Getting Error in the LSTM model for BitPredict project; Lecture Number: 326

[arrshsh]

As shown in the course, I have followed the same approach of loading the best performing model by using the model checkpoint callback. It worked well with the previous 4 experiments. But, when I did the same for LSTM, I am getting these errors:

---------------------------------------------------------------------------

NameError                                 Traceback (most recent call last)

[<ipython-input-386-3ddaf9b594a0>](https://localhost:8080/#) in <cell line: 0>()
      4 # Load
      5 loaded_model = tf.keras.models.load_model("./model_experiments/model_5_LSTM.keras", safe_mode=False)
----> 6 loaded_model.evaluate(test_windows, test_labels)

1 frames

[/usr/local/lib/python3.11/dist-packages/keras/src/utils/traceback_utils.py](https://localhost:8080/#) in error_handler(*args, **kwargs)
    120             # To get the full stack trace, call:
    121             # `keras.config.disable_traceback_filtering()`
--> 122             raise e.with_traceback(filtered_tb) from None
    123         finally:
    124             del filtered_tb

[/usr/local/lib/python3.11/dist-packages/keras/src/utils/python_utils.py](https://localhost:8080/#) in <lambda>(x)
      9 inputs = layers.Input(shape=(WINDOW_SIZE,))
     10 x = layers.Lambda(
---> 11     lambda x: tf.expand_dims(x, axis=1),
     12     output_shape=(1, WINDOW_SIZE)
     13 )(inputs)

NameError: Exception encountered when calling Lambda.call().

name 'tf' is not defined

Arguments received by Lambda.call():
  • inputs=tf.Tensor(shape=(None, 7), dtype=float32)
  • mask=None
  • training=False

I know it may seem as if TensorFlow is not imported, but the thing is that I ave imported it. This is my code:

passing shape = (WINDOW_SIZE,) instead of shape = (WINDOW_SIZE) because the shape is one-dimensional and TensorFlow expects a tuple

Explicit output shape# x = layers.LSTM(128,

While normally we do not need to explicitly mention the output shape for a lamda layer, sometimes TensorFlow does need it

x = layers.LSTM(128,return_sequences = True)(x)

import tensorflow as tf
from tensorflow.keras import layers

Model definition

inputs = layers.Input(shape=(WINDOW_SIZE,))
x = layers.Lambda(
lambda x: tf.expand_dims(x, axis=1),
output_shape=(1, WINDOW_SIZE)
)(inputs)
x = layers.LSTM(128, activation="relu")(x)
x = layers.Dense(32, activation="relu")(x)
output = layers.Dense(HORIZON)(x)
model_5 = tf.keras.Model(inputs, output, name="model_5_LSTM.keras")

Compile the model

model_5.compile(
loss="mae",
optimizer=tf.keras.optimizers.Adam()
)

Fit the model

model_5.fit(train_windows, train_labels, epochs = 100, verbose = 1, batch_size = 128, validation_data = (test_windows, test_labels), callbacks = [create_model_checkpoint(model_name = model_5.name)])

Load

loaded_model = tf.keras.models.load_model("./model_experiments/model_5_LSTM.keras", safe_mode=False)
loaded_model.evaluate(test_windows, test_labels)

It might be a little bit different from the course one because I had to change some parts as they were giving errors in the latest version of TensorFlow.