import numpy as np
import yaml
import pickle
from scipy.optimize import minimize
import torch
from torch import nn
from torch.utils.data import DataLoader
from torch.cuda.amp import autocast
from mykaggle.util.logger import get_logger
from mykaggle.model.model import get_model
from mykaggle.trainer.base import Mode
from mykaggle.dataloader.dataloader import get_dataloader
from mykaggle.dataloader.dataset import get_dataset
LOGGER = get_logger(__name__)
def predict(model: nn.Module,
df: pd.DataFrame,
dataloader: DataLoader,
batch_size: int,
num_classes: int,
use_amp: bool = True) -> np.ndarray:
if num_classes > 1:
preds = np.zeros((len(df), num_classes), dtype=np.float32)
else:
preds = np.zeros((len(df)), dtype=np.float32)
device = torch.device('cuda')
model.to(device)
model.eval()
from typing import Any, List, Dict, Optional
import time
import pandas as pd
from mykaggle.feature.base import Feature
from mykaggle.util.logger import get_logger
logger = get_logger(__name__)
class FeatureFactory:
@classmethod
def run(cls,
names: List[str],
params: Dict[str, Dict[str, Any]],
base: pd.DataFrame,
others: pd.DataFrame,
train: bool,
use_cache: bool = False,
save_cache: bool = False) -> pd.DataFrame:
'''
与えられた特徴の名前とパラメータのリストから train/test の特徴を作って返す
'''
features = cls.get_features(names, params, train=train)
df = cls.create_feature(features, base, others, use_cache, save_cache,
params)
return df
@classmethod
def get_features(cls, names: List[str], params: Dict[str, Dict[str, Any]],
train: bool) -> List[Feature]:
'''