Scipy Optimizer Interface for Keras

Question

I tried to use the implementation of Scipy optimizer interface for keras (https://github.com/pedro-r-marques/keras-opt) like its example but I got the following error:

'Sequential' object has no attribute '_configure_steps_per_execution'

I've searched a lot but I couldn't find anything related to _configure_steps_per_execution function.

Can anyone help me?

Answer 1

Works fine. Make sure you fully fill this requirement.

!git clone https://github.com/pedro-r-marques/keras-opt.git
import sys 
sys.path.insert(0, "/content/keras-opt")

from keras_opt import scipy_optimizer
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, InputLayer

model = Sequential()
model.add(InputLayer(input_shape=(4,)))
model.add(Dense(1, use_bias=False))
model.compile(loss='mse')

#%%
# Generate test data

import numpy as np

np.random.seed(42)
X = np.random.uniform(size=40).reshape(10, 4)
y = np.dot(X, np.array([1, 2, 3, 4])[:, np.newaxis])

#%%
# Use scipy.optimize to minimize the cost
model.train_function = scipy_optimizer.make_train_function(
            model, maxiter=20)
history = model.fit(X, y)

#%%
# Show weights.
model.trainable_weights

0/Unknown - 0s 0s/step - loss: 2.1390e-11Optimization terminated successfully.
Current function value: 0.000000
Iterations: 6
Function evaluations: 14
Gradient evaluations: 14
1/1 [==============================] - 0s 313ms/step - loss: 2.1390e-11
[<tf.Variable 'dense_1/kernel:0' shape=(4, 1) dtype=float32, numpy=
 array([[1.0000011],
        [2.000016 ],
        [2.9999967],
        [3.9999871]], dtype=float32)>]