An implementation of Feedforward Neural Networks for quick applications.
Project description
QuickNN
An implementation of Feedforward Neural Networks for quick applications.
- The training phase can be stopped, some parameters on fit method can be changed and then the training can be resumed with the same weights of the last interruption.
- If feed with pandas objects it can handle categorical variables with one-hot-encoding(OHE) method batch-wise as well as continuous variables.
- Easy visualization in Tensorboard of the metrics provided.
- Inner management of the validation set in the training phase.
Example
from quicknn import QuickNN
from sklearn.datasets import load_boston
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
X, y = load_boston(return_X_y=True)
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.25)
qnn = QuickNN(list_neurons=[100, 200, 1])
qnn.fit(X_train, y_train, n_epochs=10) ## In IPython session you can stop-change-resume the training
qnn.fit(X_train, y_train, n_epochs=20) ## just increasing the n_epochs.
qnn.fit(X_train, y_train, learning_rate=0.01) ## you can change e.g., the learning_rate param while training
y_pred = qnn.predict(X_val)
score = mean_squared_error(y_val, y_pred)
Installing
The dependencies are showed in requirements.txt, which can be installed with the command:
$ pip install -r requirements.txt
Then the library can easily downloaded through pip:
$ pip install quicknn
License
This project is licensed under the MIT License - see the LICENSE.md file for details.
Reference
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
quicknn-1.0.8.tar.gz
(6.4 kB
view hashes)