TypeError: 'NoneType' object is not iterable

[AlexYiLuo]

Hello,

I have been having the following errors, and not sure what went wrong. Can you help with the troubleshooting? Thank you very much!

neural-admixture train --k 5 --supervised --populations_path label.txt --name RUN_NAME --data_path phase3_kin_prune.bed --save_dir SAVE_PATH
INFO:neural_admixture.entry:Neural ADMIXTURE - Version 1.3.0
INFO:neural_admixture.entry:[CHANGELOG] Default P initialization was changed to 'pckmeans' in version 1.3.0.
INFO:neural_admixture.entry:[CHANGELOG] Warmup training for initialization of Q was added in version 1.3.0 to improve training stability (only for pckmeans).
INFO:neural_admixture.entry:[CHANGELOG] Convergence check changed so it is performed after 15 epochs in version 1.3.0 to improve training stability.
INFO:neural_admixture.entry:[CHANGELOG] Default learning rate was changed to 1e-5 instead of 1e-4 in version 1.3.0 to improve training stability.
INFO:neural_admixture.src.utils:Reading data...
INFO:neural_admixture.src.snp_reader:Input format is BED.
Mapping files: 100%|███████████████████████████████████████████████████████████████████████████| 3/3 [00:01<00:00, 1.61it/s]
INFO:neural_admixture.src.utils:Data contains 2492 samples and 842416 SNPs.
INFO:neural_admixture.src.utils:Data loaded.
INFO:neural_admixture.src.train:Job args: Namespace(learning_rate=1e-05, max_epochs=50, initialization='pckmeans', optimizer='adam', save_every=10, l2_penalty=0.0005, activation='gelu', seed=42, k=5, min_k=None, max_k=None, hidden_size=64, init_file=None, freeze_decoder=False, supervised=True, validation_data_path='', populations_path='label.txt', validation_populations_path='', wandb_log=False, wandb_user=None, wandb_project=None, pca_path=None, pca_components=2, tol=1e-06, save_dir='SAVE_PATH', data_path='phase3_kin_prune.bed', name='RUN_NAME', batch_size=400, supervised_loss_weight=0.05, warmup_epochs=10)
INFO:neural_admixture.src.train:Will use GPU.
INFO:neural_admixture.src.train:Initializing...
WARNING:neural_admixture.src.train:Initialization filename not provided. Going to store it to SAVE_PATH/RUN_NAME.pkl
INFO:neural_admixture.model.initializations:Running supervised initialization...
INFO:neural_admixture.model.initializations:Weights initialized in 19.29151487350464 seconds.
INFO:neural_admixture.src.train:Variants: 842416
INFO:neural_admixture.src.train:Optimizer successfully loaded.
INFO:neural_admixture.src.train:Going to train 1 head: K=5.
INFO:neural_admixture.src.train:Fitting...
INFO:neural_admixture.model.neural_admixture:Will stop optimization when difference in objective function between two subsequent iterations is < 1e-06 or after 50 epochs.
INFO:neural_admixture.model.neural_admixture:Going to train on supervised mode.
INFO:neural_admixture.model.neural_admixture:Bringing training data into memory...
INFO:neural_admixture.model.neural_admixture:Running warmup epochs...
Warmup epoch 1: 0%| | 0/7 [00:00<?, ?it/s]
Traceback (most recent call last):
File "/home/yluo/.conda/envs/ndam/bin/neural-admixture", line 8, in
sys.exit(main())
File "/home/yluo/.conda/envs/ndam/lib/python3.9/site-packages/neural_admixture/entry.py", line 18, in main
sys.exit(train.main(arg_list[2:]))
File "/home/yluo/.conda/envs/ndam/lib/python3.9/site-packages/neural_admixture/src/train.py", line 162, in main
model, device, _ = fit_model(trX, args, valX, trY, valY)
File "/home/yluo/.conda/envs/ndam/lib/python3.9/site-packages/neural_admixture/src/train.py", line 116, in fit_model
actual_num_epochs = model.launch_training(trX, optimizer, loss_f, num_max_epochs, device, valX=valX,
File "/home/yluo/.conda/envs/ndam/lib/python3.9/site-packages/neural_admixture/model/neural_admixture.py", line 135, in launch_training
_, _ = self._run_warmup_epoch(trX, Q_inits, opt_warmup, loss_f_warmup, batch_size, device, shuffle, epoch_num=wep+1)
File "/home/yluo/.conda/envs/ndam/lib/python3.9/site-packages/neural_admixture/model/neural_admixture.py", line 284, in _run_warmup_epoch
Q_inits=[Y.to(device) for Y in Ys])
TypeError: 'NoneType' object is not iterable

[AlbertDominguez]

Thanks for reporting this! This was caused due to the network trying to run the warmup training, however it should only be available when using the PCK-Means, and not on the supervised mode as in your case.
Fixed this in the latest release (v1.3.1) by skipping warmup unless the initialization is PCK-Means. Please run pip install -U neural-admixture to update the library!