Training a DAENN model
Train a model with DeepCV is very easy. All parameters needed to train a model are defined in one input file, that is all you need.
Note that the current version of DeepCV accepts only datasets that are in NumPy's compressed file formats (.npz).
Call DeepCV's module
All DeepCV's modules can be called via main.py API script (Do not forget to activate to the environment where you installed dependencies packages for DeepCV).
$ cd deepcv/src/
$ python main.py
------------------------------------------------
DeepCV : Deep Learning for Collective Variables
-------------------------------------------------
version 1.0 : February 2022
University of Zurich, Switzerland
https://gitlab.uzh.ch/lubergroup/deepcv
Module Description
------------- ----------------------------------------------
calc_rep Calculate molecular representation
gen_input Neural network input generator
single_train Single-data fully-connected neural network
multi_train Multi-data fully-connected neural network
daenn Deep autoencoder neural network
gan_train Training generative adversarial network (GAN)
gan_predict Generating samples using trained GAN
deepcv2plumed Create PLUMED input file
analyze_FES FES validation
analyze_model DAENN model analysis and parameters extraction
explor_abi Calculate exploration ability
For more detail, please review 'README' in the repository.
Train model
Execute the main.py with argument daenn, like below. Then it will start to train the model.
The training time depends on the size of the dataset and networks, the number of epochs, etc.
$ python main.py daenn -i input/input_ae_DA.json
============================== Program started ==============================
Project: Demo
=== Shape of dataset before splitting ===
>>> 1. Dataset: (99000, 15)
>>> 2. Dataset: (99000, 14)
>>> 3. Dataset: (99000, 13)
=== Shape of dataset after splitting ===
>>> 1. Train: (79200, 15) & Test: (19800, 15)
>>> 2. Train: (79200, 14) & Test: (19800, 14)
>>> 3. Train: (79200, 13) & Test: (19800, 13)
...
...
...
1/1 [==============================] - ETA: 0s - loss: 0.1283 - out_1_loss: 0.1268 - out_2_loss: 0.0154 - out_1_mse: 4.0202e-04 - out_2_mse: 61/1█ ETA: 00:00s - loss: 0.1283 - out_1_loss: 0.1268 - out_2_loss: 0.0154 - out_1_mse: 0.0004 - out_2_mse: 67.5435 - val_loss: 0.1176 - val_ou1/1 [==============================] - 0s 21ms/step - loss: 0.1283 - out_1_loss: 0.1268 - out_2_loss: 0.0154 - out_1_mse: 4.0202e-04 - out_2_mse: 67.5435 - val_loss: 0.1176 - val_out_1_loss: 0.1160 - val_out_2_loss: 0.0156 - val_out_1_mse: 0.0013 - val_out_2_mse: 66.8437
Training: 100%|█████████████████████████████████████████████████████████████████████████████████████████ 1000/1000 ETA: 00:00s, 42.66epochs/sDeepCV:INFO >>> Congrats! Training model is completed.
DeepCV:INFO >>> Congrats! Training model is completed.
DeepCV:INFO >>> Model has been saved to /home/rketka/github/deepcv/output/model.h5
DeepCV:INFO >>> Weights of model have been saved to /home/rketka/github/deepcv/output/model_weights.h5
DeepCV:INFO >>> Weights of model have been saved to /home/rketka/github/deepcv/output/model_weights.npz
DeepCV:INFO >>> Biases of model have been saved to /home/rketka/github/deepcv/output/model_biases.npz
DeepCV:INFO >>> Directed graphs of all model have been saved to /mnt/c/Users/Nutt/Desktop/daenn-test
DeepCV:INFO >>> Loss history plot has been saved to /home/rketka/github/deepcv/output/loss.png
DeepCV:INFO >>> Metric accuracy history plot has been saved to /home/rketka/github/deepcv/output/metrics.png
============================== DONE ==============================
Once you see the line === DONE=== it means that the training is completed.
You can then use the output saved in the folder you specified in the input file for further works, i.e., generating CVs.
Outputs
DeepCV save autoencoder, encoder, and decoder models separately in output directory.
Example of a list of output generated and saved by DeepCV (sorted by name).
assets
latent_space_100_epochs.png
latent_space_120_epochs.png
latent_space_140_epochs.png
latent_space_160_epochs.png
latent_space_180_epochs.png
latent_space_200_epochs.png
latent_space_20_epochs.png
latent_space_40_epochs.png
latent_space_60_epochs.png
latent_space_80_epochs.png
loss.png
metrics.png
model_biases.npz
model.h5
model.png
model_summary.txt
model_weights.h5
model_weights.npz
saved_model.pb
variables