Functions — github.com/athms/learning-from-brains

↓ 14 callersFunctionrun_train_process

( config, use_deepspeed=DEEPSPEED_MODE )

tests/__init__.py:137

↓ 9 callersMethod_pad_seq_right_to_n

( seq: np.ndarray, n: int, pad_value: float = 0 )

src/batcher/base.py:158

↓ 7 callersMethodget_subject_deriv_files

( self, subject, filters, )

src/preprocessor.py:201

↓ 7 callersFunctionrun_tests

()

tests/__init__.py:161

↓ 5 callersMethod_pad_tensor_left_by_n

( tensor, n, pad_value )

src/embedder/base.py:128

↓ 5 callersMethoddataset

Create Pytorch dataset that can be used for training. Args: ----- tarfiles: list List of paths to data fi

src/batcher/bert.py:292

↓ 5 callersFunctionget_all_architectures_with_all_train_styles

()

tests/__init__.py:79

↓ 5 callersMethodget_subject_source_files

( self, subject, filters )

src/preprocessor.py:212

↓ 4 callersMethod_init_embeds

(self)

src/embedder/csm.py:36

↓ 4 callersMethod_move_batch_to_device

( self, batch )

src/trainer/base.py:71

↓ 4 callersMethod_pad_seqs_right_to_n

( self, seqs, n: int=None )

src/batcher/bert.py:230

↓ 4 callersMethodfrom_pretrained

Load pretrained model from pretrained_path. Needs to point to pytorch_model.bin file.

src/model.py:49

↓ 4 callersMethodmake_key

(self, bold_path)

src/preprocessor.py:369

↓ 4 callersMethodpreprocess_bold

( self, bold_path: str=None, t_r: float=None)

src/preprocessor.py:477

↓ 3 callersMethod_combine_seqs_as_is_next

( sample_a: Dict[str,np.ndarray], sample_b: Dict[str,np.ndarray], key: str )

src/batcher/bert.py:155

↓ 3 callersMethod_combine_seqs_as_not_next

( sample_a: Dict[str, np.ndarray], sample_b: Dict[str, np.ndarray], key: str )

src/batcher/bert.py:181

↓ 3 callersMethod_order_seqs_by_is_next

( seqs_next: np.ndarray, seqs_not_next: np.ndarray, is_next: np.ndarray, batch

src/batcher/bert.py:207

↓ 3 callersMethodadd_cls_embed

( self, batch: Dict[str, torch.tensor] )

src/embedder/csm.py:143

↓ 3 callersMethodadd_pooler_layer

(self)

src/decoder/gpt.py:66

↓ 3 callersMethoddecode

( self, outputs: torch.tensor, attention_mask: torch.tensor, )

src/decoder/gpt.py:104

↓ 3 callersMethodforward

Forward pass of model. Args ---- batch: Dict[str, torch.tensor] Input batch (as generated by src

src/model.py:157

↓ 3 callersFunctionget_architectures_with_pretrain_styles

()

tests/__init__.py:41

↓ 3 callersMethodget_bold_tr

(bold_path)

src/preprocessor.py:465

↓ 3 callersFunctionmake_batcher

Make a batcher object. The batcher is used to generate batches of input data for training and evaluation. Args: -----

src/batcher/make.py:6

↓ 3 callersMethodmask_inputs

( self, batch: Dict[str, torch.tensor], )

src/embedder/csm.py:62

↓ 3 callersMethodmasking_loss

( self, masked_inputs, outputs, modelling_mask )

src/embedder/csm.py:214

↓ 3 callersMethodprep_batch

Prepare input batch for forward pass. Calls src.embedder.prep_batch. Args ---- batch: Dict[str, torch.tensor]

src/model.py:143

↓ 3 callersMethodwrite_bold_to_tar

( bold=None, path=None, key=None, sink=None, sample_dict=None,

src/preprocessor.py:386

↓ 2 callersMethod__init__

( self, num_hidden_layers: int = 4, embed_dim: int = 768, dropout: float = 0.2

src/decoder/autoencoder.py:10

↓ 2 callersMethod_get_subject_files

( self, subject, filters, basedir )

src/preprocessor.py:223

↓ 2 callersMethod_identify_bold_markers

( filepath: str=None )

src/preprocessor.py:69

↓ 2 callersMethod_make_bold_filters

( self, filepath: str )

src/preprocessor.py:78

↓ 2 callersMethod_make_dataloader

( self, files, repeat: bool = True, n_shuffle_shards: int = 1000, n_sh

src/batcher/bert.py:264

↓ 2 callersFunction_pad_seq_right_to_n

( seq: np.ndarray, n: int, pad_value: float = 0. )

src/batcher/base.py:10

↓ 2 callersMethod_sample_seq_on_and_len

( self, bold_len: int )

src/batcher/base.py:169

↓ 2 callersMethodadd_decoding_head

( self, num_decoding_classes: int )

src/decoder/linear_baseline.py:42

↓ 2 callersFunctionam_first_deepspeed_child

()

tests/__init__.py:102

↓ 2 callersFunctionclear_matplotlib_fig_cache

()

src/tools/brainmap.py:15

↓ 2 callersFunctionevs_for_task

( task: str )

scripts/dataprep/downstream/hcp_to_bids.py:329

↓ 2 callersFunctionfig_downstream_performance

Script's main function; creates Figure 5 of the manuscript.

scripts/analyses/fig5_downstream_performance.py:20

↓ 2 callersMethodforward_step

( self, inputs: torch.tensor = None, hidden: torch.tensor = None, cell_state:

src/decoder/autoencoder.py:167

↓ 2 callersFunctionget_BERT_with_pretrain_styles

()

tests/__init__.py:22

↓ 2 callersFunctionget_GPT_with_pretrain_styles

()

tests/__init__.py:16

↓ 2 callersFunctionget_NetBERT_with_pretrain_styles

()

tests/__init__.py:29

↓ 2 callersFunctionget_args

Get command line arguments

scripts/train.py:464

↓ 2 callersFunctionget_autoencoder_with_pretrain_styles

()

tests/__init__.py:35

↓ 2 callersFunctionget_config

Make config from command line arguments (as created by get_args()). Performs additional formating of args required for calling train().

scripts/train.py:394

↓ 2 callersMethodget_subjects

( self, derivatives: str=None )

src/preprocessor.py:52

↓ 2 callersMethodmake_bold_dummy

( self, bold_shape: Tuple[int, int], t_r: float, # in seconds f_s: Tuple[float

src/batcher/base.py:207

↓ 2 callersFunctionmake_model

Make model from model_config (as generated by get_config()).

scripts/train.py:314

↓ 2 callersMethodreconstruction_loss

( self, input, target, **kwargs )

src/embedder/base.py:212

↓ 2 callersFunctionsfig_downstream_learning_curves

( config: Dict=None, datasets: Tuple[str, ...]=('HCP', 'ds002105'), base_figname: str='sfig_downst

scripts/analyses/sfig4_5_downstream_learning_curves.py:23

↓ 2 callersFunctiontrain

Model training according to config. -> see get_args() below for all command line arguments.

scripts/train.py:61

↓ 1 callersMethod__init__

(self, in_dim: int = 1024, embed_dim: int = 768, num_hidden_layers: int = 1, d

src/embedder/base.py:83

↓ 1 callersMethod__init__

( self, **kwargs )

src/decoder/bert.py:10

↓ 1 callersMethod__init__

( self, num_hidden_layers: int = 6, num_attention_heads: int = 12, embed_dim:

src/decoder/gpt.py:10

↓ 1 callersMethod__init__

( self, dataloader_a, dataloader_b, length, seq_max, sample_ke

src/batcher/bert.py:11

↓ 1 callersMethod_check_derivatives_path

identify derivatives directory

src/preprocessor.py:118

↓ 1 callersMethod_clear_masker_cache

(self)

src/preprocessor.py:328

↓ 1 callersMethod_collect_preproc_bold_files

(self)

src/preprocessor.py:157

↓ 1 callersMethod_collect_source_bold_files

(self)

src/preprocessor.py:170

↓ 1 callersMethod_combine_to_sample

( self, sample_a: Dict[str, np.ndarray], sample_b: Dict[str, np.ndarray] )

src/batcher/bert.py:45

↓ 1 callersMethod_get_filepaths

( self, path, filters )

src/preprocessor.py:247

↓ 1 callersMethod_identify_func_subdirs

( dir: str=None )

src/preprocessor.py:96

↓ 1 callersMethod_make_dataloader

( self, files, repeat: bool = True, n_shuffle_shards: int = 1000, n_sh

src/batcher/base.py:86

↓ 1 callersMethod_root_loss

( self, inputs, outputs, attention_mask, **kwargs )

src/embedder/base.py:251

↓ 1 callersMethod_root_loss

( self, inputs, outputs, attention_mask, **kwargs )

src/embedder/dummy.py:87

↓ 1 callersMethod_round_to_precision

( x: torch.tensor, precision: float, )

src/embedder/base.py:151

↓ 1 callersFunction_save_colorbar_to_workdir

( workdir, axsize=(7,5), cmap='inferno', vmax=None, vmin=None, dpi=300, )

src/tools/brainmap.py:149

↓ 1 callersFunction_save_surfaces_to_workdir

( img, workdir, mask_img=None, axsize=(7,5), dpi=300, cmap='inferno', vmax=None,

src/tools/brainmap.py:23

↓ 1 callersMethod_stack_inputs

( self, tensor )

src/embedder/base.py:47

↓ 1 callersMethod_unstack_inputs

( self, tensor, b )

src/embedder/base.py:57

↓ 1 callersMethodadd_decoding_head

( self, num_decoding_classes: int )

src/decoder/autoencoder.py:304

↓ 1 callersMethodadd_decoding_head

( self, num_decoding_classes: int )

src/decoder/gpt.py:80

↓ 1 callersMethodadd_sep_embeds

add sep token embed at the end of each of two sequences contained in inputs

src/embedder/bert.py:47

↓ 1 callersMethodadd_sep_embeds

add sep token embed at the end of each of two sequences contained in inputs

src/embedder/netbert.py:47

↓ 1 callersMethodcompute_loss

( self, model, batch, return_outputs=False, **kwargs )

src/trainer/base.py:40

↓ 1 callersMethodconvert_t_rs_to_position_ids

( self, t_rs: torch.tensor )

src/embedder/base.py:157

↓ 1 callersMethodcount_preproc_bold

(self)

src/preprocessor.py:181

↓ 1 callersFunctiondataprep

Script's main function; additional preprocessing of fmriprep derivatives with Preprocessor from src/

scripts/dataprep/upstream/dataprep.py:13

↓ 1 callersMethoddecode

( self, outputs: torch.tensor )

src/decoder/bert.py:47

↓ 1 callersMethoddecode

( self, pooler_outputs: torch.tensor = None, **kwargs )

src/decoder/autoencoder.py:323

↓ 1 callersMethoddecoding_loss

( self, decoding_logits, labels, **kwargs )

src/embedder/base.py:198

↓ 1 callersMethoddecoding_loss

( self, decoding_logits, labels, **kwargs )

src/embedder/dummy.py:34

↓ 1 callersFunctionderiv_bold_iterator

iterate BOLD derivatives that are to be preprocessed.

scripts/dataprep/upstream/dataprep.py:64

↓ 1 callersFunctiondownstream_chance_accuracies

Script's main function; computes downstream chance decoding accuracy for HCP and MDTB downstream datasets.

scripts/analyses/downstream_chance_accuracies.py:16

↓ 1 callersMethodembed_inputs

( self, inputs: torch.tensor )

src/embedder/base.py:174

↓ 1 callersMethodembed_t_rs

( self, t_rs: torch.tensor )

src/embedder/base.py:180

↓ 1 callersFunctioneval_error_brain_maps

Script's main function; computes brain map for reconstruction error of given upstream model in upstream validation data

scripts/analyses/sfig2_eval_error_brain_maps.py:24

↓ 1 callersFunctionfig_hyperopt

Script's main function; creates Figure 3 of the manuscript.

scripts/analyses/fig3_hyperopt.py:20

↓ 1 callersFunctionfig_upstream_data_overview

Script's main function; creates Figure 2 of the manuscript.

scripts/analyses/fig2_upstream_data_overview.py:19

↓ 1 callersFunctionfig_upstream_performance

Script's main function; creates Figure 4 of the manuscript.

scripts/analyses/fig4_upstream_performance.py:19

↓ 1 callersFunctionget_all_architectures_with_decoding

()

tests/__init__.py:85

↓ 1 callersFunctionget_architectures_with_all_train_styles

()

tests/__init__.py:50

↓ 1 callersFunctionget_argparse

()

scripts/analyses/sfig4_5_downstream_learning_curves.py:200

↓ 1 callersFunctionget_args

()

scripts/analyses/fig5_downstream_performance.py:124

↓ 1 callersFunctionget_args

()

scripts/analyses/sfig3_upstream_performance_pretrained_lms.py:107

↓ 1 callersFunctionget_args

()

scripts/analyses/downstream_chance_accuracies.py:69

↓ 1 callersFunctionget_args

()

scripts/analyses/fig4_upstream_performance.py:108

↓ 1 callersFunctionget_args

()

scripts/analyses/sfig1_training_curve_largest_sequence_BERT.py:72

Functions222 in github.com/athms/learning-from-brains