Python lfads.LFADS属性代码示例

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def write_model_runs(hps, datasets, output_fname=None):
  """Run the model on the data in data_dict, and save the computed values.

  LFADS generates a number of outputs for each examples, and these are all
  saved.  They are:
    The mean and variance of the prior of g0.
    The mean and variance of approximate posterior of g0.
    The control inputs (if enabled)
    The initial conditions, g0, for all examples.
    The generator states for all time.
    The factors for all time.
    The rates for all time.

  Args:
    hps: The dictionary of hyperparameters.
    datasets: A dictionary of data dictionaries.  The dataset dict is simply a
      name(string)-> data dictionary mapping (See top of lfads.py).
    output_fname (optional): output filename stem to write the model runs.
  """
  model = build_model(hps, kind=hps.kind, datasets=datasets)
  model.write_model_runs(datasets, output_fname) 

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def write_model_parameters(hps, output_fname=None, datasets=None):
  """Save all the model parameters

  Save all the parameters to hps.lfads_save_dir.

  Args:
    hps: The dictionary of hyperparameters.
    output_fname: The prefix of the file in which to save the generated
      samples.
    datasets: A dictionary of data dictionaries.  The dataset dict is simply a
      name(string)-> data dictionary mapping (See top of lfads.py).
  """
  if not output_fname:
    output_fname = "model_params"
  else:
    output_fname = output_fname + "_model_params"
  fname = os.path.join(hps.lfads_save_dir, output_fname)
  print("Writing model parameters to: ", fname)
  # save the optimizer params as well
  model = build_model(hps, kind="write_model_params", datasets=datasets) 
  model_params = model.eval_model_parameters(use_nested=False,
                                             include_strs="LFADS")
  utils.write_data(fname, model_params, compression=None)
  print("Done.") 

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def write_model_parameters(hps, output_fname=None, datasets=None):
  """Save all the model parameters

  Save all the parameters to hps.lfads_save_dir.

  Args:
    hps: The dictionary of hyperparameters.
    output_fname: The prefix of the file in which to save the generated
      samples.
    datasets: A dictionary of data dictionaries.  The dataset dict is simply a
      name(string)-> data dictionary mapping (See top of lfads.py).
  """
  if not output_fname:
    output_fname = "model_params"
  else:
    output_fname = output_fname + "_model_params"
  fname = os.path.join(hps.lfads_save_dir, output_fname)
  print("Writing model parameters to: ", fname)
  # save the optimizer params as well
  model = build_model(hps, kind="write_model_params", datasets=datasets)
  model_params = model.eval_model_parameters(use_nested=False,
                                             include_strs="LFADS")
  utils.write_data(fname, model_params, compression=None)
  print("Done.") 

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def train(hps, datasets):
  """Train the LFADS model.

  Args:
    hps: The dictionary of hyperparameters.
    datasets: A dictionary of data dictionaries.  The dataset dict is simply a
      name(string)-> data dictionary mapping (See top of lfads.py).
  """
  model = build_model(hps, kind="train", datasets=datasets)
  if hps.do_reset_learning_rate:
    sess = tf.get_default_session()
    sess.run(model.learning_rate.initializer)

  model.train_model(datasets) 

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def write_model_samples(hps, datasets, dataset_name=None, output_fname=None):
  """Use the prior distribution to generate samples from the model.
  Generates batch_size number of samples (set through FLAGS).

  LFADS generates a number of outputs for each examples, and these are all
  saved.  They are:
    The mean and variance of the prior of g0.
    The control inputs (if enabled)
    The initial conditions, g0, for all examples.
    The generator states for all time.
    The factors for all time.
    The output distribution parameters (e.g. rates) for all time.

  Args:
    hps: The dictionary of hyperparameters.
    datasets: A dictionary of data dictionaries.  The dataset dict is simply a
      name(string)-> data dictionary mapping (See top of lfads.py).
    dataset_name: The name of the dataset to grab the factors -> rates
      alignment matrices from. Only a concern with models trained on
      multi-session data. By default, uses the first dataset in the data dict.
    output_fname: The name prefix of the file in which to save the generated
      samples.
  """
  if not output_fname:
    output_fname = "model_runs_" + hps.kind
  else:
    output_fname = output_fname + "model_runs_" + hps.kind
  if not dataset_name:
    dataset_name = datasets.keys()[0]
  else:
    if dataset_name not in datasets.keys():
      raise ValueError("Invalid dataset name '%s'."%(dataset_name))
  model = build_model(hps, kind=hps.kind, datasets=datasets)
  model.write_model_samples(dataset_name, output_fname) 

# 需要导入模块: import lfads [as 别名]
# 或者: from lfads import LFADS [as 别名]
def clean_data_dict(data_dict):
  """Add some key/value pairs to the data dict, if they are missing.
  Args:
    data_dict - dictionary containing data for LFADS
  Returns:
    data_dict with some keys filled in, if they are absent.
  """

  keys = ['train_truth', 'train_ext_input', 'valid_data',
          'valid_truth', 'valid_ext_input', 'valid_train']
  for k in keys:
    if k not in data_dict:
      data_dict[k] = None

  return data_dict