Functions — github.com/google-research/timesfm

hub / github.com/google-research/timesfm / functions

Functions402 in github.com/google-research/timesfm

⨍ Functions 402◇ Types & classes 88

↓ 1 callersFunctiontry_gc

()

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:48

↓ 1 callersFunctionwrite_csv_output

Write forecast results to CSV.

timesfm-forecasting/scripts/forecast_csv.py:144

↓ 1 callersFunctionwrite_json_output

Write forecast results to JSON.

timesfm-forecasting/scripts/forecast_csv.py:187

↓ 1 callersFunctionzero_pad_time_series

If time_series length is less than min_length, front pad it with zeros.

v1/experiments/baselines/timegpt_pipeline.py:46

Method__call__

( self, inputs: Float[Array, "b n p"], masks: Bool[Array, "b n p"], decode_cache: util.DecodeC

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:126

Method__call__

(self, x: Float[Array, "b ... i"])

src/timesfm/flax/dense.py:66

Method__call__

(self, x: Float[Array, "b ... i"])

src/timesfm/flax/dense.py:100

Method__call__

Generates a JTensor of sinusoids with different frequencies.

src/timesfm/flax/transformer.py:80

Method__call__

(self, x: Float[Array, "b ... d"])

src/timesfm/flax/transformer.py:128

Method__call__

Applies multi-head dot product attention on the input data.

src/timesfm/flax/transformer.py:207

Method__call__

( self, input_embeddings: Float[Array, "b n d"], patch_mask: Bool[Array, "b n"], decode_cache:

src/timesfm/flax/transformer.py:338

Method__call__

(self, inputs: Float[Array, "b ... d"])

src/timesfm/flax/normalization.py:46

Method__call__

(self, inputs: Float[Array, "b ... d"])

src/timesfm/flax/normalization.py:65

Method__call__

(self, inputs: JTensor)

v1/src/timesfm/patched_decoder.py:146

Method__call__

PatchTST call. Args: inputs: A NestedMap containing (1) input_ts: input sequence of shape [B, T] where T must be multiple of patch_

v1/src/timesfm/patched_decoder.py:365

Method__get__

(self, obj, objtype=None)

v1/src/adapter/lora_layers.py:70

Method__get__

(self, obj, objtype=None)

v1/src/adapter/dora_layers.py:79

Method__getitem__

(self, i: int)

timesfm-forecasting/examples/finetuning/finetune_lora.py:101

Method__getitem__

(self, i: int)

timesfm-forecasting/examples/finetuning/finetune_lora.py:135

Method__getitem__

( self, index: int )

v1/src/finetuning/finetuning_example.py:91

Method__getitem__

(self, k)

v1/src/adapter/lora_layers.py:56

Method__getitem__

(self, k)

v1/src/adapter/dora_layers.py:65

Method__init__

(self, config: configs.RandomFourierFeaturesConfig)

src/timesfm/torch/dense.py:62

Method__init__

( self, embedding_dims: int, min_timescale: float = 1.0, max_timescale: float = 10000.0, )

src/timesfm/torch/transformer.py:59

Method__init__

( self, num_heads: int, in_features: int, *, use_per_dim_scale: bool = True, use_rotar

src/timesfm/torch/transformer.py:172

Method__init__

(self, config: configs.TransformerConfig)

src/timesfm/torch/transformer.py:310

Method__init__

( self, num_features: int, *, epsilon: float = 1e-6, )

src/timesfm/torch/normalization.py:24

Method__init__

Initializes with the exogenous covariate inputs. Here we use model fitting language to refer to the context as 'train' and the horizon as 'te

src/timesfm/utils/xreg_lib.py:97

Method__init__

(self)

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:96

Method__init__

(self)

src/timesfm/timesfm_2p5/timesfm_2p5_torch.py:41

Method__init__

(self, config: ResidualBlockConfig, *, rngs=nnx.Rngs(42))

src/timesfm/flax/dense.py:37

Method__init__

(self, config: RandomFourierFeaturesConfig, *, rngs=nnx.Rngs(42))

src/timesfm/flax/dense.py:77

Method__init__

( self, embedding_dims: int, min_timescale: int = 1, max_timescale: int = 10000, )

src/timesfm/flax/transformer.py:70

Method__init__

(self, num_dims: int, *, rngs=nnx.Rngs(42))

src/timesfm/flax/transformer.py:123

Method__init__

( self, num_heads: int, in_features: int, *, use_per_dim_scale: bool = True, use_rotar

src/timesfm/flax/transformer.py:137

Method__init__

(self, config: TransformerConfig, *, rngs=nnx.Rngs(42))

src/timesfm/flax/transformer.py:294

Method__init__

( self, num_features: int, *, epsilon: float = 1e-6, rngs=nnx.Rngs(42), )

src/timesfm/flax/normalization.py:34

Method__init__

(self, num_features: int, *, epsilon: float = 1e-6, rngs=nnx.Rngs(42))

src/timesfm/flax/normalization.py:58

Method__init__

( self, series_list: list[np.ndarray], context_len: int, horizon_len: int,

timesfm-forecasting/examples/finetuning/finetune_lora.py:68

Method__init__

( self, series_list: list[np.ndarray], context_len: int, horizon_len: int,

timesfm-forecasting/examples/finetuning/finetune_lora.py:118

Method__init__

( self, api_key: str | None = None, base_url: Optional[str] = None, max_retries: int =

v1/experiments/baselines/timegpt_pipeline.py:159

Method__init__

( self, dataset: str, quantiles: List[float] = list(np.arange(1, 10) / 10.0), results_

v1/experiments/extended_benchmarks/utils.py:61

Method__init__

Initializes the TimesFM forecast API. Args: hparams: Hyperparameters of the model. checkpoint: Checkpoint to load. Notice `checkpoint

v1/src/timesfm/timesfm_base.py:221

Method__init__

Initializes with the exogenous covariate inputs. Here we use model fitting language to refer to the context as 'train' and the horizon as 'te

v1/src/timesfm/xreg_lib.py:81

Method__init__

Initialize objects. Args: data_path: path to csv file datetime_col: column name for datetime col num_cov_cols: list of numerica

v1/src/timesfm/data_loader.py:30

Method__init__

Init function. Args: datetimes: pandas DatetimeIndex (lowest granularity supported is min) normalized: whether to normalize features

v1/src/timesfm/time_features.py:115

Method__init__

( self, input_dims, hidden_dims, output_dims, )

v1/src/timesfm/pytorch_patched_decoder.py:242

Method__init__

( self, hidden_size: int, intermediate_size: int, )

v1/src/timesfm/pytorch_patched_decoder.py:300

Method__init__

( self, hidden_size: int, num_heads: int, num_kv_heads: int, head_dim: int, )

v1/src/timesfm/pytorch_patched_decoder.py:323

Method__init__

( self, hidden_size: int, intermediate_size: int, num_heads: int, num_kv_heads:

v1/src/timesfm/pytorch_patched_decoder.py:421

Method__init__

( self, hidden_size: int, intermediate_size: int, num_heads: int, num_kv_heads:

v1/src/timesfm/pytorch_patched_decoder.py:471

Method__init__

( self, embedding_dims: int, min_timescale: int = 1, max_timescale: int = 10_000, )

v1/src/timesfm/pytorch_patched_decoder.py:529

Method__init__

(self, config: TimesFMConfig)

v1/src/timesfm/pytorch_patched_decoder.py:577

Method__init__

Initialize dataset. Args: series: Time series data context_length: Number of past timesteps to use as input

v1/src/finetuning/finetuning_example.py:54

Method__init__

(self, project: str, config: Dict[str, Any], rank: int = 0)

v1/src/finetuning/finetuning_torch.py:55

Method__init__

( self, world_size: int, rank: int, master_addr: str = "localhost", master_port:

v1/src/finetuning/finetuning_torch.py:89

Method__init__

( self, model: nn.Module, config: FinetuningConfig, rank: int = 0, loss_fn: Opti

v1/src/finetuning/finetuning_torch.py:173

Method__init__

(self, module)

v1/src/adapter/lora_layers.py:24

Method__init__

(self, module)

v1/src/adapter/dora_layers.py:24

Method__len__

(self)

timesfm-forecasting/examples/finetuning/finetune_lora.py:98

Method__len__

(self)

timesfm-forecasting/examples/finetuning/finetune_lora.py:132

Method__len__

(self)

v1/src/finetuning/finetuning_example.py:88

Method__post_init__

(self)

v1/src/timesfm/timesfm_jax.py:85

Method__post_init__

(self)

v1/src/timesfm/timesfm_torch.py:33

Method__str__

(self)

timesfm-forecasting/scripts/check_system.py:85

Method_ar_decode

(module, carry, unused_iter)

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:190

Method_compiled_decode

(horizon, inputs, masks)

src/timesfm/timesfm_2p5/timesfm_2p5_torch.py:421

Function_create_quantiles

Returns the quantiles for forecasting.

v1/src/timesfm/patched_decoder.py:206

Method_decode

(inputs)

v1/src/timesfm/timesfm_jax.py:182

Function_distance_to_day

(index)

v1/src/timesfm/time_features.py:48

Function_dot_product_attention

Computes dot-product attention given query, key, and value.

src/timesfm/torch/transformer.py:114

Method_forecast

Forecasts on a list of time series. Args: inputs: list of time series forecast contexts. Each context time series should be in a fo

v1/src/timesfm/timesfm_jax.py:239

Method_forecast

Forecasts on a list of time series. Args: inputs: list of time series forecast contexts. Each context time series should be in a fo

v1/src/timesfm/timesfm_torch.py:72

Method_from_pretrained

Loads a PyTorch safetensors TimesFM model from a local path or the Hugging Face Hub. This method is the backend for the `from_pretrained` cla

src/timesfm/timesfm_2p5/timesfm_2p5_torch.py:305

Function_reshape

(input_tensor)

v1/peft/finetune.py:322

Function_torch_dot_product_attention

Performs the exact same (unscaled) attention as the above function, but using the fast and fused F.scaled_dot_product_attention kernel.

src/timesfm/torch/transformer.py:132

Functionapply_mask_to_logits

Applies a floating-point mask to a set of logits. Args: logits: A torch.Tensor of logit values. mask: A torch.Tensor (float32) of mask

v1/src/timesfm/pytorch_patched_decoder.py:155

Methodclose

Clean up any resources used by the logger.

v1/src/finetuning/finetuning_torch.py:41

Methodcompile

( self, context: int, horizon: int, per_core_batch_size: int = 1, )

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:237

Methodcompile

( self, forecast_config: configs.ForecastConfig, dryrun: bool = True, **kwargs )

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:506

Methodcompiled_decode_kernel

(model, horizon, inputs, masks)

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:270

Methodcompiled_decode_kernel

(fc, horizon, inputs, masks)

src/timesfm/timesfm_2p5/timesfm_2p5_flax.py:555

Methodcompute_loss

(self, prediction_output: NestedMap, input_batch: NestedMap)

v1/src/timesfm/patched_decoder.py:532

Methodcompute_predictions

(self, input_batch: NestedMap)

v1/src/timesfm/patched_decoder.py:496

Methodcross_validation

( self, df: pd.DataFrame, h: int, freq: str, n_windows: int = 1, step_size

v1/experiments/baselines/timegpt_pipeline.py:98

Methoddecode

Auto-regressive decoding without caching. Args: input_ts: input time-series and paddings. Time-series shape B x C. paddings: padding

v1/src/timesfm/pytorch_patched_decoder.py:712

Functioneval_step

(states, prng_key, inputs)

v1/peft/finetune.py:308

Methodevaluate_models

(self, models: List[str])

v1/experiments/extended_benchmarks/utils.py:213

Functionfinetune

( *, model_name: Annotated[ str, typer.Option(help="Specify the name of the huggingface model.

v1/peft/finetune.py:63

Methodfit

Fits a linear model for in-context regression. Args: ridge: A non-negative value for specifying the ridge regression penalty. If 0

src/timesfm/utils/xreg_lib.py:414

Methodfit

Fits a linear model for in-context regression. Args: ridge: A non-negative value for specifying the ridge regression penalty. If 0

v1/src/timesfm/xreg_lib.py:392

Methodflip_quantile_fn

(x)

src/timesfm/timesfm_2p5/timesfm_2p5_torch.py:453

Methodforecast_with_covariates

Forecasts on a list of time series with covariates. To optimize inference speed, avoid string valued categorical covariates. Args: inp

src/timesfm/timesfm_2p5/timesfm_2p5_base.py:198

Methodforecast_with_covariates

Forecasts on a list of time series with covariates. To optimize inference speed, avoid string valued categorical covariates. Args: inp

v1/src/timesfm/timesfm_base.py:429

Methodforward

(self, x: torch.Tensor)

src/timesfm/torch/dense.py:53

Methodforward

(self, x: torch.Tensor)

src/timesfm/torch/dense.py:84

Methodforward

Generates a JTensor of sinusoids with different frequencies.

src/timesfm/torch/transformer.py:70

Methodforward

(self, x: torch.Tensor)

src/timesfm/torch/transformer.py:162

Methodforward

( self, inputs_q: torch.Tensor, *, decode_cache: DecodeCache | None = None, patch_mask: to

src/timesfm/torch/transformer.py:224

Methodforward

( self, input_embeddings: torch.Tensor, patch_mask: torch.Tensor, decode_cache: DecodeCache |

src/timesfm/torch/transformer.py:354

← previous next →201–300 of 402, ranked by callers