adding demo details

demo/README.md

@@ -0,0 +1,13 @@
+## Model Training Demo
+
+We provide the following basic demo as a starting point to training perturbation models using this repo.
+
+To train a SAMS-VAE model on the Replogle dataset, run:
+`python ../train.py --config sams_vae_replogle.yaml`
+
+The example config file, `sams_vae_replogle.yaml`, has been annotated with additional explanation regarding config structure
+and can be used as a starting point for setting up new training runs.
+
+As the model trains, the training metrics and the model checkpoints can be visualized using `visualize_results.ipynb`
+
+For examples of training model sweeps, see the sweep configs and instructions in `../paper/experiments/`

demo/sams_vae_replogle.yaml

@@ -0,0 +1,61 @@
+# Experiment hyperparameters
+name: sams_vae_replogle
+seed: 0
+max_epochs: 1000
+
+# WandB hyperparameters
+# can set to True to save metrics and checkpoints to WandB
+use_wandb: False
+wandb_kwargs.name: sams_vae_replogle_filtered_example
+wandb_kwargs.project: debug
+
+# Data module class + hyperparameters
+# current data module options are:
+#   - ReplogleDataModule
+#   - NormanOODCombinationDataModule
+#   - NormanDataEfficiency
+#   - NormanDataEfficiencyDataModule
+#   - SAMSVAESimulationDataModule
+# data_module_kwargs are the arguments for the __init__
+# function in the corresponding data module classes (see sams_vae/data/)
+data_module: ReplogleDataModule
+data_module_kwargs.batch_size: 512
+
+# Model class + hyperparameters
+# current model options are:
+#   - SAMSVAEModel
+#   - CPAVAEModel
+#   - SVAEPlusModel
+#   - ConditionalVAEModel
+# model_kwargs are the arguments for the __init__ function
+# of the corresponding model classes (see sams_vae/models/)
+model: SAMSVAEModel
+model_kwargs.n_latent: 100
+model_kwargs.mask_prior_prob: 0.01
+model_kwargs.embedding_prior_scale: 1
+model_kwargs.likelihood_key: library_nb
+model_kwargs.decoder_n_layers: 1
+model_kwargs.decoder_n_hidden: 350
+
+# Guide class + hyperparameters
+# Like the models, these correspond to the guide classes
+# in sams_vae/models/
+# Note that the guide must match the model (will be under the same subdirectory,
+# eg sams_vae/models/sams_vae/ or sams_vae/models/cpa_vae/)
+guide: SAMSVAEMeanFieldNormalGuide
+guide_kwargs.n_latent: 100
+guide_kwargs.basal_encoder_n_layers: 1
+guide_kwargs.basal_encoder_n_hidden: 350
+guide_kwargs.basal_encoder_input_normalization: log_standardize
+
+# Loss module class + hyperparameters
+# from same subdirectory as model and guide
+loss_module: SAMSVAE_ELBOLossModule
+
+# Lightning module hyperparameters
+lightning_module_kwargs.lr: 0.001
+lightning_module_kwargs.n_particles: 5
+
+# Predictor class + hyperparameters (used to evaluation)
+# also much match model / guide, in same subdirectory
+predictor: SAMSVAEPredictor