Initialization issues and model.predict doesn't work
edavidk7 opened this issue · comments
I am new to both TensorFlow and machine learning, I built working code on my windows system and wanted to try it on M1. I used the virtual environment created by the installer, but it seems to have issues especially with fitting (model.predict) my own data.
Here is the code:
`import tensorflow as tf
from tensorflow import keras
#import matplotlib.pyplot as plt
#%matplotlib inline
import numpy as np
(X_train, y_train), (X_test, y_test) = keras.datasets.mnist.load_data()
X_train, X_test = X_train/255, X_test/255
X_train_flattened = X_train.reshape(len(X_train),2828)
X_test_flattened = X_test.reshape(len(X_test),2828)
model = keras.Sequential([
keras.layers.Dense(600,input_shape=(784,), activation = "relu"),
keras.layers.Dense(400, activation = "sigmoid"),
keras.layers.Dense(200, activation = "sigmoid"),
keras.layers.Dense(75, activation = "sigmoid"),
keras.layers.Dense(10, activation = "sigmoid")
])
model.compile(
optimizer = "adam", loss = "sparse_categorical_crossentropy", metrics = ["accuracy"]
)
model.fit(X_train_flattened, y_train, epochs=28)
model.evaluate(X_test_flattened,y_test)
print("Accuracy assesment:")
y_predictions = model.predict(X_test_flattened)
print(y_test[:15])
y_predicted_labels = [np.argmax(j) for j in y_predictions]
print(y_predicted_labels[:15])
cm = tf.math.confusion_matrix(labels=y_test, predictions=y_predicted_labels)
#plt.matshow(cm)`
and this is the console output:
(tensorflow_macos_venv) davidkorcak@David-Air Documents % /usr/bin/env /Users/davidkorcak/Documents/tensorflow_macos_venv/bin/python /Users/davidkorcak/.vscode/extensions/ms-python.python-2021.5.842923320/pythonFiles/lib/python/debugpy/launcher 60087 -- /Users/davidkorcak/Documents/test.py (tensorflow_macos_venv) davidkorcak@David-Air Documents % cd /Users/davidkorcak/Documents ; /usr/bin/env /Users/davidkorcak/Documents/tensorflow_macos_venv/bin/python /Users/davidkorcak/.vscode/extensions/ms-python.python-2021.5.842923320/pythonFiles/lib/python/debugpy/launcher 60109 -- /Users/davidkorcak/Documents/test.py 2021-05-19 11:53:53.554146: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:116] None of the MLIR optimization passes are enabled (registered 2) 2021-05-19 11:53:53.554439: W tensorflow/core/platform/profile_utils/cpu_utils.cc:126] Failed to get CPU frequency: 0 Hz Epoch 1/28 1875/1875 [==============================] - 4s 2ms/step - loss: 0.6058 - accuracy: 0.8328 Epoch 2/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.1043 - accuracy: 0.9695 Epoch 3/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0654 - accuracy: 0.9807 Epoch 4/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0516 - accuracy: 0.9850 Epoch 5/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0371 - accuracy: 0.9887 Epoch 6/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0304 - accuracy: 0.9913 Epoch 7/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0233 - accuracy: 0.9928 Epoch 8/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0250 - accuracy: 0.9917 Epoch 9/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0176 - accuracy: 0.9946 Epoch 10/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0148 - accuracy: 0.9954 Epoch 11/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0118 - accuracy: 0.9965 Epoch 12/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0143 - accuracy: 0.9954 Epoch 13/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0105 - accuracy: 0.9965 Epoch 14/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0091 - accuracy: 0.9975 Epoch 15/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0120 - accuracy: 0.9962 Epoch 16/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0084 - accuracy: 0.9975 Epoch 17/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0099 - accuracy: 0.9970 Epoch 18/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0068 - accuracy: 0.9978 Epoch 19/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0072 - accuracy: 0.9978 Epoch 20/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0073 - accuracy: 0.9977 Epoch 21/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0077 - accuracy: 0.9978 Epoch 22/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0045 - accuracy: 0.9988 Epoch 23/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0048 - accuracy: 0.9983 Epoch 24/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0046 - accuracy: 0.9987 Epoch 25/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0058 - accuracy: 0.9985 Epoch 26/28 1875/1875 [==============================] - 3s 1ms/step - loss: 0.0041 - accuracy: 0.9987 Epoch 27/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0059 - accuracy: 0.9980 Epoch 28/28 1875/1875 [==============================] - 3s 2ms/step - loss: 0.0041 - accuracy: 0.9988 313/313 [==============================] - 0s 607us/step - loss: 0.0869 - accuracy: 0.9829 Accuracy assesment: 2021-05-19 11:55:15.583944: I tensorflow/compiler/tf2mlcompute/kernels/mlc_subgraph_op.cc:326] Compute: Failed in processing TensorFlow graph sequential/MLCSubgraphOp_2_0 with frame_id = 0 and iter_id = 0 with error: Internal: ExecuteMLCInferenceGraph: Failed to execute MLC inference graph. (error will be reported 5 times unless TF_MLC_LOGGING=1). 2021-05-19 11:55:15.587260: F tensorflow/core/framework/op_kernel.cc:983] Check failed: outputs_[index].tensor == nullptr (0x124715490 vs. nullptr)
Any ideas on what to do? Besides the tensorflow-macos dependencies, I also have Jupyter installed
I have a same issue. MacTF a1,a2,a3 are same. Please suggest the environment for MacTF, For example numpy etc.
It is working fine for me.
Try this ->
- If you have numpy 1.20 then try to install numpy 1.19.5
- Check if tf.executing_eagerly() is set to True as it should be
Thank you, @devnev39.
I tried your code on my M1 mac. It is works! I understand this issue that is graph executing or eagerly executing. I change to design eagerly executable models. Thank you again.
@devnev39 I have the numpy version supplied with TensorFlow-macOS, which is 1.18.5. I will try your recommended code and the changing the eagerly execution parameter. Thank you.
@devnev39 It seems the activation function in the last layer is the issue here. I changed the code line by line to match yours and it only works once the activation function parameter is removed in the last layer.
Edit: linear activation for output layer works fine, issue seems to be only with sigmoid
I guess there is some error in your side of code. I just did all the possibilities with sigmoid and relu and also setting from_logits=True (generally used in loss function when there is no activation at output layer) and they all executed the same code in the pictures without any error.