Note: This environment variable is required for fully deterministic CuBLAS ops on CUDA >= 10.2 when reproducible=True is set below. Without it, PyTorch raises a RuntimeError instead of training deterministically. It must be set before torch is imported. See the README FAQ for details.

In [ ]:

%env CUBLAS_WORKSPACE_CONFIG=:16:8

%env CUBLAS_WORKSPACE_CONFIG=:16:8

How to Save the Pipeline Memory Efficiently¶

Our pipeline offers extensive insights into the dynamics of training and other intermediate results.
When re-using the pipeline (with .save and .load functionality), this can create very large binary files, especially when working with large datasets.

Therefore, the default way of saving the pipeline is by discarding most of the training dynamics and intermediate results.
In this tutorial, we show:

• Which data is kept and which is discarded.
  
  
• The trade-offs in visualizing and evaluating a saved pipeline.
  
  
• How to save the complete pipeline including all training dynamics and intermediate results.
  
  

Theory: What Data is Kept?¶

Result Object¶

The main driver of memory is the result attribute and the _datasets (depending on the data size) in the pipeline and the trainer.

In the result attribute, we save the following:

• model
• adata_latent
  
  All other attributes are set to empty default values.

Pipeline¶

Here we keep all attributes excpept:

• .visualizer
• _datasets We remove the .visualizer, because this stores plots which can be misleading when rerunning the pipeline and seeing old plots that do not fit the new data.
  We remove the _datasets attribute, because it is often uses the most memory. All other attributes are kept. This is especially relevant for the config and the _preprocessor.
  Whenever you perform any pipeline step, it will use the initial config you passed when first creating the pipeline.
  The preprocessor is necessary because when running predict with new data, this should be preprocessed the same way as initially (e.g., which genes to keep, how the scalers were fitted, etc.).

Trainer¶

We remove most attribute of the _trainer via the method purge.

IMPORTANT

When calling .save, the result attribute is modified in-place (to avoid copies).
This means after calling .save, your current pipeline object does not contain information from the other result objects.
This applies to the pipeline object in current memory, not only to the pipeline object after loading.
If you want to avoid this, please run .save(save_all=True).

Practical Examples¶

First we run a varix pipeline¶

In [1]:

from autoencodix.utils.example_data import EXAMPLE_MULTI_SC
from autoencodix.configs.varix_config import VarixConfig
from autoencodix.configs.default_config import DataCase, DataConfig, DataInfo
import autoencodix as acx

my_config = VarixConfig(
    learning_rate=0.001,
    epochs=30,
    checkpoint_interval=5,
    default_vae_loss="kl",  # kl or mmd possible
    data_case=DataCase.MULTI_SINGLE_CELL,
)
print("\n")
print("Starting Pipeline")
print("-" * 50)
print("-" * 50)
varix = acx.Varix(
    data=EXAMPLE_MULTI_SC,
    config=my_config,
)
result = varix.run()

from autoencodix.utils.example_data import EXAMPLE_MULTI_SC from autoencodix.configs.varix_config import VarixConfig from autoencodix.configs.default_config import DataCase, DataConfig, DataInfo import autoencodix as acx my_config = VarixConfig( learning_rate=0.001, epochs=30, checkpoint_interval=5, default_vae_loss="kl", # kl or mmd possible data_case=DataCase.MULTI_SINGLE_CELL, ) print("\n") print("Starting Pipeline") print("-" * 50) print("-" * 50) varix = acx.Varix( data=EXAMPLE_MULTI_SC, config=my_config, ) result = varix.run()

Starting Pipeline
--------------------------------------------------
--------------------------------------------------
in handle_direct_user_data with data: <class 'autoencodix.data.datapackage.DataPackage'>
mudata: View of MuData object with n_obs × n_vars = 1000 × 700
  2 modalities
    rna:	1000 x 500
      obs:	'cell_type', 'batch', 'donor', 'cell_cycle'
    protein:	1000 x 200
      obs:	'cell_type', 'batch', 'donor', 'cell_cycle'
Processing 1 MuData objects: ['multi_sc']
Processing train modality: multi_sc
Processing valid split
Processing valid modality: multi_sc
Processing test split
Processing test modality: multi_sc

OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.

Epoch 1 - Train Loss: 783.8270
Sub-losses: recon_loss: 783.8269, var_loss: 0.0000, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 1 - Valid Loss: 674.2708
Sub-losses: recon_loss: 674.2707, var_loss: 0.0001, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 2 - Train Loss: 699.6010
Sub-losses: recon_loss: 699.6008, var_loss: 0.0002, anneal_factor: 0.0001, effective_beta_factor: 0.0000
Epoch 2 - Valid Loss: 652.1881
Sub-losses: recon_loss: 652.1878, var_loss: 0.0003, anneal_factor: 0.0001, effective_beta_factor: 0.0000
Epoch 3 - Train Loss: 656.4180
Sub-losses: recon_loss: 656.4168, var_loss: 0.0012, anneal_factor: 0.0002, effective_beta_factor: 0.0000
Epoch 3 - Valid Loss: 625.0739
Sub-losses: recon_loss: 625.0734, var_loss: 0.0006, anneal_factor: 0.0002, effective_beta_factor: 0.0000
Epoch 4 - Train Loss: 637.0107
Sub-losses: recon_loss: 637.0058, var_loss: 0.0050, anneal_factor: 0.0003, effective_beta_factor: 0.0000
Epoch 4 - Valid Loss: 615.8611
Sub-losses: recon_loss: 615.8604, var_loss: 0.0007, anneal_factor: 0.0003, effective_beta_factor: 0.0000
Epoch 5 - Train Loss: 627.9194
Sub-losses: recon_loss: 627.9152, var_loss: 0.0042, anneal_factor: 0.0007, effective_beta_factor: 0.0001
Epoch 5 - Valid Loss: 605.3825
Sub-losses: recon_loss: 605.3809, var_loss: 0.0016, anneal_factor: 0.0007, effective_beta_factor: 0.0001
Epoch 6 - Train Loss: 612.5574
Sub-losses: recon_loss: 612.5484, var_loss: 0.0090, anneal_factor: 0.0013, effective_beta_factor: 0.0001
Epoch 6 - Valid Loss: 589.9656
Sub-losses: recon_loss: 589.9628, var_loss: 0.0028, anneal_factor: 0.0013, effective_beta_factor: 0.0001
Epoch 7 - Train Loss: 593.6844
Sub-losses: recon_loss: 593.6668, var_loss: 0.0177, anneal_factor: 0.0025, effective_beta_factor: 0.0002
Epoch 7 - Valid Loss: 560.1090
Sub-losses: recon_loss: 560.1044, var_loss: 0.0045, anneal_factor: 0.0025, effective_beta_factor: 0.0002
Epoch 8 - Train Loss: 561.6479
Sub-losses: recon_loss: 561.6288, var_loss: 0.0191, anneal_factor: 0.0048, effective_beta_factor: 0.0005
Epoch 8 - Valid Loss: 507.7625
Sub-losses: recon_loss: 507.7530, var_loss: 0.0096, anneal_factor: 0.0048, effective_beta_factor: 0.0005
Epoch 9 - Train Loss: 534.9153
Sub-losses: recon_loss: 534.8818, var_loss: 0.0335, anneal_factor: 0.0093, effective_beta_factor: 0.0009
Epoch 9 - Valid Loss: 445.4993
Sub-losses: recon_loss: 445.4802, var_loss: 0.0191, anneal_factor: 0.0093, effective_beta_factor: 0.0009
Epoch 10 - Train Loss: 497.7102
Sub-losses: recon_loss: 497.6525, var_loss: 0.0577, anneal_factor: 0.0180, effective_beta_factor: 0.0018
Epoch 10 - Valid Loss: 438.6540
Sub-losses: recon_loss: 438.6202, var_loss: 0.0338, anneal_factor: 0.0180, effective_beta_factor: 0.0018
Epoch 11 - Train Loss: 483.5498
Sub-losses: recon_loss: 483.4291, var_loss: 0.1206, anneal_factor: 0.0344, effective_beta_factor: 0.0034
Epoch 11 - Valid Loss: 416.3960
Sub-losses: recon_loss: 416.3270, var_loss: 0.0690, anneal_factor: 0.0344, effective_beta_factor: 0.0034
Epoch 12 - Train Loss: 470.8179
Sub-losses: recon_loss: 470.5980, var_loss: 0.2199, anneal_factor: 0.0650, effective_beta_factor: 0.0065
Epoch 12 - Valid Loss: 414.0788
Sub-losses: recon_loss: 413.9213, var_loss: 0.1575, anneal_factor: 0.0650, effective_beta_factor: 0.0065
Epoch 13 - Train Loss: 469.9154
Sub-losses: recon_loss: 469.5027, var_loss: 0.4127, anneal_factor: 0.1192, effective_beta_factor: 0.0119
Epoch 13 - Valid Loss: 403.3598
Sub-losses: recon_loss: 403.0631, var_loss: 0.2967, anneal_factor: 0.1192, effective_beta_factor: 0.0119
Epoch 14 - Train Loss: 462.5418
Sub-losses: recon_loss: 461.8312, var_loss: 0.7105, anneal_factor: 0.2086, effective_beta_factor: 0.0209
Epoch 14 - Valid Loss: 394.7311
Sub-losses: recon_loss: 394.1951, var_loss: 0.5361, anneal_factor: 0.2086, effective_beta_factor: 0.0209
Epoch 15 - Train Loss: 473.5944
Sub-losses: recon_loss: 472.4635, var_loss: 1.1309, anneal_factor: 0.3392, effective_beta_factor: 0.0339
Epoch 15 - Valid Loss: 399.1707
Sub-losses: recon_loss: 398.3019, var_loss: 0.8688, anneal_factor: 0.3392, effective_beta_factor: 0.0339
Epoch 16 - Train Loss: 460.3056
Sub-losses: recon_loss: 458.5930, var_loss: 1.7126, anneal_factor: 0.5000, effective_beta_factor: 0.0500
Epoch 16 - Valid Loss: 397.4022
Sub-losses: recon_loss: 395.9919, var_loss: 1.4102, anneal_factor: 0.5000, effective_beta_factor: 0.0500
Epoch 17 - Train Loss: 440.7799
Sub-losses: recon_loss: 438.4338, var_loss: 2.3461, anneal_factor: 0.6608, effective_beta_factor: 0.0661
Epoch 17 - Valid Loss: 394.7011
Sub-losses: recon_loss: 392.7804, var_loss: 1.9207, anneal_factor: 0.6608, effective_beta_factor: 0.0661
Epoch 18 - Train Loss: 454.1740
Sub-losses: recon_loss: 451.1796, var_loss: 2.9944, anneal_factor: 0.7914, effective_beta_factor: 0.0791
Epoch 18 - Valid Loss: 394.9608
Sub-losses: recon_loss: 392.6119, var_loss: 2.3489, anneal_factor: 0.7914, effective_beta_factor: 0.0791
Epoch 19 - Train Loss: 441.0030
Sub-losses: recon_loss: 437.6811, var_loss: 3.3219, anneal_factor: 0.8808, effective_beta_factor: 0.0881
Epoch 19 - Valid Loss: 393.5808
Sub-losses: recon_loss: 391.0548, var_loss: 2.5260, anneal_factor: 0.8808, effective_beta_factor: 0.0881
Epoch 20 - Train Loss: 443.8722
Sub-losses: recon_loss: 440.2967, var_loss: 3.5755, anneal_factor: 0.9350, effective_beta_factor: 0.0935
Epoch 20 - Valid Loss: 395.2028
Sub-losses: recon_loss: 392.4612, var_loss: 2.7416, anneal_factor: 0.9350, effective_beta_factor: 0.0935
Epoch 21 - Train Loss: 445.0039
Sub-losses: recon_loss: 441.2810, var_loss: 3.7229, anneal_factor: 0.9656, effective_beta_factor: 0.0966
Epoch 21 - Valid Loss: 397.5640
Sub-losses: recon_loss: 394.4998, var_loss: 3.0642, anneal_factor: 0.9656, effective_beta_factor: 0.0966
Epoch 22 - Train Loss: 436.5999
Sub-losses: recon_loss: 432.9243, var_loss: 3.6755, anneal_factor: 0.9820, effective_beta_factor: 0.0982
Epoch 22 - Valid Loss: 396.8338
Sub-losses: recon_loss: 393.9197, var_loss: 2.9141, anneal_factor: 0.9820, effective_beta_factor: 0.0982
Epoch 23 - Train Loss: 436.5542
Sub-losses: recon_loss: 432.7557, var_loss: 3.7985, anneal_factor: 0.9907, effective_beta_factor: 0.0991
Epoch 23 - Valid Loss: 393.0197
Sub-losses: recon_loss: 389.9772, var_loss: 3.0426, anneal_factor: 0.9907, effective_beta_factor: 0.0991
Epoch 24 - Train Loss: 440.4967
Sub-losses: recon_loss: 436.5647, var_loss: 3.9319, anneal_factor: 0.9952, effective_beta_factor: 0.0995
Epoch 24 - Valid Loss: 392.2895
Sub-losses: recon_loss: 389.0881, var_loss: 3.2015, anneal_factor: 0.9952, effective_beta_factor: 0.0995
Epoch 25 - Train Loss: 439.7828
Sub-losses: recon_loss: 435.8818, var_loss: 3.9011, anneal_factor: 0.9975, effective_beta_factor: 0.0998
Epoch 25 - Valid Loss: 392.1140
Sub-losses: recon_loss: 388.8866, var_loss: 3.2273, anneal_factor: 0.9975, effective_beta_factor: 0.0998
Epoch 26 - Train Loss: 436.1428
Sub-losses: recon_loss: 432.1413, var_loss: 4.0016, anneal_factor: 0.9987, effective_beta_factor: 0.0999
Epoch 26 - Valid Loss: 393.1209
Sub-losses: recon_loss: 389.9928, var_loss: 3.1280, anneal_factor: 0.9987, effective_beta_factor: 0.0999
Epoch 27 - Train Loss: 451.9407
Sub-losses: recon_loss: 448.0566, var_loss: 3.8841, anneal_factor: 0.9993, effective_beta_factor: 0.0999
Epoch 27 - Valid Loss: 392.9781
Sub-losses: recon_loss: 389.4436, var_loss: 3.5345, anneal_factor: 0.9993, effective_beta_factor: 0.0999
Epoch 28 - Train Loss: 433.2443
Sub-losses: recon_loss: 429.2605, var_loss: 3.9838, anneal_factor: 0.9997, effective_beta_factor: 0.1000
Epoch 28 - Valid Loss: 396.1637
Sub-losses: recon_loss: 392.5587, var_loss: 3.6050, anneal_factor: 0.9997, effective_beta_factor: 0.1000
Epoch 29 - Train Loss: 432.2076
Sub-losses: recon_loss: 428.2002, var_loss: 4.0074, anneal_factor: 0.9998, effective_beta_factor: 0.1000
Epoch 29 - Valid Loss: 392.4899
Sub-losses: recon_loss: 389.0604, var_loss: 3.4295, anneal_factor: 0.9998, effective_beta_factor: 0.1000
Epoch 30 - Train Loss: 428.7175
Sub-losses: recon_loss: 424.6904, var_loss: 4.0272, anneal_factor: 0.9999, effective_beta_factor: 0.1000
Epoch 30 - Valid Loss: 391.6893
Sub-losses: recon_loss: 388.3713, var_loss: 3.3179, anneal_factor: 0.9999, effective_beta_factor: 0.1000

Saving and Loading Explained¶

Here we first check the latent space (could be any other training dynamic) before saving.
After saving, we see that the result object got cleaned and only model, adata_latent, and embedding_evaluation are kept.

In [2]:

ls_before_save = result.latentspaces.get(epoch=-1, split="test")
print("Length of latentspace before saving")
print(len(ls_before_save))
varix.save("varix_backup.pkl")
ls_after_save = result.latentspaces.get()
print("Length of latentspace after saving")
print(len(ls_after_save))

ls_before_save = result.latentspaces.get(epoch=-1, split="test") print("Length of latentspace before saving") print(len(ls_before_save)) varix.save("varix_backup.pkl") ls_after_save = result.latentspaces.get() print("Length of latentspace after saving") print(len(ls_after_save))

Length of latentspace before saving
200
Preprocessor saved successfully.
saving memory efficient
Pipeline object saved successfully.
Length of latentspace after saving
0

Implications for Visualization¶

We load the model and see the implications of this in visualize.

In [3]:

varix_loaded = acx.Varix().load("varix_backup.pkl")

varix_loaded = acx.Varix().load("varix_backup.pkl")

Attempting to load a pipeline from ./varix_backup.pkl...
Pipeline object loaded successfully. Actual type: Varix
Preprocessor loaded successfully.

In [4]:

varix_loaded.visualize()

varix_loaded.visualize()

<Figure size 0x500 with 0 Axes>

Note that we code a UserWarning. This is expected, since it tells us that we cannot visualize loss plots anymore because the data for this is no longer in the result object.
However, we could run a predict step on the trained model to get some visualizations.

In [5]:

res_loaded = varix_loaded.predict(data=EXAMPLE_MULTI_SC)

res_loaded = varix_loaded.predict(data=EXAMPLE_MULTI_SC)

in handle_direct_user_data with data: <class 'autoencodix.data.datapackage.DataPackage'>
mudata: View of MuData object with n_obs × n_vars = 1000 × 700
  2 modalities
    rna:	1000 x 500
      obs:	'cell_type', 'batch', 'donor', 'cell_cycle', 'n_genes'
    protein:	1000 x 200
      obs:	'cell_type', 'batch', 'donor', 'cell_cycle', 'n_genes'
Processing 1 MuData objects: ['multi_sc']
Processing test split
Processing test modality: multi_sc
n_samples in format recon: 1000
train
n_samples from datatpackge: {'paired_count': 1000}

In [6]:

varix_loaded.visualize()
# we use test split here, because predict with new data is equal to test split
varix_loaded.show_result(split="test")

varix_loaded.visualize() # we use test split here, because predict with new data is equal to test split varix_loaded.show_result(split="test")

Creating plots ...

<Figure size 0x500 with 0 Axes>

Implications for Evaluate¶

The evaluate step is not supported for this memory-efficient saving as of now, and you'll get the following warning and error:

In [7]:

res_eval = varix_loaded.evaluate(params=["cell_type"])

res_eval = varix_loaded.evaluate(params=["cell_type"])

Perform ML task with feature df: Latent
Latent
Perform ML task for target parameter: cell_type

How to Keep All Data¶

When you really want to investigate all training dynamics and results, we recommend setting save_all=True. See below:

In [8]:

from autoencodix.utils.example_data import EXAMPLE_MULTI_BULK

my_config = VarixConfig(
    learning_rate=0.001,
    epochs=30,
    checkpoint_interval=5,
    default_vae_loss="kl",  # kl or mmd possible
    data_case=DataCase.MULTI_BULK,
)
n_varix = acx.Varix(data=EXAMPLE_MULTI_BULK, config=my_config)
result_new = n_varix.run()
n_varix.save(file_path="new_varix.pkl", save_all=True)

from autoencodix.utils.example_data import EXAMPLE_MULTI_BULK my_config = VarixConfig( learning_rate=0.001, epochs=30, checkpoint_interval=5, default_vae_loss="kl", # kl or mmd possible data_case=DataCase.MULTI_BULK, ) n_varix = acx.Varix(data=EXAMPLE_MULTI_BULK, config=my_config) result_new = n_varix.run() n_varix.save(file_path="new_varix.pkl", save_all=True)

in handle_direct_user_data with data: <class 'autoencodix.data.datapackage.DataPackage'>
anno key: transcriptomics
anno key: proteomics
Epoch 1 - Train Loss: 214.2306
Sub-losses: recon_loss: 214.2306, var_loss: 0.0000, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 1 - Valid Loss: 185.5984
Sub-losses: recon_loss: 185.5983, var_loss: 0.0000, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 2 - Train Loss: 203.4312
Sub-losses: recon_loss: 203.4311, var_loss: 0.0001, anneal_factor: 0.0001, effective_beta_factor: 0.0000
Epoch 2 - Valid Loss: 187.5715
Sub-losses: recon_loss: 187.5715, var_loss: 0.0000, anneal_factor: 0.0001, effective_beta_factor: 0.0000
Epoch 3 - Train Loss: 197.6002
Sub-losses: recon_loss: 197.6001, var_loss: 0.0001, anneal_factor: 0.0002, effective_beta_factor: 0.0000
Epoch 3 - Valid Loss: 185.5950
Sub-losses: recon_loss: 185.5950, var_loss: 0.0000, anneal_factor: 0.0002, effective_beta_factor: 0.0000
Epoch 4 - Train Loss: 195.3141
Sub-losses: recon_loss: 195.3139, var_loss: 0.0002, anneal_factor: 0.0003, effective_beta_factor: 0.0000
Epoch 4 - Valid Loss: 184.0191
Sub-losses: recon_loss: 184.0190, var_loss: 0.0001, anneal_factor: 0.0003, effective_beta_factor: 0.0000
Epoch 5 - Train Loss: 191.4216
Sub-losses: recon_loss: 191.4212, var_loss: 0.0004, anneal_factor: 0.0007, effective_beta_factor: 0.0001
Epoch 5 - Valid Loss: 184.4212
Sub-losses: recon_loss: 184.4210, var_loss: 0.0002, anneal_factor: 0.0007, effective_beta_factor: 0.0001
Epoch 6 - Train Loss: 190.2712
Sub-losses: recon_loss: 190.2704, var_loss: 0.0008, anneal_factor: 0.0013, effective_beta_factor: 0.0001
Epoch 6 - Valid Loss: 183.9692
Sub-losses: recon_loss: 183.9686, var_loss: 0.0006, anneal_factor: 0.0013, effective_beta_factor: 0.0001
Epoch 7 - Train Loss: 187.3219
Sub-losses: recon_loss: 187.3202, var_loss: 0.0018, anneal_factor: 0.0025, effective_beta_factor: 0.0002
Epoch 7 - Valid Loss: 182.1401
Sub-losses: recon_loss: 182.1388, var_loss: 0.0013, anneal_factor: 0.0025, effective_beta_factor: 0.0002
Epoch 8 - Train Loss: 185.5204
Sub-losses: recon_loss: 185.5167, var_loss: 0.0036, anneal_factor: 0.0048, effective_beta_factor: 0.0005
Epoch 8 - Valid Loss: 179.8548
Sub-losses: recon_loss: 179.8524, var_loss: 0.0024, anneal_factor: 0.0048, effective_beta_factor: 0.0005
Epoch 9 - Train Loss: 184.3709
Sub-losses: recon_loss: 184.3626, var_loss: 0.0083, anneal_factor: 0.0093, effective_beta_factor: 0.0009
Epoch 9 - Valid Loss: 181.4105
Sub-losses: recon_loss: 181.4053, var_loss: 0.0052, anneal_factor: 0.0093, effective_beta_factor: 0.0009
Epoch 10 - Train Loss: 182.7771
Sub-losses: recon_loss: 182.7612, var_loss: 0.0159, anneal_factor: 0.0180, effective_beta_factor: 0.0018
Epoch 10 - Valid Loss: 178.7134
Sub-losses: recon_loss: 178.7029, var_loss: 0.0104, anneal_factor: 0.0180, effective_beta_factor: 0.0018
Epoch 11 - Train Loss: 179.8746
Sub-losses: recon_loss: 179.8428, var_loss: 0.0318, anneal_factor: 0.0344, effective_beta_factor: 0.0034
Epoch 11 - Valid Loss: 175.9603
Sub-losses: recon_loss: 175.9353, var_loss: 0.0250, anneal_factor: 0.0344, effective_beta_factor: 0.0034
Epoch 12 - Train Loss: 177.6577
Sub-losses: recon_loss: 177.5888, var_loss: 0.0689, anneal_factor: 0.0650, effective_beta_factor: 0.0065
Epoch 12 - Valid Loss: 178.2489
Sub-losses: recon_loss: 178.2000, var_loss: 0.0489, anneal_factor: 0.0650, effective_beta_factor: 0.0065
Epoch 13 - Train Loss: 177.2356
Sub-losses: recon_loss: 177.1096, var_loss: 0.1260, anneal_factor: 0.1192, effective_beta_factor: 0.0119
Epoch 13 - Valid Loss: 174.5106
Sub-losses: recon_loss: 174.4299, var_loss: 0.0807, anneal_factor: 0.1192, effective_beta_factor: 0.0119
Epoch 14 - Train Loss: 176.5659
Sub-losses: recon_loss: 176.3190, var_loss: 0.2469, anneal_factor: 0.2086, effective_beta_factor: 0.0209
Epoch 14 - Valid Loss: 174.3225
Sub-losses: recon_loss: 174.1803, var_loss: 0.1422, anneal_factor: 0.2086, effective_beta_factor: 0.0209
Epoch 15 - Train Loss: 173.8222
Sub-losses: recon_loss: 173.4113, var_loss: 0.4109, anneal_factor: 0.3392, effective_beta_factor: 0.0339
Epoch 15 - Valid Loss: 176.3353
Sub-losses: recon_loss: 176.0875, var_loss: 0.2478, anneal_factor: 0.3392, effective_beta_factor: 0.0339
Epoch 16 - Train Loss: 172.2794
Sub-losses: recon_loss: 171.6635, var_loss: 0.6159, anneal_factor: 0.5000, effective_beta_factor: 0.0500
Epoch 16 - Valid Loss: 174.5671
Sub-losses: recon_loss: 174.2156, var_loss: 0.3515, anneal_factor: 0.5000, effective_beta_factor: 0.0500
Epoch 17 - Train Loss: 170.2685
Sub-losses: recon_loss: 169.4155, var_loss: 0.8530, anneal_factor: 0.6608, effective_beta_factor: 0.0661
Epoch 17 - Valid Loss: 173.7806
Sub-losses: recon_loss: 173.2976, var_loss: 0.4829, anneal_factor: 0.6608, effective_beta_factor: 0.0661
Epoch 18 - Train Loss: 170.7247
Sub-losses: recon_loss: 169.6572, var_loss: 1.0675, anneal_factor: 0.7914, effective_beta_factor: 0.0791
Epoch 18 - Valid Loss: 173.0292
Sub-losses: recon_loss: 172.3982, var_loss: 0.6310, anneal_factor: 0.7914, effective_beta_factor: 0.0791
Epoch 19 - Train Loss: 169.7708
Sub-losses: recon_loss: 168.5970, var_loss: 1.1737, anneal_factor: 0.8808, effective_beta_factor: 0.0881
Epoch 19 - Valid Loss: 174.4595
Sub-losses: recon_loss: 173.7771, var_loss: 0.6825, anneal_factor: 0.8808, effective_beta_factor: 0.0881
Epoch 20 - Train Loss: 168.6560
Sub-losses: recon_loss: 167.3547, var_loss: 1.3013, anneal_factor: 0.9350, effective_beta_factor: 0.0935
Epoch 20 - Valid Loss: 173.9368
Sub-losses: recon_loss: 173.2402, var_loss: 0.6966, anneal_factor: 0.9350, effective_beta_factor: 0.0935
Epoch 21 - Train Loss: 168.6468
Sub-losses: recon_loss: 167.3661, var_loss: 1.2808, anneal_factor: 0.9656, effective_beta_factor: 0.0966
Epoch 21 - Valid Loss: 173.9302
Sub-losses: recon_loss: 173.1697, var_loss: 0.7605, anneal_factor: 0.9656, effective_beta_factor: 0.0966
Epoch 22 - Train Loss: 167.6626
Sub-losses: recon_loss: 166.2957, var_loss: 1.3668, anneal_factor: 0.9820, effective_beta_factor: 0.0982
Epoch 22 - Valid Loss: 171.7076
Sub-losses: recon_loss: 170.9284, var_loss: 0.7792, anneal_factor: 0.9820, effective_beta_factor: 0.0982
Epoch 23 - Train Loss: 166.7637
Sub-losses: recon_loss: 165.4086, var_loss: 1.3551, anneal_factor: 0.9907, effective_beta_factor: 0.0991
Epoch 23 - Valid Loss: 171.9471
Sub-losses: recon_loss: 171.1123, var_loss: 0.8348, anneal_factor: 0.9907, effective_beta_factor: 0.0991
Epoch 24 - Train Loss: 164.6296
Sub-losses: recon_loss: 163.2627, var_loss: 1.3669, anneal_factor: 0.9952, effective_beta_factor: 0.0995
Epoch 24 - Valid Loss: 170.6566
Sub-losses: recon_loss: 169.8342, var_loss: 0.8224, anneal_factor: 0.9952, effective_beta_factor: 0.0995
Epoch 25 - Train Loss: 165.6159
Sub-losses: recon_loss: 164.1586, var_loss: 1.4573, anneal_factor: 0.9975, effective_beta_factor: 0.0998
Epoch 25 - Valid Loss: 170.8879
Sub-losses: recon_loss: 170.0140, var_loss: 0.8739, anneal_factor: 0.9975, effective_beta_factor: 0.0998
Epoch 26 - Train Loss: 163.7932
Sub-losses: recon_loss: 162.3245, var_loss: 1.4687, anneal_factor: 0.9987, effective_beta_factor: 0.0999
Epoch 26 - Valid Loss: 169.5822
Sub-losses: recon_loss: 168.7289, var_loss: 0.8533, anneal_factor: 0.9987, effective_beta_factor: 0.0999
Epoch 27 - Train Loss: 161.9896
Sub-losses: recon_loss: 160.4601, var_loss: 1.5295, anneal_factor: 0.9993, effective_beta_factor: 0.0999
Epoch 27 - Valid Loss: 169.6479
Sub-losses: recon_loss: 168.8129, var_loss: 0.8350, anneal_factor: 0.9993, effective_beta_factor: 0.0999
Epoch 28 - Train Loss: 162.4318
Sub-losses: recon_loss: 160.8788, var_loss: 1.5529, anneal_factor: 0.9997, effective_beta_factor: 0.1000
Epoch 28 - Valid Loss: 169.6541
Sub-losses: recon_loss: 168.7791, var_loss: 0.8749, anneal_factor: 0.9997, effective_beta_factor: 0.1000
Epoch 29 - Train Loss: 162.1046
Sub-losses: recon_loss: 160.5550, var_loss: 1.5496, anneal_factor: 0.9998, effective_beta_factor: 0.1000
Epoch 29 - Valid Loss: 168.5911
Sub-losses: recon_loss: 167.6339, var_loss: 0.9572, anneal_factor: 0.9998, effective_beta_factor: 0.1000
Epoch 30 - Train Loss: 160.8321
Sub-losses: recon_loss: 159.2459, var_loss: 1.5862, anneal_factor: 0.9999, effective_beta_factor: 0.1000
Epoch 30 - Valid Loss: 165.0649
Sub-losses: recon_loss: 164.0801, var_loss: 0.9848, anneal_factor: 0.9999, effective_beta_factor: 0.1000
Preprocessor saved successfully.
Pipeline object saved successfully.

Now we can load the pipeline and all intermediate data is kept.

In [9]:

loaded_all_varix = acx.Varix().load(file_path="new_varix.pkl")

loaded_all_varix = acx.Varix().load(file_path="new_varix.pkl")

Attempting to load a pipeline from ./new_varix.pkl...
Pipeline object loaded successfully. Actual type: Varix
Preprocessor loaded successfully.

In [10]:

loaded_all_varix.visualize()

loaded_all_varix.visualize()

In [11]:

loaded_all_varix.show_result()

loaded_all_varix.show_result()

Creating plots ...

And we see that the loss plots from the initial training are still present.

Implications for Ontix and XModalix¶

These two pipelines offer more/other visualizations.

❗❗ Requirements: Getting Tutorial Data for XModalixj ❗❗¶

The data for this tutorial is hosted on Hugging Face Hub (autoencodix/tcga) and is downloaded automatically in the cell below on first run.

XModalix¶

In [ ]:

import os
import autoencodix as acx

from autoencodix.configs.xmodalix_config import XModalixConfig
from autoencodix.configs.default_config import DataCase

p = os.getcwd()
d = "autoencodix_package"
if d not in p:
    raise FileNotFoundError(f"'{d}' not found in path: {p}")
os.chdir(os.sep.join(p.split(os.sep)[: p.split(os.sep).index(d) + 1]))
print(f"Changed to: {os.getcwd()}")

# ---------------------------------------------------------------------
# Data is hosted on Hugging Face Hub and downloaded (and locally cached)
# automatically, then placed under the paths used below
# ---------------------------------------------------------------------
import shutil
import zipfile
import pandas as pd
from huggingface_hub import hf_hub_download

HF_REPO_ID = "autoencodix/tcga"
xmod_data_root = "data/XModalix-Tut-data"
os.makedirs(xmod_data_root, exist_ok=True)

rna_path = hf_hub_download(repo_id=HF_REPO_ID, repo_type="dataset", filename="rna.parquet")
shutil.copyfile(rna_path, os.path.join(xmod_data_root, "combined_rnaseq_formatted.parquet"))

clin_path = hf_hub_download(repo_id=HF_REPO_ID, repo_type="dataset", filename="clinical.parquet")
mapping_path = hf_hub_download(
    repo_id=HF_REPO_ID, repo_type="dataset", filename="xmodalix/tcga_image_mappings.txt"
)
# The base clinical file has no image column; the XModalix image mapping is a separate,
# per-architecture extension that we join in here (left join: not every sample has an image).
clin_df = pd.read_parquet(clin_path)
mapping_df = pd.read_csv(mapping_path, sep="\t", index_col="sample_ids")
clin_df["img_paths"] = mapping_df["img_paths"]
clin_df.to_parquet(os.path.join(xmod_data_root, "combined_clin_formatted.parquet"))

images_zip = hf_hub_download(
    repo_id=HF_REPO_ID, repo_type="dataset", filename="xmodalix/tcga_fake_images.zip"
)
images_extract_dir = os.path.join(xmod_data_root, "images")
if not os.path.isdir(os.path.join(images_extract_dir, "tcga_fake")):
    with zipfile.ZipFile(images_zip) as zf:
        zf.extractall(images_extract_dir)

clin_file = os.path.join("data/XModalix-Tut-data/combined_clin_formatted.parquet")
rna_file = os.path.join("data/XModalix-Tut-data/combined_rnaseq_formatted.parquet")
img_root = os.path.join("data/XModalix-Tut-data/images/tcga_fake")

xmodalix_config = XModalixConfig(
    checkpoint_interval=5,
    class_param="CANCER_TYPE_ACRONYM",
    epochs=30,
    latent_dim=8,
    requires_paired=False,
    pretrain_epochs=2,
    data_case=DataCase.IMG_TO_BULK,
    data_config=DataConfig(
        annotation_columns=["CANCER_TYPE_ACRONYM"],
        data_info={
            "img": DataInfo(
                file_path=img_root,
                data_type="IMG",
                scaling="MINMAX",
                translate_direction="to",
                pretrain_epochs=2,
                # extra_anno_file=imganno_file,
            ),
            "rna": DataInfo(
                file_path=rna_file,
                data_type="NUMERIC",
                scaling="MINMAX",
                translate_direction="from",
            ),
            "anno": DataInfo(file_path=clin_file, data_type="ANNOTATION", sep="\t"),
        },
    ),
)

xmodalix = acx.XModalix(config=xmodalix_config)
result = xmodalix.run()
outpath = os.path.join("tutorial_res", "xmodalix.pkl")
xmodalix.save(file_path=outpath)

import os import autoencodix as acx from autoencodix.configs.xmodalix_config import XModalixConfig from autoencodix.configs.default_config import DataCase p = os.getcwd() d = "autoencodix_package" if d not in p: raise FileNotFoundError(f"'{d}' not found in path: {p}") os.chdir(os.sep.join(p.split(os.sep)[: p.split(os.sep).index(d) + 1])) print(f"Changed to: {os.getcwd()}") # --------------------------------------------------------------------- # Data is hosted on Hugging Face Hub and downloaded (and locally cached) # automatically, then placed under the paths used below # --------------------------------------------------------------------- import shutil import zipfile import pandas as pd from huggingface_hub import hf_hub_download HF_REPO_ID = "autoencodix/tcga" xmod_data_root = "data/XModalix-Tut-data" os.makedirs(xmod_data_root, exist_ok=True) rna_path = hf_hub_download(repo_id=HF_REPO_ID, repo_type="dataset", filename="rna.parquet") shutil.copyfile(rna_path, os.path.join(xmod_data_root, "combined_rnaseq_formatted.parquet")) clin_path = hf_hub_download(repo_id=HF_REPO_ID, repo_type="dataset", filename="clinical.parquet") mapping_path = hf_hub_download( repo_id=HF_REPO_ID, repo_type="dataset", filename="xmodalix/tcga_image_mappings.txt" ) # The base clinical file has no image column; the XModalix image mapping is a separate, # per-architecture extension that we join in here (left join: not every sample has an image). clin_df = pd.read_parquet(clin_path) mapping_df = pd.read_csv(mapping_path, sep="\t", index_col="sample_ids") clin_df["img_paths"] = mapping_df["img_paths"] clin_df.to_parquet(os.path.join(xmod_data_root, "combined_clin_formatted.parquet")) images_zip = hf_hub_download( repo_id=HF_REPO_ID, repo_type="dataset", filename="xmodalix/tcga_fake_images.zip" ) images_extract_dir = os.path.join(xmod_data_root, "images") if not os.path.isdir(os.path.join(images_extract_dir, "tcga_fake")): with zipfile.ZipFile(images_zip) as zf: zf.extractall(images_extract_dir) clin_file = os.path.join("data/XModalix-Tut-data/combined_clin_formatted.parquet") rna_file = os.path.join("data/XModalix-Tut-data/combined_rnaseq_formatted.parquet") img_root = os.path.join("data/XModalix-Tut-data/images/tcga_fake") xmodalix_config = XModalixConfig( checkpoint_interval=5, class_param="CANCER_TYPE_ACRONYM", epochs=30, latent_dim=8, requires_paired=False, pretrain_epochs=2, data_case=DataCase.IMG_TO_BULK, data_config=DataConfig( annotation_columns=["CANCER_TYPE_ACRONYM"], data_info={ "img": DataInfo( file_path=img_root, data_type="IMG", scaling="MINMAX", translate_direction="to", pretrain_epochs=2, # extra_anno_file=imganno_file, ), "rna": DataInfo( file_path=rna_file, data_type="NUMERIC", scaling="MINMAX", translate_direction="from", ), "anno": DataInfo(file_path=clin_file, data_type="ANNOTATION", sep="\t"), }, ), ) xmodalix = acx.XModalix(config=xmodalix_config) result = xmodalix.run() outpath = os.path.join("tutorial_res", "xmodalix.pkl") xmodalix.save(file_path=outpath)

Now we can load the pipeline again

In [13]:

xmodalix_loaded = acx.XModalix.load(outpath)

xmodalix_loaded = acx.XModalix.load(outpath)

Attempting to load a pipeline from tutorial_res/xmodalix.pkl...
Pipeline object loaded successfully. Actual type: XModalix
Preprocessor loaded successfully.

Since we used memory efficient saving, we did not save the data, thus we need to run preprocessing again, this will read and preprocess the files defined in the config.

In [14]:

xmodalix_loaded.preprocess()

xmodalix_loaded.preprocess()

reading parquet: data/XModalix-Tut-data/combined_rnaseq_formatted.parquet
reading parquet: data/XModalix-Tut-data/combined_clin_formatted.parquet
Given image size is possible, rescaling images to: 64x64
Successfully loaded 3230 images for img
anno key: rna
anno key: img
Converting 2261 images to torch.float32 tensors...
Converting 646 images to torch.float32 tensors...
Converting 323 images to torch.float32 tensors...
key: train, type: <class 'dict'>
key: valid, type: <class 'dict'>
key: test, type: <class 'dict'>

No we can use the trained model to predict again:

In [15]:

# now you can use the model to predict with a different pair again:

r = xmodalix_loaded.predict(from_key="rna", to_key="img")

# now you can use the model to predict with a different pair again: r = xmodalix_loaded.predict(from_key="rna", to_key="img")

Prediction complete.
Processing latent space results into a single AnnData object...
Identified source modality for latent space: 'multi_bulk.rna'
  - Added 17450 source feature IDs to .uns
Finished processing latent results.

In [16]:

xmodalix_loaded.visualize()

xmodalix_loaded.visualize()

In [17]:

xmodalix_loaded.show_result()

xmodalix_loaded.show_result()

Creating plots ...
Absolute loss plot not found in the plots dictionary
This happens, when you did not run visualize() or if you saved and loaded the model with `save_all=False`

Ontix¶

In [18]:

import autoencodix as acx
from autoencodix.configs.default_config import DefaultConfig
from autoencodix.configs.ontix_config import OntixConfig

from autoencodix.utils.example_data import (
    EXAMPLE_PROCESSED_DATA,
)

# EXAMPLE_DATA hold PyTorch Datasets (child with extra info) with metdata for train, test and valid splits
processed_data = EXAMPLE_PROCESSED_DATA


ont_lvl1 = dict()
ont_lvl2 = dict()

ont_lvl1["pwy-1"] = ["sub-pwy-1", "sub-pwy-2"]
ont_lvl1["pwy-2"] = ["sub-pwy-2"]
ont_lvl1["pwy-3"] = ["sub-pwy-1", "sub-pwy-3"]
# first third of feature ids in processed_data.train.feature_ids
ont_lvl2["sub-pwy-1"] = processed_data.train.feature_ids[
    : int(len(processed_data.train.feature_ids) / 3)
]
# second third of feature ids in processed_data.train.feature_ids
ont_lvl2["sub-pwy-2"] = processed_data.train.feature_ids[
    int(len(processed_data.train.feature_ids) / 3) : int(
        2 * len(processed_data.train.feature_ids) / 3
    )
]
# last third of feature ids in processed_data.train.feature_ids
ont_lvl2["sub-pwy-3"] = processed_data.train.feature_ids[
    int(2 * len(processed_data.train.feature_ids) / 3) : int(
        len(processed_data.train.feature_ids)
    )
]

# ont_lvl2["sub-pwy-1"] = ["gene-1", "gene-2"]
# ont_lvl2["sub-pwy-2"] = ["gene-3", "gene-4"]
# ont_lvl2["sub-pwy-3"] = ["gene-2", "gene-6"]

ontology_tuple = (ont_lvl1, ont_lvl2)

# Write each dictionary in ontology_tuple to a separate text file
for i, ont_dict in enumerate(ontology_tuple):
    file_name = f"ontology_level_{i + 1}.txt"
    with open(file_name, "w") as f:
        for key, values in ont_dict.items():
            for value in values:
                f.write(f"{value}\t{key}\n")
print("Ontology dictionaries written to ontology_level_1.txt and ontology_level_2.txt")

ont_files = ["ontology_level_1.txt", "ontology_level_2.txt"]

import autoencodix as acx from autoencodix.configs.default_config import DefaultConfig from autoencodix.configs.ontix_config import OntixConfig from autoencodix.utils.example_data import ( EXAMPLE_PROCESSED_DATA, ) # EXAMPLE_DATA hold PyTorch Datasets (child with extra info) with metdata for train, test and valid splits processed_data = EXAMPLE_PROCESSED_DATA ont_lvl1 = dict() ont_lvl2 = dict() ont_lvl1["pwy-1"] = ["sub-pwy-1", "sub-pwy-2"] ont_lvl1["pwy-2"] = ["sub-pwy-2"] ont_lvl1["pwy-3"] = ["sub-pwy-1", "sub-pwy-3"] # first third of feature ids in processed_data.train.feature_ids ont_lvl2["sub-pwy-1"] = processed_data.train.feature_ids[ : int(len(processed_data.train.feature_ids) / 3) ] # second third of feature ids in processed_data.train.feature_ids ont_lvl2["sub-pwy-2"] = processed_data.train.feature_ids[ int(len(processed_data.train.feature_ids) / 3) : int( 2 * len(processed_data.train.feature_ids) / 3 ) ] # last third of feature ids in processed_data.train.feature_ids ont_lvl2["sub-pwy-3"] = processed_data.train.feature_ids[ int(2 * len(processed_data.train.feature_ids) / 3) : int( len(processed_data.train.feature_ids) ) ] # ont_lvl2["sub-pwy-1"] = ["gene-1", "gene-2"] # ont_lvl2["sub-pwy-2"] = ["gene-3", "gene-4"] # ont_lvl2["sub-pwy-3"] = ["gene-2", "gene-6"] ontology_tuple = (ont_lvl1, ont_lvl2) # Write each dictionary in ontology_tuple to a separate text file for i, ont_dict in enumerate(ontology_tuple): file_name = f"ontology_level_{i + 1}.txt" with open(file_name, "w") as f: for key, values in ont_dict.items(): for value in values: f.write(f"{value}\t{key}\n") print("Ontology dictionaries written to ontology_level_1.txt and ontology_level_2.txt") ont_files = ["ontology_level_1.txt", "ontology_level_2.txt"]

Ontology dictionaries written to ontology_level_1.txt and ontology_level_2.txt

In [19]:

ontix = acx.Ontix(
    ontologies=ont_files,
    sep="\t",
    config=OntixConfig(epochs=30, learning_rate=0.005, n_layers=1),
    data=processed_data,
)
result_onitx = ontix.run()

ontix = acx.Ontix( ontologies=ont_files, sep="\t", config=OntixConfig(epochs=30, learning_rate=0.005, n_layers=1), data=processed_data, ) result_onitx = ontix.run()

Ontix checks:
All possible feature names length: 30
Feature order length: 30
Feature names without filtering: 30
Mask layer 0 with shape torch.Size([3, 3]) and 5.0 connections
Mask layer 1 with shape torch.Size([30, 3]) and 30.0 connections
Latent Dim: 3
Epoch 1 - Train Loss: 1221.0144
Sub-losses: recon_loss: 1221.0143, var_loss: 0.0001, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 1 - Valid Loss: 1201.5706
Sub-losses: recon_loss: 1201.5705, var_loss: 0.0002, anneal_factor: 0.0000, effective_beta_factor: 0.0000
Epoch 2 - Train Loss: 1197.5059
Sub-losses: recon_loss: 1197.5055, var_loss: 0.0004, anneal_factor: 0.0001, effective_beta_factor: 0.0001
Epoch 2 - Valid Loss: 1176.2030
Sub-losses: recon_loss: 1176.2024, var_loss: 0.0006, anneal_factor: 0.0001, effective_beta_factor: 0.0001
Epoch 3 - Train Loss: 1162.7103
Sub-losses: recon_loss: 1162.7088, var_loss: 0.0015, anneal_factor: 0.0002, effective_beta_factor: 0.0002
Epoch 3 - Valid Loss: 1132.3454
Sub-losses: recon_loss: 1132.3434, var_loss: 0.0020, anneal_factor: 0.0002, effective_beta_factor: 0.0002
Epoch 4 - Train Loss: 1114.2964
Sub-losses: recon_loss: 1114.2919, var_loss: 0.0045, anneal_factor: 0.0003, effective_beta_factor: 0.0003
Epoch 4 - Valid Loss: 1078.6582
Sub-losses: recon_loss: 1078.6521, var_loss: 0.0062, anneal_factor: 0.0003, effective_beta_factor: 0.0003
Epoch 5 - Train Loss: 1065.5264
Sub-losses: recon_loss: 1065.5127, var_loss: 0.0137, anneal_factor: 0.0007, effective_beta_factor: 0.0007
Epoch 5 - Valid Loss: 1031.6363
Sub-losses: recon_loss: 1031.6191, var_loss: 0.0172, anneal_factor: 0.0007, effective_beta_factor: 0.0007
Epoch 6 - Train Loss: 1025.8335
Sub-losses: recon_loss: 1025.7974, var_loss: 0.0360, anneal_factor: 0.0013, effective_beta_factor: 0.0013
Epoch 6 - Valid Loss: 1003.2790
Sub-losses: recon_loss: 1003.2419, var_loss: 0.0371, anneal_factor: 0.0013, effective_beta_factor: 0.0013
Epoch 7 - Train Loss: 1007.7999
Sub-losses: recon_loss: 1007.7208, var_loss: 0.0791, anneal_factor: 0.0025, effective_beta_factor: 0.0025
Epoch 7 - Valid Loss: 981.5320
Sub-losses: recon_loss: 981.4466, var_loss: 0.0855, anneal_factor: 0.0025, effective_beta_factor: 0.0025
Epoch 8 - Train Loss: 990.8652
Sub-losses: recon_loss: 990.6891, var_loss: 0.1762, anneal_factor: 0.0048, effective_beta_factor: 0.0048
Epoch 8 - Valid Loss: 974.6721
Sub-losses: recon_loss: 974.5013, var_loss: 0.1707, anneal_factor: 0.0048, effective_beta_factor: 0.0048
Epoch 9 - Train Loss: 977.5746
Sub-losses: recon_loss: 977.2104, var_loss: 0.3642, anneal_factor: 0.0093, effective_beta_factor: 0.0093
Epoch 9 - Valid Loss: 955.1447
Sub-losses: recon_loss: 954.7737, var_loss: 0.3711, anneal_factor: 0.0093, effective_beta_factor: 0.0093
Epoch 10 - Train Loss: 974.9158
Sub-losses: recon_loss: 974.1722, var_loss: 0.7436, anneal_factor: 0.0180, effective_beta_factor: 0.0180
Epoch 10 - Valid Loss: 951.4543
Sub-losses: recon_loss: 950.7306, var_loss: 0.7237, anneal_factor: 0.0180, effective_beta_factor: 0.0180
Epoch 11 - Train Loss: 956.8427
Sub-losses: recon_loss: 955.2451, var_loss: 1.5976, anneal_factor: 0.0344, effective_beta_factor: 0.0344
Epoch 11 - Valid Loss: 941.5319
Sub-losses: recon_loss: 939.9884, var_loss: 1.5435, anneal_factor: 0.0344, effective_beta_factor: 0.0344
Epoch 12 - Train Loss: 955.6375
Sub-losses: recon_loss: 952.4723, var_loss: 3.1652, anneal_factor: 0.0650, effective_beta_factor: 0.0650
Epoch 12 - Valid Loss: 935.6899
Sub-losses: recon_loss: 932.8172, var_loss: 2.8727, anneal_factor: 0.0650, effective_beta_factor: 0.0650
Epoch 13 - Train Loss: 951.1653
Sub-losses: recon_loss: 945.4156, var_loss: 5.7497, anneal_factor: 0.1192, effective_beta_factor: 0.1192
Epoch 13 - Valid Loss: 930.1309
Sub-losses: recon_loss: 924.4477, var_loss: 5.6832, anneal_factor: 0.1192, effective_beta_factor: 0.1192
Epoch 14 - Train Loss: 947.8176
Sub-losses: recon_loss: 937.2625, var_loss: 10.5552, anneal_factor: 0.2086, effective_beta_factor: 0.2086
Epoch 14 - Valid Loss: 928.7065
Sub-losses: recon_loss: 918.8953, var_loss: 9.8112, anneal_factor: 0.2086, effective_beta_factor: 0.2086
Epoch 15 - Train Loss: 952.3984
Sub-losses: recon_loss: 936.3096, var_loss: 16.0888, anneal_factor: 0.3392, effective_beta_factor: 0.3392
Epoch 15 - Valid Loss: 929.8787
Sub-losses: recon_loss: 914.7768, var_loss: 15.1018, anneal_factor: 0.3392, effective_beta_factor: 0.3392
Epoch 16 - Train Loss: 949.2610
Sub-losses: recon_loss: 927.5621, var_loss: 21.6989, anneal_factor: 0.5000, effective_beta_factor: 0.5000
Epoch 16 - Valid Loss: 933.7573
Sub-losses: recon_loss: 913.9474, var_loss: 19.8100, anneal_factor: 0.5000, effective_beta_factor: 0.5000
Epoch 17 - Train Loss: 950.3796
Sub-losses: recon_loss: 924.5725, var_loss: 25.8071, anneal_factor: 0.6608, effective_beta_factor: 0.6608
Epoch 17 - Valid Loss: 928.1723
Sub-losses: recon_loss: 905.7023, var_loss: 22.4700, anneal_factor: 0.6608, effective_beta_factor: 0.6608
Epoch 18 - Train Loss: 951.6113
Sub-losses: recon_loss: 923.3481, var_loss: 28.2631, anneal_factor: 0.7914, effective_beta_factor: 0.7914
Epoch 18 - Valid Loss: 928.5229
Sub-losses: recon_loss: 903.8937, var_loss: 24.6293, anneal_factor: 0.7914, effective_beta_factor: 0.7914
Epoch 19 - Train Loss: 941.8345
Sub-losses: recon_loss: 913.8564, var_loss: 27.9781, anneal_factor: 0.8808, effective_beta_factor: 0.8808
Epoch 19 - Valid Loss: 926.0874
Sub-losses: recon_loss: 902.1671, var_loss: 23.9202, anneal_factor: 0.8808, effective_beta_factor: 0.8808
Epoch 20 - Train Loss: 940.5006
Sub-losses: recon_loss: 913.8868, var_loss: 26.6137, anneal_factor: 0.9350, effective_beta_factor: 0.9350
Epoch 20 - Valid Loss: 920.9032
Sub-losses: recon_loss: 897.3171, var_loss: 23.5861, anneal_factor: 0.9350, effective_beta_factor: 0.9350
Epoch 21 - Train Loss: 938.2790
Sub-losses: recon_loss: 913.3347, var_loss: 24.9443, anneal_factor: 0.9656, effective_beta_factor: 0.9656
Epoch 21 - Valid Loss: 918.9359
Sub-losses: recon_loss: 895.8992, var_loss: 23.0367, anneal_factor: 0.9656, effective_beta_factor: 0.9656
Epoch 22 - Train Loss: 929.9743
Sub-losses: recon_loss: 905.7796, var_loss: 24.1947, anneal_factor: 0.9820, effective_beta_factor: 0.9820
Epoch 22 - Valid Loss: 918.0283
Sub-losses: recon_loss: 895.6837, var_loss: 22.3446, anneal_factor: 0.9820, effective_beta_factor: 0.9820
Epoch 23 - Train Loss: 933.9519
Sub-losses: recon_loss: 910.1971, var_loss: 23.7548, anneal_factor: 0.9907, effective_beta_factor: 0.9907
Epoch 23 - Valid Loss: 911.7701
Sub-losses: recon_loss: 891.0258, var_loss: 20.7443, anneal_factor: 0.9907, effective_beta_factor: 0.9907
Epoch 24 - Train Loss: 926.4578
Sub-losses: recon_loss: 903.6484, var_loss: 22.8094, anneal_factor: 0.9952, effective_beta_factor: 0.9952
Epoch 24 - Valid Loss: 914.0869
Sub-losses: recon_loss: 893.2252, var_loss: 20.8617, anneal_factor: 0.9952, effective_beta_factor: 0.9952
Epoch 25 - Train Loss: 923.0073
Sub-losses: recon_loss: 900.3552, var_loss: 22.6521, anneal_factor: 0.9975, effective_beta_factor: 0.9975
Epoch 25 - Valid Loss: 905.8131
Sub-losses: recon_loss: 885.7856, var_loss: 20.0275, anneal_factor: 0.9975, effective_beta_factor: 0.9975
Epoch 26 - Train Loss: 920.1220
Sub-losses: recon_loss: 898.1062, var_loss: 22.0159, anneal_factor: 0.9987, effective_beta_factor: 0.9987
Epoch 26 - Valid Loss: 905.8649
Sub-losses: recon_loss: 884.6357, var_loss: 21.2292, anneal_factor: 0.9987, effective_beta_factor: 0.9987
Epoch 27 - Train Loss: 919.9186
Sub-losses: recon_loss: 898.6025, var_loss: 21.3161, anneal_factor: 0.9993, effective_beta_factor: 0.9993
Epoch 27 - Valid Loss: 905.4550
Sub-losses: recon_loss: 885.2871, var_loss: 20.1680, anneal_factor: 0.9993, effective_beta_factor: 0.9993
Epoch 28 - Train Loss: 913.9538
Sub-losses: recon_loss: 891.5910, var_loss: 22.3627, anneal_factor: 0.9997, effective_beta_factor: 0.9997
Epoch 28 - Valid Loss: 896.2276
Sub-losses: recon_loss: 876.1357, var_loss: 20.0919, anneal_factor: 0.9997, effective_beta_factor: 0.9997
Epoch 29 - Train Loss: 918.2080
Sub-losses: recon_loss: 897.0868, var_loss: 21.1212, anneal_factor: 0.9998, effective_beta_factor: 0.9998
Epoch 29 - Valid Loss: 895.7843
Sub-losses: recon_loss: 875.7118, var_loss: 20.0725, anneal_factor: 0.9998, effective_beta_factor: 0.9998
Epoch 30 - Train Loss: 910.1649
Sub-losses: recon_loss: 889.1332, var_loss: 21.0318, anneal_factor: 0.9999, effective_beta_factor: 0.9999
Epoch 30 - Valid Loss: 897.4055
Sub-losses: recon_loss: 877.5325, var_loss: 19.8730, anneal_factor: 0.9999, effective_beta_factor: 0.9999

In [20]:

ontix.save(file_path="ontix_backup.pkl")

ontix.save(file_path="ontix_backup.pkl")

Preprocessor saved successfully.
saving memory efficient
Ontix checks:
All possible feature names length: 30
Feature order length: 30
Feature names without filtering: 30
Mask layer 0 with shape torch.Size([3, 3]) and 5.0 connections
Mask layer 1 with shape torch.Size([30, 3]) and 30.0 connections
Latent Dim: 3
Pipeline object saved successfully.

In [21]:

ontix_loaded = acx.Ontix(ontologies=ont_files).load(file_path="ontix_backup.pkl")

ontix_loaded = acx.Ontix(ontologies=ont_files).load(file_path="ontix_backup.pkl")

Attempting to load a pipeline from ./ontix_backup.pkl...
Pipeline object loaded successfully. Actual type: Ontix
Preprocessor loaded successfully.

In [22]:

pred_res = ontix_loaded.predict(EXAMPLE_PROCESSED_DATA)

pred_res = ontix_loaded.predict(EXAMPLE_PROCESSED_DATA)

In [23]:

ontix_loaded.visualize()

ontix_loaded.evaluate(params=["cluster"])

ontix_loaded.visualize() ontix_loaded.evaluate(params=["cluster"])

Perform ML task with feature df: Latent
Latent
Perform ML task for target parameter: cluster
Perform ML task with feature df: Latent
Latent
Perform ML task for target parameter: cluster

Out[23]:

Result Object Public Attributes:
------------------------------
latentspaces: TrainingDynamics object
sample_ids: TrainingDynamics object
reconstructions: TrainingDynamics object
mus: TrainingDynamics object
sigmas: TrainingDynamics object
losses: TrainingDynamics object
sub_losses: LossRegistry(_losses={})
preprocessed_data: Tensor of shape (0,)
model: OntixArchitecture
model_checkpoints: TrainingDynamics object
datasets: DatasetContainer(train=None, valid=None, test=None)
new_datasets: DatasetContainer(train=<autoencodix.data._numeric_dataset.NumericDataset object at 0x30eb85ab0>, valid=<autoencodix.data._numeric_dataset.NumericDataset object at 0x30e8bf8b0>, test=<autoencodix.data._numeric_dataset.NumericDataset object at 0x30e8bf670>)
adata_latent: AnnData object with n_obs × n_vars = 201 × 3
    uns: 'var_names'
final_reconstruction: None
sub_results: None
sub_reconstructions: None
embedding_evaluation:   cv_run score_split CLINIC_PARAM       metric     value  \
0   CV_1        test      cluster  roc_auc_ovo  0.967187   
1   CV_2        test      cluster  roc_auc_ovo  0.978906   
2   CV_3        test      cluster  roc_auc_ovo  0.989062   
3   CV_4        test      cluster  roc_auc_ovo  0.976562   
4   CV_5        test      cluster  roc_auc_ovo  0.951562   
5   CV_1       train      cluster  roc_auc_ovo  0.975391   
6   CV_2       train      cluster  roc_auc_ovo  0.973736   
7   CV_3       train      cluster  roc_auc_ovo  0.971341   
8   CV_4       train      cluster  roc_auc_ovo  0.972955   
9   CV_5       train      cluster  roc_auc_ovo  0.980029   
0   CV_1        test      cluster  roc_auc_ovo  0.967187   
1   CV_2        test      cluster  roc_auc_ovo  0.978906   
2   CV_3        test      cluster  roc_auc_ovo  0.989062   
3   CV_4        test      cluster  roc_auc_ovo  0.976562   
4   CV_5        test      cluster  roc_auc_ovo  0.951562   
5   CV_1       train      cluster  roc_auc_ovo  0.975391   
6   CV_2       train      cluster  roc_auc_ovo  0.973736   
7   CV_3       train      cluster  roc_auc_ovo  0.971341   
8   CV_4       train      cluster  roc_auc_ovo  0.972955   
9   CV_5       train      cluster  roc_auc_ovo  0.980029   

                 ML_ALG         ML_TYPE ML_TASK ML_SUBTASK  
0  LogisticRegression()  classification  Latent     Latent  
1  LogisticRegression()  classification  Latent     Latent  
2  LogisticRegression()  classification  Latent     Latent  
3  LogisticRegression()  classification  Latent     Latent  
4  LogisticRegression()  classification  Latent     Latent  
5  LogisticRegression()  classification  Latent     Latent  
6  LogisticRegression()  classification  Latent     Latent  
7  LogisticRegression()  classification  Latent     Latent  
8  LogisticRegression()  classification  Latent     Latent  
9  LogisticRegression()  classification  Latent     Latent  
0  LogisticRegression()  classification  Latent     Latent  
1  LogisticRegression()  classification  Latent     Latent  
2  LogisticRegression()  classification  Latent     Latent  
3  LogisticRegression()  classification  Latent     Latent  
4  LogisticRegression()  classification  Latent     Latent  
5  LogisticRegression()  classification  Latent     Latent  
6  LogisticRegression()  classification  Latent     Latent  
7  LogisticRegression()  classification  Latent     Latent  
8  LogisticRegression()  classification  Latent     Latent  
9  LogisticRegression()  classification  Latent     Latent  
embedding_attributions: Empty DataFrame
Columns: []
Index: []
embedding_explanations: Dict with 0 items

In [24]:

ontix_loaded.show_result()

ontix_loaded.show_result()

Creating plots ...

<Figure size 0x500 with 0 Axes>