keras-one-cycle
keras-one-cycle copied to clipboard
titu1994
Implementation of One-Cycle Learning rate policy (adapted from Fast.ai lib)
The initialization in clr.py is: `def __init__(self, num_samples, batch_size, max_lr, end_percentage=0.1, scale_percentage=None, maximum_momentum=0.95, minimum_momentum=0.85, verbose=True)` The initialization in the documentation is: `lr_manager = OneCycleLR(num_samples, num_epoch, batch_size, max_lr end_percentage=0.1, scale_percentage=None, maximum_momentum=0.95,...
I received this output when using data generators: ``` --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in 1 model.compile(optimizer='adam', loss=losses.CategoricalCrossentropy(),metrics=['accuracy']) 2 lr_callback = LRFinder(train_generator.n, train_generator.batch_size) ----> 3 history = model.fit_generator(train_generator,...
Currently, it seems this tool suppose validation to be NumPy arrays or Keras generator, while the tensorflow.data.Dataset is also one of the most frequently used input data structure which is...
When verbose is used, LR finder prints progress in multiple lines which is ugly. There progress bar should be printed in the same line 256/298 [========================>.....] - ETA: 38s -...
I know there is a "plot interpretation" section already. But I still cannot get the idea about what your callback did and what is the meaning of the plot. Why...
Have defined my lr callback as follows lr_manager = OneCycleLR(max_lr,end_percentage, scale_percentage,maximum_momentum, minimum_momentum,verbose=True) Then have further created the callback callbackscustomx = [lr_manager, ModelCheckpoint(filepathx, monitor='val_acc', verbose=1, save_best_only=True)] But then model.fit is failing...
My model isn't using any regularizers. Do I need weight decay? Or Should add l2? (Model in questions is a mix of 3d convolutions and 2d convolutions)
when I use clr_triangular = OneCycleLR(NUM_SAMPLES, NUM_EPOCHS, BATCH_SIZE, MAX_LR, end_percentage=0.2, scale_percentage=0.2) in your plot_clr.py I got the following errors: __init__() got multiple values for argument 'and percentage' But if I...
Hi, I'm visually impaired, so I can't look at the plots. I'm wondering if it's possible to implement a feature for the script to find the right values, so users...
