Files
2020-04-28 07:58:14 +08:00

263 lines
10 KiB
Python

import os
import numpy as np
import pandas as pd
import torch
from tqdm.auto import tqdm
from torch import nn
from torch.nn import functional as F
from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
from test_tube import Experiment, HyperOptArgumentParser
from neural_processes.data.smart_meter import (
collate_fns,
SmartMeterDataSet,
get_smartmeter_df,
)
import torchvision.transforms as transforms
from neural_processes.plot import plot_from_loader_to_tensor, plot_from_loader
from argparse import ArgumentParser
import json
import pytorch_lightning as pl
import math
from matplotlib import pyplot as plt
import torch
import io
import PIL
import optuna
from torchvision.transforms import ToTensor
from neural_processes.data.smart_meter import get_smartmeter_df
from neural_processes.modules import BatchNormSequence, LSTMBlock, NPBlockRelu2d
from neural_processes.utils import ObjectDict
from neural_processes.lightning import PL_Seq2Seq
from ..logger import logger
from ..utils import hparams_power
class TransformerSeq2SeqNet(nn.Module):
def __init__(self, hparams):
super().__init__()
hparams = hparams_power(hparams)
self.hparams = hparams
self._min_std = hparams.min_std
hidden_out_size = self.hparams.hidden_out_size
y_size = self.hparams.input_size - self.hparams.input_size_decoder
x_size = self.hparams.input_size_decoder
# Sometimes input normalisation can be important, an initial batch norm is a nice way to ensure this https://stackoverflow.com/a/46772183/221742
self.x_norm = BatchNormSequence(x_size, affine=False)
self.y_norm = BatchNormSequence(y_size, affine=False)
# TODO embedd both X's the same
if self.hparams.get('use_lstm', False):
self.x_emb = LSTMBlock(x_size, x_size)
self.y_emb = LSTMBlock(y_size, y_size)
self.enc_emb = nn.Linear(self.hparams.input_size, hidden_out_size)
self.dec_emb = nn.Linear(self.hparams.input_size_decoder, hidden_out_size)
encoder_norm = nn.LayerNorm(hidden_out_size)
layer_enc = nn.TransformerEncoderLayer(
d_model=hidden_out_size,
dim_feedforward=hidden_out_size*4,
dropout=self.hparams.attention_dropout,
nhead=self.hparams.nhead,
# activation
)
self.encoder = nn.TransformerEncoder(
layer_enc, num_layers=self.hparams.nlayers, norm=encoder_norm
)
layer_dec = nn.TransformerDecoderLayer(
d_model=hidden_out_size,
dim_feedforward=hidden_out_size*4,
dropout=self.hparams.attention_dropout,
nhead=self.hparams.nhead,
)
decoder_norm = nn.LayerNorm(hidden_out_size)
self.decoder = nn.TransformerDecoder(
layer_dec, num_layers=self.hparams.nlayers, norm=decoder_norm
)
self.mean = NPBlockRelu2d(hidden_out_size, self.hparams.output_size)
self.std = NPBlockRelu2d(hidden_out_size, self.hparams.output_size)
self._use_lvar = False
# self._reset_parameters()
def _reset_parameters(self):
r"""Initiate parameters in the transformer model."""
for p in self.parameters():
if p.dim() > 1:
torch.nn.init.xavier_uniform_(p)
def forward(self, context_x, context_y, target_x, target_y=None, mask_context=True, mask_target=True):
device = next(self.parameters()).device
tgt_key_padding_mask = None
# if target_y is not None and mask_target:
# # Mask nan's
# target_mask = torch.isfinite(target_y)# & (target_y!=self.hparams.nan_value)
# target_y[~target_mask] = 0
# target_y = target_y.detach()
# tgt_key_padding_mask = ~target_mask.any(-1)
src_key_padding_mask = None
# if mask_context:
# # Mask nan's
# context_mask = torch.isfinite(context_y)# & (context_y!=self.hparams.nan_value)
# context_y[~context_mask] = 0
# context_y = context_y.detach()
# src_key_padding_mask = ~context_mask.any(-1)# * float('-inf')
# Norm
context_x = self.x_norm(context_x)
target_x = self.x_norm(target_x)
context_y = self.y_norm(context_y)
# if target_y is not None:
# target_y = self.y_norm(target_y)
# LSTM
if self.hparams.get('use_lstm', False):
context_x = self.x_emb(context_x)
target_x = self.x_emb(target_x)
# Size([B, C, X]) -> Size([B, C, X])
context_y = self.y_emb(context_y)
# Size([B, T, Y]) -> Size([B, T, Y])
# Embed
x = torch.cat([context_x, context_y], -1)
x = self.enc_emb(x)
# Size([B, C, X]) -> Size([B, C, hidden_dim])
target_x = self.dec_emb(target_x)
# Size([B, C, T]) -> Size([B, C, hidden_dim])
x = x.permute(1, 0, 2) # (B,C,hidden_dim) -> (C,B,hidden_dim)
target_x = target_x.permute(1, 0, 2)
# requires (C, B, hidden_dim)
memory = self.encoder(x, src_key_padding_mask=src_key_padding_mask)
# In transformers the memory and target_x need to be the same length. Lets use a permutation invariant agg on the context
# Then expand it, so it's available as we decode, conditional on target_x
# (C, B, emb_dim) -> (B, emb_dim) -> (T, B, emb_dim)
# In transformers the memory and target_x need to be the same length. Lets use a permutation invariant agg on the context
# Then expand it, so it's available as we decode, conditional on target_x
memory_max = memory.max(dim=0, keepdim=True)[0].expand_as(target_x)
memory_mean = memory.mean(dim=0, keepdim=True)[0].expand_as(target_x)
memory_last = memory[-1:, :, :].expand_as(target_x)
memory_all = memory_max + memory_last
if self.hparams.agg == 'max':
memory = memory_max
elif self.hparams.agg == 'last':
memory = memory_last
elif self.hparams.agg == 'all':
memory = memory_all
elif self.hparams.agg == 'mean':
memory = memory_mean
else:
raise Exception(f"hparams.agg should be in ['last', 'max', 'mean', 'all'] not '{self.hparams.agg}''")
outputs = self.decoder(target_x, memory, tgt_key_padding_mask=tgt_key_padding_mask)
# [T, B, emb_dim] -> [B, T, emb_dim]
outputs = outputs.permute(1, 0, 2).contiguous()
# Size([B, T, emb_dim])
mean = self.mean(outputs)
log_sigma = self.std(outputs)
if self._use_lvar:
log_sigma = torch.clamp(
log_sigma, math.log(self._min_std), -math.log(self._min_std)
)
sigma = torch.exp(log_sigma)
else:
sigma = self._min_std + (1 - self._min_std) * F.softplus(log_sigma)
y_dist = torch.distributions.Normal(mean, sigma)
# Loss
loss_mse = loss_p = loss_p_weighted = None
if target_y is not None:
loss_mse = F.mse_loss(mean, target_y, reduction="none")
if self._use_lvar:
loss_p = -log_prob_sigma(target_y, mean, log_sigma)
else:
loss_p = -y_dist.log_prob(target_y).mean(-1)
if self.hparams["context_in_target"]:
loss_p[: context_x.size(1)] /= 100
loss_mse[: context_x.size(1)] /= 100
# Weight loss nearer to prediction time?
weight = (torch.arange(loss_p.shape[1]) + 1).float().to(device)[None, :]
loss_p_weighted = loss_p / torch.sqrt(weight) # We want to weight nearer stuff more
y_pred = y_dist.rsample if self.training else y_dist.loc
return (
y_pred,
dict(loss=loss_p.mean(), loss_p=loss_p.mean(), loss_mse=loss_mse.mean(), loss_p_weighted=loss_p_weighted.mean()),
dict(log_sigma=log_sigma, y_dist=y_dist),
)
class TransformerSeq2Seq_PL(PL_Seq2Seq):
def __init__(self, hparams, MODEL_CLS=TransformerSeq2SeqNet, **kwargs):
super().__init__(hparams, MODEL_CLS=MODEL_CLS, **kwargs)
DEFAULT_ARGS = {
"agg": "max",
"attention_dropout": 0.2,
"hidden_out_size_power": 4,
"hidden_size_power": 5,
"learning_rate": 0.002,
"nhead_power": 3,
"nlayers": 2,
"use_lstm": False
}
@staticmethod
def add_suggest(trial: optuna.Trial, user_attrs={}):
"""
Add hyperparam ranges to an optuna trial and typical user attrs.
Usage:
trial = optuna.trial.FixedTrial(
params={
'hidden_size': 128,
}
)
trial = add_suggest(trial)
trainer = pl.Trainer()
model = LSTM_PL(dict(**trial.params, **trial.user_attrs), dataset_train,
dataset_test, cache_base_path, norm)
trainer.fit(model)
"""
trial.suggest_loguniform("learning_rate", 1e-6, 1e-2)
trial.suggest_uniform("attention_dropout", 0, 0.75)
# we must have nhead<==hidden_size
# so nhead_power.max()<==hidden_size_power.min()
trial.suggest_discrete_uniform("hidden_size_power", 4, 10, 1)
trial.suggest_discrete_uniform("hidden_out_size_power", 4, 9, 1)
trial.suggest_discrete_uniform("nhead_power", 1, 4, 1)
trial.suggest_int("nlayers", 1, 12)
trial.suggest_categorical("use_lstm", [False, True])
trial.suggest_categorical("agg", ['last', 'max', 'mean', 'all'])
user_attrs_default = {
"batch_size": 16,
"grad_clip": 40,
"max_nb_epochs": 200,
"num_workers": 4,
"num_extra_target": 24 * 4,
"vis_i": "670",
"num_context": 24 * 4,
"input_size": 18,
"input_size_decoder": 17,
"context_in_target": False,
"output_size": 1,
"patience": 3,
'min_std': 0.005,
}
[trial.set_user_attr(k, v) for k, v in user_attrs_default.items()]
[trial.set_user_attr(k, v) for k, v in user_attrs.items()]
return trial