default.yaml

# YAML default config file for multimodalecho project
# This file stores all default values 
# and can be used as a template for new config files.

# extract_videos and extract_Mmode
# ================================================================= #
extract_videos:
  # directory where Videos folder is stored
  # default: /cluster/dataset/vogtlab/Projects/EchoNet/
  data_dir: "/cluster/dataset/vogtlab/Projects/EchoNet/"
  # directory where output should be stored
  # default: None (is set in script depending on other params)
  output: null
  # split ("all", "TRAIN", "VAL", "TEST") to perform operations on
  # default: all
  split: all
  # length which videos should be cut to (shorter videos are discarded)
  # default: 112
  length: 112

extract_Mmode:
  # directory where Videos folder is stored
  # default: /cluster/dataset/vogtlab/Projects/EchoNet/
  data_dir: "/cluster/dataset/vogtlab/Projects/EchoNet/"
  # directory where output should be stored
  # default: None (is set in script depending on other params)
  output: null
  # split ("all", "TRAIN", "VAL", "TEST") to perform operations on
  # default: all
  split: all
  # type of axes to extract ("default", "informed", "random")
  # default: default
  axis: default
  # number of M-modes to extract
  # default: 2
  num_modes: 2
  # width of resulting M-mode images
  # default: 112
  width: 112
  # seed for random number generators
  # default: 0
  seed: 0

# run_model
# ================================================================= #
run_model:
  # directory where Videos folder is stored
  # default: /cluster/dataset/vogtlab/Projects/EchoNet/
  data_dir: "/cluster/dataset/vogtlab/Projects/EchoNet/"
  # directory where output should be stored
  # default: None (is set in script depending on other params)
  output: null
  # specifies model type to run
  # default: 2d_resnet34
  model_type: "2d_resnet34"
  # whether to use pretrained weights (available only for 2d models)
  # default: False
  pretrained: False
  # whether to perform classification (as opposed to regression)
  # default: False
  classification: False
  # datatype of input data (one of "tensors", "videos", "images")
  # default: "videos"
  datatype: "videos"
  # threshold to use for classification
  # default: 50
  thresh: 50
  # number of epochs
  # default: 90
  num_epochs: 90
  # learning rate
  # default: 1e-4
  lr: 0.0001
  # step size to use for lr scheduler
  # default: 15
  step_size: 15
  # batch size to use during TRAIN and VAL
  # default: 64
  batch_size: 64
  # whether to extract frames (as opposed to M-modes)
  # default: False
  frames: False
  # axis to use for M-mode extraction
  # "default" creates linearly spaced angles around the vertical axis
  # "informed" creates linearly spaced angles around the axis through the left ventricle
  # "random_start" creates linearly spaced angles around a random starting axis
  # "random" leads to randomly selected angles
  # default: "default"
  axis: "default"
  # number of modes (frames or M-modes) to extract
  # default: 2
  num_modes: 2
  # width of M-mode images
  # default 112 (resulting in square 112x112 inputs)
  width: 112
  # whether to perform feature selection (currently removes image with specified index)
  # default: None
  feature_selection: null
  # which type of curriculum learning to perform (if any)
  # default: None
  curriculum: null
  # number of warmup epochs for self-paced learning
  # default: 5
  spl_warmup: 5
  # initial loss threshold for self-paced learning
  # default: 100
  spl_thresh: 100
  # threshold growing factor for self-paced learning
  # default: 1.3
  spl_factor: 1.3
  # quantile to be used for self-paced learning (alternative to the other three params)
  # default: 0.75
  spl_quantile: 0.75
  # initial fraction of train size to use for training
  # default: 0.2
  init_fraction: 0.2
  # fraction of total epochs after which the entire train set is used for training
  # default: 0.8
  epoch_fraction: 0.8
  # device to run computations on 
  # default: None (runs on CUDA whenever available)
  device: null
  # global random seed for numpy and torch
  # default: 0
  seed: 0
  # whether to record tensorboard data (TRAIN and VAL loss over epochs)
  # default: False
  tensorboard: False

# collect_results
# ================================================================= #
collect_results:
  # directory where Videos folder is stored
  # default: output
  data_dir: "output"
  # directory where output should be stored
  # default: None (is set in script depending on other params)
  output: null
  # whether to produce plots
  # default: True
  plot: True
  # whether to print table
  # default: True
  table: True
  # if specified, only results from this model type are collected
  # default: None (means all model types)
  model_type: null
  # if specified, only relevant results are collected
  # default: None
  pretrained: null
    # if specified, only relevant results are collected
  # default: None
  classification: None
  # threshold to use for classification
  # default: 50
  thresh: 50
  # number of epochs
  # default: 90
  num_epochs: 90
  # learning rate
  # default: 1e-4
  lr: 0.0001
  # step size to use for lr scheduler
  # default: 15
  step_size: 15
  # batch size to use during TRAIN and VAL
  # default: 64
  batch_size: 64
  # if specified, only relevant results are collected
  # default: None
  axis: null
  # if specified, only relevant results are collected
  # default: None
  feature_selection: null
  # if specified, only relevant results are collected
  # default: None
  curriculum: null
  # if specified, only relevant results are collected
  # default: None
  spl_warmup: null
  # if specified, only relevant results are collected
  # default: None
  spl_thresh: null
  # if specified, only relevant results are collected
  # default: None
  spl_factor: null
  # quantile to be used for self-paced learning (alternative to the other three params)
  # default: None
  spl_quantile: null
  # if specified, only relevant results are collected
  # default: None
  init_fraction: null
  # if specified, only relevant results are collected
  # default: None
  epoch_fraction: null
  # global random seed for numpy and torch
  # default: 0
  seed: 0


# epoch_plots
# ================================================================= #
epoch_plots:
  # directory where Videos folder is stored
  # default: output
  data_dir: "output"
  # directory where output should be stored
  # default: None (is set in script depending on other params)
  output: "plots"
  # if specified, only results from this model type are collected
  # default: None (means all model types)
  model_type: null
  # if specified, only relevant results are collected
  # default: None
  pretrained: null
    # if specified, only relevant results are collected
  # default: None
  classification: None
  # learning rate
  # default: 1e-4
  lr: 0.0001
  # step size to use for lr scheduler
  # default: 15
  step_size: 15
  # batch size to use during TRAIN and VAL
  # default: 64
  batch_size: 64
  # if specified, only relevant results are collected
  # default: None
  axis: null
  # if specified, only relevant results are collected
  # default: None
  num_modes: null
  # if specified, only relevant results are collected
  # default: None
  feature_selection: null
  # if specified, only relevant results are collected
  # default: None
  curriculum: null
  # if specified, only relevant results are collected
  # default: None
  spl_warmup: null
  # if specified, only relevant results are collected
  # default: None
  spl_thresh: null
  # if specified, only relevant results are collected
  # default: None
  spl_factor: null
  # quantile to be used for self-paced learning (alternative to the other three params)
  # default: None
  spl_quantile: null
  # if specified, only relevant results are collected
  # default: None
  init_fraction: null
  # if specified, only relevant results are collected
  # default: None
  epoch_fraction: null
  # global random seed for numpy and torch
  # default: 0
  seed: 0