use a pre-trained model about calamari HOT 3 CLOSED

You should see a huge difference between training with and without using --weights, especially in the time it consumes to get the first meaningful results.
However, since the last layer of the neural net is not 'copied' fully (possible changes of the alphabet) and your lines are not fully similar to the training data set of the pretrainined model, the pretraining yields also 'garbage' for the first few iterations.
However, as said, training using a pretrained model should converge way faster.

from calamari.

@ChWick
thank you very much

from calamari.

@ChWick
Dear,
Please, if I want to execute Calamari with only one layer LSTM , is the following command correct :

calamari-train --files DatasetPath –network lstm=200 --checkpoint_frequency 1000 --output_dir modelsLSTM200

Thank you

from calamari.