  
__________________________________________________________________________________________________
conv1d_18 (Conv1D)              (None, 50, 32)       5152        conv1d_17[0][0]                  
__________________________________________________________________________________________________
conv1d_22 (Conv1D)              (None, 50, 32)       5152        conv1d_21[0][0]                  
__________________________________________________________________________________________________
conv1d_15 (Conv1D)              (None, 25, 32)       5152        conv1d_14[0][0]                  
__________________________________________________________________________________________________
conv1d_19 (Conv1D)              (None, 25, 32)       5152        conv1d_18[0][0]                  
__________________________________________________________________________________________________
conv1d_23 (Conv1D)              (None, 25, 32)       5152        conv1d_22[0][0]                  
__________________________________________________________________________________________________
conv1d_16 (Conv1D)              (None, 11, 32)       5152        conv1d_15[0][0]                  
__________________________________________________________________________________________________
conv1d_20 (Conv1D)              (None, 11, 32)       5152        conv1d_19[0][0]                  
__________________________________________________________________________________________________
conv1d_24 (Conv1D)              (None, 11, 32)       5152        conv1d_23[0][0]                  
__________________________________________________________________________________________________
flatten_4 (Flatten)             (None, 352)          0           conv1d_16[0][0]                  
__________________________________________________________________________________________________
flatten_5 (Flatten)             (None, 352)          0           conv1d_20[0][0]                  
__________________________________________________________________________________________________
flatten_6 (Flatten)             (None, 352)          0           conv1d_24[0][0]                  
__________________________________________________________________________________________________
feature_x (Dense)               (None, 64)           22592       flatten_4[0][0]                  
__________________________________________________________________________________________________
feature_y (Dense)               (None, 64)           22592       flatten_5[0][0]                  
__________________________________________________________________________________________________
feature_z (Dense)               (None, 64)           22592       flatten_6[0][0]                  
__________________________________________________________________________________________________
q (InputLayer)                  (None, 4)            0                                            
__________________________________________________________________________________________________
concatenate_2 (Concatenate)     (None, 196)          0           feature_x[0][0]                  
                                                                 feature_y[0][0]                  
                                                                 feature_z[0][0]                  
                                                                 q[0][0]                          
__________________________________________________________________________________________________
dense_2 (Dense)                 (None, 32)           6304        concatenate_2[0][0]              
__________________________________________________________________________________________________
readout (Dense)                 (None, 3)            99          dense_2[0][0]                    
==================================================================================================
Total params: 121,123
Trainable params: 121,123
Non-trainable params: 0
__________________________________________________________________________________________________
Train on 9379 samples, validate on 2345 samples
Epoch 1/500
 - 3s - loss: 0.0905 - mean_absolute_error: 0.2260 - r2_keras: 0.2083 - val_loss: 0.0542 - val_mean_absolute_error: 0.1868 - val_r2_keras: 0.5264
Epoch 2/500
 - 3s - loss: 0.0552 - mean_absolute_error: 0.1869 - r2_keras: 0.5240 - val_loss: 0.0509 - val_mean_absolute_error: 0.1773 - val_r2_keras: 0.5562
Epoch 3/500
 - 3s - loss: 0.0517 - mean_absolute_error: 0.1783 - r2_keras: 0.5550 - val_loss: 0.0499 - val_mean_absolute_error: 0.1721 - val_r2_keras: 0.5638
Epoch 4/500
 - 3s - loss: 0.0503 - mean_absolute_error: 0.1748 - r2_keras: 0.5664 - val_loss: 0.0509 - val_mean_absolute_error: 0.1757 - val_r2_keras: 0.5543
Epoch 5/500
 - 3s - loss: 0.0495 - mean_absolute_error: 0.1728 - r2_keras: 0.5736 - val_loss: 0.0510 - val_mean_absolute_error: 0.1713 - val_r2_keras: 0.5549
Epoch 6/500
 - 3s - loss: 0.0481 - mean_absolute_error: 0.1692 - r2_keras: 0.5868 - val_loss: 0.0490 - val_mean_absolute_error: 0.1681 - val_r2_keras: 0.5725
Epoch 7/500
 - 3s - loss: 0.0465 - mean_absolute_error: 0.1654 - r2_keras: 0.5999 - val_loss: 0.0528 - val_mean_absolute_error: 0.1785 - val_r2_keras: 0.5370
Epoch 8/500
 - 3s - loss: 0.0450 - mean_absolute_error: 0.1626 - r2_keras: 0.6125 - val_loss: 0.0514 - val_mean_absolute_error: 0.1673 - val_r2_keras: 0.5548
Epoch 9/500
 - 3s - loss: 0.0432 - mean_absolute_error: 0.1578 - r2_keras: 0.6281 - val_loss: 0.0483 - val_mean_absolute_error: 0.1633 - val_r2_keras: 0.5810
Epoch 10/500
 - 3s - loss: 0.0414 - mean_absolute_error: 0.1531 - r2_keras: 0.6436 - val_loss: 0.0507 - val_mean_absolute_error: 0.1640 - val_r2_keras: 0.5581
Epoch 11/500
 - 3s - loss: 0.0382 - mean_absolute_error: 0.1460 - r2_keras: 0.6709 - val_loss: 0.0523 - val_mean_absolute_error: 0.1709 - val_r2_keras: 0.5457
Epoch 12/500
 - 3s - loss: 0.0360 - mean_absolute_error: 0.1410 - r2_keras: 0.6904 - val_loss: 0.0527 - val_mean_absolute_error: 0.1696 - val_r2_keras: 0.5398
Best val_r2_keras score: 0.5809796204190772 Epochs: 12
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-28-c4eaafcd036c> in <module>()
      2 
      3 }
----> 4 h = ta.Scan(X_train.values, y_train.values, params, build_deep_model, val_split=0.2, dataset_name='em-wave-pos', experiment_no='1')

c:\users\admin\anaconda3\lib\site-packages\talos\scan\Scan.py in __init__(self, x, y, params, model, dataset_name, experiment_no, x_val, y_val, val_split, shuffle, round_limit, grid_downsample, random_method, seed, search_method, reduction_method, reduction_interval, reduction_window, reduction_threshold, reduction_metric, reduce_loss, last_epoch_value, talos_log_name, clear_tf_session, functional_model, disable_progress_bar, print_params, debug)
    164         # input parameters section ends
    165 
--> 166         self._null = self.runtime()
    167 
    168     def runtime(self):

c:\users\admin\anaconda3\lib\site-packages\talos\scan\Scan.py in runtime(self)
    169 
    170         self = scan_prepare(self)
--> 171         self = scan_run(self)

c:\users\admin\anaconda3\lib\site-packages\talos\scan\scan_run.py in scan_run(self)
     19                      disable=self.disable_progress_bar)
     20     while len(self.param_log) != 0:
---> 21         self = scan_round(self)
     22         self.pbar.update(1)
     23     self.pbar.close()

c:\users\admin\anaconda3\lib\site-packages\talos\scan\scan_round.py in scan_round(self)
     45 
     46     _hr_out = run_round_results(self, _hr_out)
---> 47     self._val_score = get_score(self)
     48     write_log(self)
     49     self.result.append(_hr_out)

c:\users\admin\anaconda3\lib\site-packages\talos\metrics\score_model.py in get_score(self)
     36         # all other cases
     37         else:
---> 38             y_pred = self.keras_model.predict_classes(self.x_val)
     39 
     40         return Performance(y_pred, self.y_val, self.shape, self.y_max).result

AttributeError: 'Model' object has no attribute 'predict_classes'

I read a stackoverflow post about this error and they said it’s because predict_classes only exist in Sequential model.

I’m also confused because I’m building a regressor not a classifier, why is talos trying to predict classes?

How do I fix this issue?

[mikkokotila]

That will not work because it will just try to install the latest version from pypi. You have to follow:

pip uninstall talos 
pip install git+https://github.com/autonomio/talos@daily-dev  

[mikkokotila]

In the older version (which I think you are in) you still have to declare that the model is functional through Scan(functional_model=True) but in the latest version (0.4.4) this is no longer needed and you can run both Sequential and Functional models as-is. Can you make sure you have the latest version:

pip uninstall talos 
pip install git+https://github.com/autonomio/talos@daily-dev 

After that you should not have this problem anymore.