ProphitBet is a Machine Learning Soccer Bet prediction application. It analyzes the form of teams, computes match statistics and predicts the outcomes of a match using Advanced Machine Learning (ML) methods. The supported algorithms in this application are Neural Networks, Random Forests & Ensembl Models.
MIT License
312
stars
113
forks
source link
NN-Training with tensorflow-metal slower than without #38
i got a new Macbook Air with the M2 Chip. To use it, i have to install tensowflow-metal. The GPU is recognized, but the Training takes longer than with the CPU only.
With GPU:
`
2023-10-02 10:47:29.111865: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.
170/170 [==============================] - ETA: 0s - loss: 8.5763 - accuracy: 0.3513
2023-10-02 10:47:53.714029: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.
22/22 [==============================] - 0s 998us/step - loss: 1.8961 - accuracy: 0.3321 - val_loss: 1.3421 - val_accuracy: 0.4000 - lr: 1.0000e-04
Epoch 6/80
`
I tried the example from here https://developer.apple.com/metal/tensorflow-plugin/ and that was significant faster.
Hi,
i got a new Macbook Air with the M2 Chip. To use it, i have to install tensowflow-metal. The GPU is recognized, but the Training takes longer than with the CPU only. With GPU:
` 2023-10-02 10:47:29.111865: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.
170/170 [==============================] - ETA: 0s - loss: 8.5763 - accuracy: 0.3513 2023-10-02 10:47:53.714029: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.
170/170 [==============================] - 28s 119ms/step - loss: 8.5763 - accuracy: 0.3513 - val_loss: 6.3053 - val_accuracy: 0.4400 - lr: 7.0000e-04 Epoch 2/80 `
With CPU:
` Epoch 4/80 22/22 [==============================] - 0s 1ms/step - loss: 1.9838 - accuracy: 0.3166 - val_loss: 1.3517 - val_accuracy: 0.3200 - lr: 1.0000e-04 Epoch 5/80
22/22 [==============================] - 0s 998us/step - loss: 1.8961 - accuracy: 0.3321 - val_loss: 1.3421 - val_accuracy: 0.4000 - lr: 1.0000e-04 Epoch 6/80 ` I tried the example from here https://developer.apple.com/metal/tensorflow-plugin/ and that was significant faster.
Anybody an Idea what to change?