search

brakid / BundesligaPrediction

Predicting results of games of the German Soccer League (Bundesliga) using a feed-forward network.
6 stars 3 forks source link

Accuracy didnt increase #1

Open Timmmeyyy opened 5 years ago

Timmmeyyy commented 5 years ago

Hi, i have a probleme that my accuracy didnt increase, it is always 0,66 after 1400 epochs. Do you have any idea whtat can i check first. I think my input data and labels are right. Greetz Tim

brakid commented 5 years ago

Hi,

maybe you need to tweak a bit the hyperparameters of the model such as the layer-count and size: https://github.com/brakid/BundesligaPrediction/blob/2b647f2b8a585b8dcbcfb91ff21434cf48b68ee0/prediction.py#L178 and following lines. You might also want to check different activation functions such as tanh or softrelu in the layers: https://beta.mxnet.io/api/ndarray/_autogen/mxnet.ndarray.Activation.html

In general, if the model fails to match your data, this is a sign that the model is not powerful enough to capture the subtleties in your dataset, adding additional layers could be helpful as well.

Another option could be that the optimizer is not calibrated correctly for your data. The learning rate might be too high or low: https://github.com/brakid/BundesligaPrediction/blob/2b647f2b8a585b8dcbcfb91ff21434cf48b68ee0/prediction.py#L191

There are various resources on hyperparameter tuning for Neural networks, such as: https://towardsdatascience.com/hyper-parameter-tuning-techniques-in-deep-learning-4dad592c63c8.

Hopefully this helps you increase the accuracy and performance of your neural network.