freqtrade_origin/freqtrade/freqai/prediction_models/PyTorchClassifierMultiTarget.py

164 lines
6.9 KiB
Python
Raw Normal View History

2023-03-05 14:59:24 +00:00
import logging
2023-03-12 14:16:04 +00:00
from typing import Any, Dict, List, Optional, Tuple
2023-03-05 14:59:24 +00:00
import numpy as np
2023-03-08 14:03:36 +00:00
import numpy.typing as npt
2023-03-05 14:59:24 +00:00
import pandas as pd
import torch
from pandas import DataFrame
from torch.nn import functional as F
2023-03-08 14:11:51 +00:00
from freqtrade.exceptions import OperationalException
from freqtrade.freqai.base_models.BasePyTorchModel import BasePyTorchModel
from freqtrade.freqai.base_models.PyTorchModelTrainer import PyTorchModelTrainer
2023-03-08 14:03:36 +00:00
from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
from freqtrade.freqai.prediction_models.PyTorchMLPModel import PyTorchMLPModel
2023-03-05 14:59:24 +00:00
logger = logging.getLogger(__name__)
class PyTorchClassifierMultiTarget(BasePyTorchModel):
2023-03-09 09:14:54 +00:00
"""
A PyTorch implementation of a multi-target classifier.
"""
2023-03-05 14:59:24 +00:00
def __init__(self, **kwargs):
2023-03-09 09:14:54 +00:00
"""
int: The number of nodes in the hidden layer of the neural network.
int: The maximum number of iterations to run during training.
int: The batch size to use during training.
float: The learning rate to use during training.
int: The number of training iterations between each evaluation.
dict: A dictionary mapping class names to their corresponding indices.
dict: A dictionary mapping indices to their corresponding class names.
"""
2023-03-09 12:55:52 +00:00
2023-03-05 14:59:24 +00:00
super().__init__(**kwargs)
2023-03-12 12:31:08 +00:00
trainer_kwargs = self.freqai_info.get("trainer_kwargs", {})
self.n_hidden: int = trainer_kwargs.get("n_hidden", 1024)
self.max_iters: int = trainer_kwargs.get("max_iters", 100)
self.batch_size: int = trainer_kwargs.get("batch_size", 64)
self.learning_rate: float = trainer_kwargs.get("learning_rate", 3e-4)
self.max_n_eval_batches: Optional[int] = trainer_kwargs.get("max_n_eval_batches", None)
self.model_kwargs: Dict = trainer_kwargs.get("model_kwargs", {})
self.class_name_to_index = None
self.index_to_class_name = None
2023-03-05 14:59:24 +00:00
def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
"""
User sets up the training and test data to fit their desired model here
2023-03-09 09:14:54 +00:00
:param data_dictionary: the dictionary constructed by DataHandler to hold
2023-03-05 14:59:24 +00:00
all the training and test data/labels.
2023-03-09 09:14:54 +00:00
:raises ValueError: If self.class_names is not defined in the parent class.
2023-03-05 14:59:24 +00:00
"""
2023-03-09 12:55:52 +00:00
if not hasattr(self, "class_names"):
raise ValueError(
"Missing attribute: self.class_names "
"set self.freqai.class_names = [\"class a\", \"class b\", \"class c\"] "
"inside IStrategy.set_freqai_targets method."
)
self.init_class_names_to_index_mapping(self.class_names)
self.encode_classes_name(data_dictionary, dk)
2023-03-09 11:29:11 +00:00
n_features = data_dictionary["train_features"].shape[-1]
model = PyTorchMLPModel(
2023-03-05 14:59:24 +00:00
input_dim=n_features,
2023-03-12 12:31:08 +00:00
output_dim=len(self.class_names),
**self.model_kwargs
2023-03-05 14:59:24 +00:00
)
model.to(self.device)
optimizer = torch.optim.AdamW(model.parameters(), lr=self.learning_rate)
criterion = torch.nn.CrossEntropyLoss()
2023-03-05 14:59:24 +00:00
init_model = self.get_init_model(dk.pair)
trainer = PyTorchModelTrainer(
model=model,
optimizer=optimizer,
criterion=criterion,
model_meta_data={"class_names": self.class_names},
device=self.device,
batch_size=self.batch_size,
max_iters=self.max_iters,
max_n_eval_batches=self.max_n_eval_batches,
init_model=init_model
)
trainer.fit(data_dictionary)
2023-03-05 14:59:24 +00:00
return trainer
def predict(
self, unfiltered_df: DataFrame, dk: FreqaiDataKitchen, **kwargs
) -> Tuple[DataFrame, npt.NDArray[np.int_]]:
"""
Filter the prediction features data and predict with it.
:param unfiltered_df: Full dataframe for the current backtest period.
:return:
:pred_df: dataframe containing the predictions
:do_predict: np.array of 1s and 0s to indicate places where freqai needed to remove
data (NaNs) or felt uncertain about data (PCA and DI index)
2023-03-09 12:55:52 +00:00
:raises ValueError: if 'class_name' doesn't exist in model meta_data.
2023-03-05 14:59:24 +00:00
"""
2023-03-09 12:55:52 +00:00
class_names = self.model.model_meta_data.get("class_names", None)
if not class_names:
raise ValueError(
"Missing class names. "
"self.model.model_meta_data[\"class_names\"] is None."
)
self.init_class_names_to_index_mapping(class_names)
2023-03-05 14:59:24 +00:00
dk.find_features(unfiltered_df)
filtered_df, _ = dk.filter_features(
unfiltered_df, dk.training_features_list, training_filter=False
)
filtered_df = dk.normalize_data_from_metadata(filtered_df)
dk.data_dictionary["prediction_features"] = filtered_df
self.data_cleaning_predict(dk)
dk.data_dictionary["prediction_features"] = torch.tensor(
dk.data_dictionary["prediction_features"].values
).float().to(self.device)
2023-03-05 14:59:24 +00:00
logits = self.model.model(dk.data_dictionary["prediction_features"])
2023-03-05 14:59:24 +00:00
probs = F.softmax(logits, dim=-1)
predicted_classes = torch.argmax(probs, dim=-1)
predicted_classes_str = self.decode_classes_name(predicted_classes)
pred_df_prob = DataFrame(probs.detach().numpy(), columns=class_names)
pred_df = DataFrame(predicted_classes_str, columns=[dk.label_list[0]])
2023-03-05 14:59:24 +00:00
pred_df = pd.concat([pred_df, pred_df_prob], axis=1)
return (pred_df, dk.do_predict)
def encode_classes_name(self, data_dictionary: Dict[str, pd.DataFrame], dk: FreqaiDataKitchen):
"""
encode class name str -> int
assuming first column of *_labels data frame to contain class names
"""
2023-03-09 12:55:52 +00:00
target_column_name = dk.label_list[0]
for split in ["train", "test"]:
label_df = data_dictionary[f"{split}_labels"]
self.assert_valid_class_names(label_df[target_column_name])
label_df[target_column_name] = list(
map(lambda x: self.class_name_to_index[x], label_df[target_column_name])
)
def assert_valid_class_names(self, labels: pd.Series):
non_defined_labels = set(labels) - set(self.class_names)
if len(non_defined_labels) != 0:
raise OperationalException(
2023-03-08 13:38:22 +00:00
f"Found non defined labels: {non_defined_labels}, ",
f"expecting labels: {self.class_names}"
)
2023-03-08 14:16:49 +00:00
def decode_classes_name(self, classes: torch.Tensor) -> List[str]:
2023-03-08 13:38:22 +00:00
"""
decode class name int -> str
"""
2023-03-09 12:55:52 +00:00
2023-03-08 13:38:22 +00:00
return list(map(lambda x: self.index_to_class_name[x.item()], classes))
def init_class_names_to_index_mapping(self, class_names):
self.class_name_to_index = {s: i for i, s in enumerate(class_names)}
self.index_to_class_name = {i: s for i, s in enumerate(class_names)}
logger.info(f"class_name_to_index: {self.class_name_to_index}")