config_anndata_eval.yml

debug: True
profile: False
data:
  output_path: './eval_pbmc_toy'
  # seq2cells model checkpoint
  model_chkpt_path: "./checkpoint.ckpt"
  # anndata object hosting the observations and precomputed embeddings
  # Requires embeddings stored in the .varm key 'seq_embedding'
  ann_data_file: "./tests/resources/single_cell_data/pbmc_toy_example.h5ad"
  # column (obs.) name indicating the training test and validation split
  split_name: 'enf_set'
  # which layer of the anndata object layers to use as observed counts
  # usually this should be a normalised counts layer e.g. 'pflog1ppf'
  # use 'X' to indicate to use whatever is stored as X in the adata object
  observed_counts_layer: 'X'
  # <predictions_layer> does not need to be present if running the predictions.
  # In that case it will be used as layer name to store the predicted counts in
  # the new data frame. If Running only the correlation computation this is
  # the layer in which the predicted counts are expected
  predictions_layer: 'predicted'
  save_predictions: True
  # full path where to store AnnData with predictions
  save_anndata_path: "./pbmc_toy_example_with_predictions.h5ad"
task:
  run_pred: True
  # if running not on 'all' than will save a subset of the anndata object
  # with genes corresponding to the chosen set
  pred_on: 'all'
  # if predictions under are found in the provided anndata object under the
  # specified 'predictions_layer' should they be overwritten?
  overwrite_pred: True
  run_eval: True
  eval_on: 'all'
  # optional set <subset_genes_column> to a anndata.obs column (0 and 1) that
  # should be used to subset the observations (genes)
  subset_genes_column: None
  eval_highly_variable: True
  eval_highly_expressed: True
resource:
  device: "cpu"
  # if <pred_batch_size> set to 0 will use
  # the entire length of the (subset) anndata object
  pred_batch_size: 10
  # if to write predictions per batch to disk (True) or
  # (False) perform all in memory
  temp_write_pred: True
  # read anndata in backed mode?
  backed_mode: False
  # is running on a subset of data in backed mode (e.g. validation only) this
  # will require to write a temporary h5ad file that will be deleted after
  # running
  backed_mode_temp_h5ad: ""