mirror of
https://github.com/microsoft/qlib.git
synced 2026-07-09 14:00:55 +08:00
cf35562e84
* Merge data selection to main * Update trainer for reweighter * Typos fixed. * update data selection interface * successfully run exp after refactor some interface * data selection share handler & trainer * fix meta model time series bug * fix online workflow set_uri bug * fix set_uri bug * updawte ds docs and delay trainer bug * docs * resume reweighter * add reweighting result * fix qlib model import * make recorder more friendly * fix experiment workflow bug * commit for merging master incase of conflictions * Successful run DDG-DA with a single command * remove unused code * asdd more docs * Update README.md * Update & fix some bugs. * Update configuration & remove debug functions * Update README.md * Modfify horizon from code rather than yaml * Update performance in README.md * fix part comments * Remove unfinished TCTS. * Fix some details. * Update meta docs * Update README.md of the benchmarks_dynamic * Update README.md files * Add README.md to the rolling_benchmark baseline. * Refine the docs and link * Rename README.md in benchmarks_dynamic. * Remove comments. * auto download data Co-authored-by: wendili-cs <wendili.academic@qq.com> Co-authored-by: demon143 <785696300@qq.com>
69 lines
2.7 KiB
Python
69 lines
2.7 KiB
Python
# Copyright (c) Microsoft Corporation.
|
|
# Licensed under the MIT License.
|
|
|
|
import pandas as pd
|
|
import numpy as np
|
|
import torch
|
|
from torch import nn
|
|
|
|
from .utils import preds_to_weight_with_clamp, SingleMetaBase
|
|
|
|
|
|
class TimeWeightMeta(SingleMetaBase):
|
|
def __init__(self, hist_step_n, clip_weight=None, clip_method="clamp"):
|
|
# clip_method includes "tanh" or "clamp"
|
|
super().__init__(hist_step_n, clip_weight, clip_method)
|
|
self.linear = nn.Linear(hist_step_n, 1)
|
|
self.k = nn.Parameter(torch.Tensor([8.0]))
|
|
|
|
def forward(self, time_perf, time_belong=None, return_preds=False):
|
|
hist_step_n = self.linear.in_features
|
|
# NOTE: the reshape order is very important
|
|
time_perf = time_perf.reshape(hist_step_n, time_perf.shape[0] // hist_step_n, *time_perf.shape[1:])
|
|
time_perf = torch.mean(time_perf, dim=1, keepdim=False)
|
|
|
|
preds = []
|
|
for i in range(time_perf.shape[1]):
|
|
preds.append(self.linear(time_perf[:, i]))
|
|
preds = torch.cat(preds)
|
|
preds = preds - torch.mean(preds) # avoid using future information
|
|
preds = preds * self.k
|
|
if return_preds:
|
|
if time_belong is None:
|
|
return preds
|
|
else:
|
|
return time_belong @ preds
|
|
else:
|
|
weights = preds_to_weight_with_clamp(preds, self.clip_weight, self.clip_method)
|
|
if time_belong is None:
|
|
return weights
|
|
else:
|
|
return time_belong @ weights
|
|
|
|
|
|
class PredNet(nn.Module):
|
|
def __init__(self, step, hist_step_n, clip_weight=None, clip_method="tanh"):
|
|
super().__init__()
|
|
self.step = step
|
|
self.twm = TimeWeightMeta(hist_step_n=hist_step_n, clip_weight=clip_weight, clip_method=clip_method)
|
|
self.init_paramters(hist_step_n)
|
|
|
|
def get_sample_weights(self, X, time_perf, time_belong, ignore_weight=False):
|
|
weights = torch.from_numpy(np.ones(X.shape[0])).float().to(X.device)
|
|
if not ignore_weight:
|
|
if time_perf is not None:
|
|
weights_t = self.twm(time_perf, time_belong)
|
|
weights = weights * weights_t
|
|
return weights
|
|
|
|
def forward(self, X, y, time_perf, time_belong, X_test, ignore_weight=False):
|
|
"""Please refer to the docs of MetaTaskDS for the description of the variables"""
|
|
weights = self.get_sample_weights(X, time_perf, time_belong, ignore_weight=ignore_weight)
|
|
X_w = X.T * weights.view(1, -1)
|
|
theta = torch.inverse(X_w @ X) @ X_w @ y
|
|
return X_test @ theta, weights
|
|
|
|
def init_paramters(self, hist_step_n):
|
|
self.twm.linear.weight.data = 1.0 / hist_step_n + self.twm.linear.weight.data * 0.01
|
|
self.twm.linear.bias.data.fill_(0.0)
|