Model Auto-naming

This example assumes you’ve read advanced.py, and covers:

• How to configure automatic model naming

import deeptrain
deeptrain.util.misc.append_examples_dir_to_sys_path()

from utils import make_autoencoder, init_session, AE_CONFIGS as C

DeepTrain auto-names model based on model_name_configs, a dict.

• Keys denote either TrainGenerator attributes, its object’s attributes (via .), or model_configs keys.
  • 'best_key_metric' reflects the actual value, if TrainGenerator checkpointed since last change.
• Values denote attribute aliases; if blank or None, will use attrs as given.

name_cfg = {'datagen.batch_size': 'BS',
            'filters': 'filt',
            'optimizer': '',
            'lr': '',
            'best_key_metric': '__max'}
C['traingen'].update({'epochs': 1,
                      'model_base_name': "AE",
                      'model_name_configs': name_cfg})
C['model']['optimizer'] = 'Adam'
C['model']['lr'] = 1e-4

tg = init_session(C, make_autoencoder)

WARNING: multiple file extensions found in `path`; only .npy will be used
Discovered 48 files with matching format
48 set nums inferred; if more are expected, ensure file names contain a common substring w/ a number (e.g. 'train1.npy', 'train2.npy', etc)
DataGenerator initiated

WARNING: multiple file extensions found in `path`; only .npy will be used
Discovered 36 files with matching format
36 set nums inferred; if more are expected, ensure file names contain a common substring w/ a number (e.g. 'train1.npy', 'train2.npy', etc)
DataGenerator initiated

NOTE: will exclude `labels` from saving when `input_as_labels=True`; to keep 'labels', add '{labels}'to `saveskip_list` instead
Preloading superbatch ... WARNING: multiple file extensions found in `path`; only .npy will be used
Discovered 48 files with matching format
................................................ finished, w/ 6144 total samples
Train initial data prepared
Preloading superbatch ... WARNING: multiple file extensions found in `path`; only .npy will be used
Discovered 36 files with matching format
.................................... finished, w/ 4608 total samples
Val initial data prepared
Logging ON; directory (new): C:\deeptrain\examples\dir\logs\M9__AE-filt6_12_2_6_12-Adam-1e-4__max999.000

tg.train()

Fitting set 1...   Loss = 0.303295
Fitting set 2...   Loss = 0.301919
Fitting set 3...   Loss = 0.301043
Fitting set 4...   Loss = 0.300576
Fitting set 5...   Loss = 0.300247
Fitting set 6...   Loss = 0.300119
Fitting set 7...   Loss = 0.299533
Fitting set 8...   Loss = 0.299117
Fitting set 9...   Loss = 0.298494
Fitting set 10...  Loss = 0.297995
Fitting set 11...  Loss = 0.297355
Fitting set 12...  Loss = 0.296885
Fitting set 13...  Loss = 0.296406
Fitting set 14...  Loss = 0.295988
Fitting set 15...  Loss = 0.295545
Fitting set 16...  Loss = 0.295146
Fitting set 17...  Loss = 0.294726
Fitting set 18...  Loss = 0.294205
Fitting set 19...  Loss = 0.293743
Fitting set 20...  Loss = 0.293248
Fitting set 21...  Loss = 0.292782
Fitting set 22...  Loss = 0.292293
Fitting set 23...  Loss = 0.291838
Fitting set 24...  Loss = 0.291419
Fitting set 25...  Loss = 0.290918
Fitting set 26...  Loss = 0.290412
Fitting set 27...  Loss = 0.289942
Fitting set 28...  Loss = 0.289465
Fitting set 29...  Loss = 0.288982
Fitting set 30...  Loss = 0.288463
Fitting set 31...  Loss = 0.287996
Fitting set 32...  Loss = 0.287530
Fitting set 33...  Loss = 0.287014
Fitting set 34...  Loss = 0.286521
Fitting set 35...  Loss = 0.286011
Fitting set 36...  Loss = 0.285539
Fitting set 37...  Loss = 0.285035
Fitting set 38...  Loss = 0.284538
Fitting set 39...  Loss = 0.284004
Fitting set 40...  Loss = 0.283527
Fitting set 41...  Loss = 0.283023
Fitting set 42...  Loss = 0.282512
Fitting set 43...  Loss = 0.282022
Fitting set 44...  Loss = 0.281532
Fitting set 45...  Loss = 0.281049
Fitting set 46...  Loss = 0.280587
Fitting set 47...  Loss = 0.280085
Fitting set 48...  Loss = 0.279605
Data set_nums shuffled


_____________________
 EPOCH 1 -- COMPLETE 



Validating...
Validating set 1...  Loss = 0.238562
Validating set 2...  Loss = 0.238878
Validating set 3...  Loss = 0.237879
Validating set 4...  Loss = 0.238345
Validating set 5...  Loss = 0.238605
Validating set 6...  Loss = 0.237900
Validating set 7...  Loss = 0.238646
Validating set 8...  Loss = 0.238011
Validating set 9...  Loss = 0.238837
Validating set 10... Loss = 0.238321
Validating set 11... Loss = 0.238181
Validating set 12... Loss = 0.238668
Validating set 13... Loss = 0.237183
Validating set 14... Loss = 0.239100
Validating set 15... Loss = 0.238098
Validating set 16... Loss = 0.237906
Validating set 17... Loss = 0.238677
Validating set 18... Loss = 0.239686
Validating set 19... Loss = 0.238267
Validating set 20... Loss = 0.238836
Validating set 21... Loss = 0.238081
Validating set 22... Loss = 0.238481
Validating set 23... Loss = 0.238832
Validating set 24... Loss = 0.238768
Validating set 25... Loss = 0.238193
Validating set 26... Loss = 0.238238
Validating set 27... Loss = 0.237997
Validating set 28... Loss = 0.238209
Validating set 29... Loss = 0.238841
Validating set 30... Loss = 0.238128
Validating set 31... Loss = 0.238263
Validating set 32... Loss = 0.238241
Validating set 33... Loss = 0.238068
Validating set 34... Loss = 0.237933
Validating set 35... Loss = 0.238575
Validating set 36... Loss = 0.238656
TrainGenerator state saved
Model report generated and saved
Best model saved to C:\deeptrain\examples\dir\models\M9__AE-filt6_12_2_6_12-Adam-1e-4__max.238
TrainGenerator state saved
Model report generated and saved

Training has concluded.

print(tg.model_name)

M9__AE-filt6_12_2_6_12-Adam-1e-4__max.238

Note that logdir and best model saves are also named with model_name; it, together with model_num, enables scalable reference to hundreds of trained models: sort through models by reading off key hyperparameters.

print(tg.logdir)
print(tg.get_last_log('state', best=True))

C:\deeptrain\examples\dir\logs\M9__AE-filt6_12_2_6_12-Adam-1e-4__max999.000
C:\deeptrain\examples\dir\models\M9__AE-filt6_12_2_6_12-Adam-1e-4__max.238__state.h5