Spaces:
Sleeping
Sleeping
File size: 22,439 Bytes
48abd32 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 |
"""
TimesFM Forecasting Module
This module provides a simplified and robust interface for TimesFM forecasting,
handling both basic and covariates-enhanced forecasting with consistent quantile output.
Key Features:
- Single forecast method with optional covariates
- Always returns quantiles (never "maybe")
- Simplified logic: IF covariates -> use covariates, ELSE -> use basic
- Consistent return format: (point_forecast, quantile_forecast)
"""
import numpy as np
import pandas as pd
import logging
from typing import List, Dict, Optional, Tuple, Any, Union
import timesfm
logger = logging.getLogger(__name__)
class Forecaster:
"""
Simplified TimesFM Forecaster with consistent quantile output.
This class provides a single forecast method that handles both basic and
covariates-enhanced forecasting, always returning quantiles.
Example:
>>> forecaster = Forecaster(model)
>>> point_forecast, quantile_forecast = forecaster.forecast(
... inputs=[1,2,3,4,5],
... use_covariates=True,
... dynamic_numerical_covariates={'feature1': [[1,2,3,4,5]]}
... )
"""
def __init__(self, model: timesfm.TimesFm):
"""
Initialize the Forecaster with a loaded TimesFM model.
Args:
model: Initialized TimesFM model instance
"""
self.model = model
self.capabilities = self._detect_capabilities()
logger.info(f"Forecaster initialized with capabilities: {list(self.capabilities.keys())}")
def _detect_capabilities(self) -> Dict[str, bool]:
"""Detect available forecasting capabilities of the model."""
return {
'basic_forecasting': True,
'quantile_forecasting': hasattr(self.model, 'experimental_quantile_forecast'),
'covariates_support': hasattr(self.model, 'forecast_with_covariates')
}
def forecast(
self,
inputs: Union[List[float], List[List[float]]],
freq: Union[int, List[int]] = 0,
dynamic_numerical_covariates: Optional[Dict[str, List[List[float]]]] = None,
dynamic_categorical_covariates: Optional[Dict[str, List[List[str]]]] = None,
static_numerical_covariates: Optional[Dict[str, List[float]]] = None,
static_categorical_covariates: Optional[Dict[str, List[str]]] = None,
use_covariates: bool = False,
xreg_mode: str = "xreg + timesfm",
ridge: float = 0.0,
normalize_xreg_target_per_input: bool = True
) -> Tuple[np.ndarray, np.ndarray]:
"""
Perform TimesFM forecasting with optional covariates support.
This is the main forecasting method that handles both basic and covariates-enhanced
forecasting. Quantiles are always returned regardless of covariates usage.
Args:
inputs: Input time series data
freq: Frequency indicator(s)
dynamic_numerical_covariates: Dynamic numerical covariates (if use_covariates=True)
dynamic_categorical_covariates: Dynamic categorical covariates (if use_covariates=True)
static_numerical_covariates: Static numerical covariates (if use_covariates=True)
static_categorical_covariates: Static categorical covariates (if use_covariates=True)
use_covariates: Whether to use covariates-enhanced forecasting
xreg_mode: Covariate integration mode ("xreg + timesfm" or "timesfm + xreg")
ridge: Ridge regression parameter for covariates
normalize_xreg_target_per_input: Whether to normalize covariates
Returns:
Tuple of (point_forecast, quantile_forecast) - both are always returned
Raises:
ValueError: If covariates are requested but not supported
Exception: If forecasting fails
"""
logger.info(f"Performing TimesFM forecasting (covariates={use_covariates})...")
# Normalize inputs format
if isinstance(inputs[0], (int, float)):
# inputs is a single list of numbers
inputs_norm = [inputs]
else:
# inputs is already a list of lists
inputs_norm = inputs
if isinstance(freq, int):
freq_norm = [freq] * len(inputs_norm)
else:
freq_norm = freq
try:
if use_covariates and any([
dynamic_numerical_covariates, dynamic_categorical_covariates,
static_numerical_covariates, static_categorical_covariates
]):
# Validate covariates support
if not self.capabilities['covariates_support']:
raise ValueError("Model does not support covariates forecasting")
# Validate covariates data structure
self._validate_covariates(
inputs_norm, dynamic_numerical_covariates, dynamic_categorical_covariates,
static_numerical_covariates, static_categorical_covariates
)
logger.info(f"Using covariates-enhanced forecasting (mode: {xreg_mode})...")
logger.info(f"Inputs shape: {[len(x) for x in inputs] if isinstance(inputs[0], list) else len(inputs)}")
logger.info(f"Inputs type: {type(inputs)}")
# Perform covariates forecasting with original mode
covariates_result = self.model.forecast_with_covariates(
inputs=inputs_norm,
dynamic_numerical_covariates=dynamic_numerical_covariates or {},
dynamic_categorical_covariates=dynamic_categorical_covariates or {},
static_numerical_covariates=static_numerical_covariates or {},
static_categorical_covariates=static_categorical_covariates or {},
freq=freq_norm,
xreg_mode=xreg_mode,
ridge=ridge,
normalize_xreg_target_per_input=normalize_xreg_target_per_input
)
# Handle return format from forecast_with_covariates
if isinstance(covariates_result, tuple) and len(covariates_result) == 2:
point_forecast, quantile_forecast = covariates_result
point_forecast = np.array(point_forecast)
quantile_forecast = np.array(quantile_forecast)
logger.info(f"β
Covariates forecasting completed.")
logger.info(f" Point forecast shape: {point_forecast.shape}")
logger.info(f" Quantile forecast shape: {quantile_forecast.shape}")
# Check if we have proper quantiles (multiple quantiles, not just 1)
if quantile_forecast.ndim == 2 and (quantile_forecast.shape[0] == 1 or quantile_forecast.shape[1] == 1):
logger.warning("β οΈ Covariates forecasting returned insufficient quantiles, falling back to basic forecast for quantiles")
# Get quantiles from basic forecast method
_, quantile_forecast = self.model.forecast(inputs=inputs_norm, freq=freq_norm)
quantile_forecast = np.array(quantile_forecast)
logger.info(f"β
Basic forecast quantiles obtained. Shape: {quantile_forecast.shape}")
else:
logger.info("β
Using quantiles from covariates forecasting")
else:
# Fallback: If forecast_with_covariates doesn't return quantiles, get them separately
logger.warning("β οΈ Covariates forecasting didn't return quantiles, getting them separately")
point_forecast = np.array(covariates_result)
_, quantile_forecast = self.model.forecast(inputs=inputs_norm, freq=freq_norm)
quantile_forecast = np.array(quantile_forecast)
else:
logger.info("Using basic forecasting...")
# Perform basic forecasting - this should return (point, quantiles)
point_forecast, quantile_forecast = self.model.forecast(inputs=inputs_norm, freq=freq_norm)
point_forecast = np.array(point_forecast)
quantile_forecast = np.array(quantile_forecast)
logger.info(f"β
Basic forecasting completed.")
return point_forecast, quantile_forecast
except Exception as e:
logger.error(f"β Forecasting failed: {str(e)}")
raise
def _validate_covariates(
self,
inputs: List[List[float]],
dynamic_numerical: Optional[Dict],
dynamic_categorical: Optional[Dict],
static_numerical: Optional[Dict],
static_categorical: Optional[Dict]
) -> None:
"""Validate covariates data structure and compatibility."""
logger.info("Validating covariates data structure...")
# Check that all covariates have the same number of series as inputs
num_series = len(inputs)
for cov_type, cov_data in [
("dynamic_numerical", dynamic_numerical),
("dynamic_categorical", dynamic_categorical),
("static_numerical", static_numerical),
("static_categorical", static_categorical)
]:
if cov_data:
for name, data in cov_data.items():
if isinstance(data[0], (list, np.ndarray)):
# Dynamic covariates
if len(data) != num_series:
raise ValueError(f"Dynamic covariate '{name}' has {len(data)} series, expected {num_series}")
else:
# Static covariates
if len(data) != num_series:
raise ValueError(f"Static covariate '{name}' has {len(data)} values, expected {num_series}")
logger.info("β
Covariates validation passed")
def run_forecast(
forecaster: 'Forecaster',
target_inputs: List[List[float]],
covariates: Optional[Dict[str, Any]] = None,
use_covariates: bool = False,
freq: Union[int, List[int]] = 0
) -> Dict[str, Any]:
"""
Centralized forecasting function that handles both basic and covariates-enhanced forecasting.
This function implements the logic to decide whether to run forecast_with_covariates
or the basic forecast, including fallback mechanisms and proper error handling.
Args:
forecaster: Initialized Forecaster instance
target_inputs: Input time series data
covariates: Dictionary containing covariate data (if use_covariates=True)
use_covariates: Whether to use covariates-enhanced forecasting
freq: Frequency indicator(s)
Returns:
Dictionary containing forecast results with keys:
- 'enhanced_forecast' or 'point_forecast': Main forecast array
- 'quantile_forecast': Quantile forecast array (always present)
- 'method': String indicating the forecasting method used
- 'metadata': Additional forecast metadata
Raises:
Exception: If forecasting fails
"""
logger.info(f"π Running centralized forecast (covariates={use_covariates})...")
try:
results = {}
if use_covariates and covariates:
logger.info("Using covariates-enhanced forecasting...")
# Extract covariate data
dynamic_numerical = covariates.get('dynamic_numerical_covariates')
dynamic_categorical = covariates.get('dynamic_categorical_covariates')
static_numerical = covariates.get('static_numerical_covariates')
static_categorical = covariates.get('static_categorical_covariates')
# Perform covariates forecasting
point_forecast, quantile_forecast = forecaster.forecast(
inputs=target_inputs,
freq=freq,
dynamic_numerical_covariates=dynamic_numerical,
dynamic_categorical_covariates=dynamic_categorical,
static_numerical_covariates=static_numerical,
static_categorical_covariates=static_categorical,
use_covariates=True
)
results['point_forecast'] = point_forecast
results['method'] = 'covariates_enhanced'
else:
logger.info("Using basic forecasting...")
# Perform basic forecasting
point_forecast, quantile_forecast = forecaster.forecast(
inputs=target_inputs,
freq=freq,
use_covariates=False
)
results['point_forecast'] = point_forecast
results['method'] = 'basic_timesfm'
# Check for NaN values before returning
if np.any(np.isnan(point_forecast)):
logger.error(f"β NaN values detected in point_forecast: {np.isnan(point_forecast).sum()} out of {point_forecast.size}")
logger.error(f"Point forecast values: {point_forecast}")
raise ValueError(f"Forecasting produced NaN values in point forecast. This may be due to insufficient data or model issues.")
if np.any(np.isnan(quantile_forecast)):
logger.error(f"β NaN values detected in quantile_forecast: {np.isnan(quantile_forecast).sum()} out of {quantile_forecast.size}")
logger.error(f"Quantile forecast shape: {quantile_forecast.shape}")
raise ValueError(f"Forecasting produced NaN values in quantile forecast. This may be due to insufficient data or model issues.")
# Quantiles are always available
results['quantile_forecast'] = quantile_forecast
logger.info(f"β
Quantile forecast obtained. Shape: {quantile_forecast.shape}")
# Add metadata
results['metadata'] = {
'input_series_count': len(target_inputs),
'forecast_length': results.get('point_forecast').shape[-1],
'covariates_used': use_covariates and covariates is not None,
'quantiles_available': True # Always true now
}
logger.info(f"β
Centralized forecast completed successfully!")
logger.info(f" Method: {results['method']}")
logger.info(f" Forecast shape: {results['metadata']['forecast_length']}")
logger.info(f" Quantiles: Yes (shape: {quantile_forecast.shape})")
logger.info(f" Point forecast range: {np.min(point_forecast):.2f} to {np.max(point_forecast):.2f}")
return results
except Exception as e:
logger.error(f"β Centralized forecasting failed: {str(e)}")
raise
def process_quantile_bands(
quantile_forecast: np.ndarray,
selected_indices: List[int] = None
) -> Dict[str, Any]:
"""
Centralized function to process quantile forecasts into quantile bands for visualization.
This function contains the logic for sorting quantiles and creating the quantile band
dictionary, as used in both the webapp and notebook.
Args:
quantile_forecast: Array of quantile forecasts with shape (horizon, num_quantiles) or (num_quantiles, horizon)
selected_indices: List of quantile indices to use for bands (default: [1, 3, 5, 7, 9])
Returns:
Dictionary of quantile bands ready for visualization with keys:
- 'quantile_band_0_lower', 'quantile_band_0_upper', 'quantile_band_0_label'
- 'quantile_band_1_lower', 'quantile_band_1_upper', 'quantile_band_1_label'
- etc.
"""
logger.info("π Processing quantile bands...")
logger.info(f"Input quantile_forecast type: {type(quantile_forecast)}")
logger.info(f"Input quantile_forecast shape: {quantile_forecast.shape if hasattr(quantile_forecast, 'shape') else 'N/A'}")
# logger.info(f"!!!!!!!!!!!!! selected_indices: {selected_indices}")
# logger.info(f"!!!!!!!!!!!!! quantile_forecast.shape: {quantile_forecast.shape}")
if quantile_forecast is None:
logger.warning("No quantile forecast provided")
return {}
try:
# logger.info(f"!!!!!!!!!!!!! selected_indices: {selected_indices}")
# logger.info(f"!!!!!!!!!!!!! quantile_forecast.shape: {quantile_forecast.shape}")
# Handle quantile indices - only use default if explicitly None (not empty list)
if selected_indices is None:
# This means no quantile selection was made, use default
selected_indices = [1, 3, 5, 7, 9] # Q10, Q30, Q50, Q70, Q90
elif selected_indices == []:
# This means user explicitly selected no quantiles, return empty
logger.info("No quantiles selected by user - returning empty quantile bands")
return {}
# Handle different array dimensions
if quantile_forecast.ndim == 3:
# Shape is (1, horizon, num_quantiles) - squeeze out first dimension
q_mat = quantile_forecast.squeeze(0)
logger.info(f"3D array detected, squeezed to shape: {q_mat.shape}")
elif quantile_forecast.ndim == 1:
# Shape is (horizon,) - reshape to (1, horizon)
q_mat = quantile_forecast.reshape(1, -1)
logger.info(f"1D array detected, reshaped to: {q_mat.shape}")
else:
# Shape is 2D - determine if we need to transpose
# For quantiles, we expect (horizon, num_quantiles) format
# If we have more horizon than quantiles, it's likely (horizon, num_quantiles) and should be kept as-is
if quantile_forecast.shape[0] > quantile_forecast.shape[1]:
# Shape is (horizon, num_quantiles) - keep as is
q_mat = quantile_forecast
logger.info(f"2D array kept as is (horizon, quantiles): {q_mat.shape}")
else:
# Shape is (num_quantiles, horizon) - transpose to (horizon, num_quantiles)
q_mat = quantile_forecast.T
logger.info(f"2D array transposed from {quantile_forecast.shape} to {q_mat.shape}")
horizon_len, num_quantiles = q_mat.shape
logger.info(f"π Available quantiles: {num_quantiles} (indices 0-{num_quantiles-1})")
logger.info(f"π Note: Index 0 is legacy mean forecast, using indices 1-{num_quantiles-1} for actual quantiles")
# Check if we have enough quantiles for band creation (need at least 3 total: 0=legacy, 1=Q10, 2=Q20)
if num_quantiles < 3:
logger.warning(f"Not enough quantiles for band creation. Have {num_quantiles}, need at least 3")
return {}
# Filter selected indices to valid range (skip index 0)
valid_indices = [idx for idx in selected_indices if 1 <= idx < num_quantiles] # Skip index 0
if not valid_indices:
logger.warning("No valid quantile indices selected (after skipping legacy index 0)")
return {}
# logger.info(f"!!!!!!!!!!!!! valid_indices: {valid_indices}")
# Sort quantiles by their median magnitude to ensure proper ordering
quantile_medians = np.median(q_mat, axis=0)
sorted_indices = np.argsort(quantile_medians)
# Create quantile bands from selected indices
quantile_bands = {}
band_count = 0
for i in range(len(valid_indices) - 1):
lower_idx = valid_indices[i]
upper_idx = valid_indices[i + 1]
# Get the sorted indices for these quantiles
lower_sorted_idx = sorted_indices[lower_idx]
upper_sorted_idx = sorted_indices[upper_idx]
# Extract quantile values
lower_quantile = q_mat[:, lower_sorted_idx]
upper_quantile = q_mat[:, upper_sorted_idx]
# Create band labels
lower_pct = idx_to_percent(lower_idx, num_quantiles)
upper_pct = idx_to_percent(upper_idx, num_quantiles)
band_label = f"Q{lower_pct:02d}βQ{upper_pct:02d}"
# Store band data
quantile_bands[f'quantile_band_{band_count}_lower'] = lower_quantile.tolist()
quantile_bands[f'quantile_band_{band_count}_upper'] = upper_quantile.tolist()
quantile_bands[f'quantile_band_{band_count}_label'] = band_label
logger.info(f" Band {band_count}: {band_label} - Lower: {len(lower_quantile)}, Upper: {len(upper_quantile)}")
band_count += 1
logger.info(f"β
Created {band_count} quantile bands from indices: {valid_indices}")
for i in range(band_count):
label = quantile_bands[f'quantile_band_{i}_label']
logger.info(f" Band {i}: {label}")
return quantile_bands
except Exception as e:
logger.error(f"β Quantile band processing failed: {str(e)}")
raise
def idx_to_percent(idx: int, num_quantiles: int) -> int:
"""
Convert quantile index to percentage for labeling.
Note: Index 0 is legacy mean forecast and should be skipped.
Actual quantiles start at index 1: 1->Q10, 2->Q20, ..., 9->Q90
Args:
idx: Quantile index (1-based for actual quantiles, 0 is legacy)
num_quantiles: Total number of quantiles (including legacy index 0)
Returns:
Percentage value (e.g., 10 for Q10, 90 for Q90)
"""
if num_quantiles == 10:
# Special case for 10 quantiles: 1->Q10, 2->Q20, ..., 9->Q90
# Index 0 is legacy mean, so actual quantiles start at index 1
return idx * 10
else:
# General case: distribute evenly, accounting for skipped index 0
# If we have 10 total quantiles (0-9), actual quantiles are 1-9
actual_quantiles = num_quantiles - 1 # Subtract 1 for legacy index 0
return int(100 * idx / actual_quantiles) |