9394559170

Committed 26 Apr 2024 06:36AM UTC coverage: 94.656% (-0.02%) from 94.674%

Build # 9394559170

Build Type

push

github

Committed by

xmatthias

Commit Message

Loader should be passed as kwarg for clarity

Run Details

20280 of 21425 relevant lines covered (94.66%)

0.95 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

96.55

/freqtrade/freqai/prediction_models/PyTorchTransformerRegressor.py

from typing import Any, Dict, Tuple

import numpy as np
import numpy.typing as npt
import pandas as pd
import torch

from freqtrade.freqai.base_models.BasePyTorchRegressor import BasePyTorchRegressor
from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
from freqtrade.freqai.torch.PyTorchDataConvertor import (DefaultPyTorchDataConvertor,
                                                         PyTorchDataConvertor)
from freqtrade.freqai.torch.PyTorchModelTrainer import PyTorchTransformerTrainer
from freqtrade.freqai.torch.PyTorchTransformerModel import PyTorchTransformerModel


class PyTorchTransformerRegressor(BasePyTorchRegressor):
    """
    This class implements the fit method of IFreqaiModel.
    in the fit method we initialize the model and trainer objects.
    the only requirement from the model is to be aligned to PyTorchRegressor
    predict method that expects the model to predict tensor of type float.
    the trainer defines the training loop.

    parameters are passed via `model_training_parameters` under the freqai
    section in the config file. e.g:
    {
        ...
        "freqai": {
            ...
            "conv_width": 30,  // PyTorchTransformer is based on windowing
            "feature_parameters": {
                ...
                "include_shifted_candles": 0,  // which removes the need for shifted candles
                ...
            },
            "model_training_parameters" : {
                "learning_rate": 3e-4,
                "trainer_kwargs": {
                    "n_steps": 5000,
                    "batch_size": 64,
                    "n_epochs": null
                },
                "model_kwargs": {
                    "hidden_dim": 512,
                    "dropout_percent": 0.2,
                    "n_layer": 1,
                },
            }
        }
    }
    """

    @property
    def data_convertor(self) -> PyTorchDataConvertor:
        return DefaultPyTorchDataConvertor(target_tensor_type=torch.float)

    def __init__(self, **kwargs) -> None:
        super().__init__(**kwargs)
        config = self.freqai_info.get("model_training_parameters", {})
        self.learning_rate: float = config.get("learning_rate",  3e-4)
        self.model_kwargs: Dict[str, Any] = config.get("model_kwargs",  {})
        self.trainer_kwargs: Dict[str, Any] = config.get("trainer_kwargs",  {})

    def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:
        """
        User sets up the training and test data to fit their desired model here
        :param data_dictionary: the dictionary holding all data for train, test,
            labels, weights
        :param dk: The datakitchen object for the current coin/model
        """

        n_features = data_dictionary["train_features"].shape[-1]
        n_labels = data_dictionary["train_labels"].shape[-1]
        model = PyTorchTransformerModel(
            input_dim=n_features,
            output_dim=n_labels,
            time_window=self.window_size,
            **self.model_kwargs
        )
        model.to(self.device)
        optimizer = torch.optim.AdamW(model.parameters(), lr=self.learning_rate)
        criterion = torch.nn.MSELoss()
        # check if continual_learning is activated, and retrieve the model to continue training
        trainer = self.get_init_model(dk.pair)
        if trainer is None:
            trainer = PyTorchTransformerTrainer(
                model=model,
                optimizer=optimizer,
                criterion=criterion,
                device=self.device,
                data_convertor=self.data_convertor,
                window_size=self.window_size,
                tb_logger=self.tb_logger,
                **self.trainer_kwargs,
            )
        trainer.fit(data_dictionary, self.splits)
        return trainer

    def predict(
        self, unfiltered_df: pd.DataFrame, dk: FreqaiDataKitchen, **kwargs
    ) -> Tuple[pd.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)
        """

        dk.find_features(unfiltered_df)
        dk.data_dictionary["prediction_features"], _ = dk.filter_features(
            unfiltered_df, dk.training_features_list, training_filter=False
        )

        dk.data_dictionary["prediction_features"], outliers, _ = dk.feature_pipeline.transform(
            dk.data_dictionary["prediction_features"], outlier_check=True)

        x = self.data_convertor.convert_x(
            dk.data_dictionary["prediction_features"],
            device=self.device
        )
        # if user is asking for multiple predictions, slide the window
        # along the tensor
        x = x.unsqueeze(0)
        # create empty torch tensor
        self.model.model.eval()
        yb = torch.empty(0).to(self.device)
        if x.shape[1] > self.window_size:
            ws = self.window_size
            for i in range(0, x.shape[1] - ws):
                xb = x[:, i:i + ws, :].to(self.device)
                y = self.model.model(xb)
                yb = torch.cat((yb, y), dim=1)
        else:
            yb = self.model.model(x)

        yb = yb.cpu().squeeze(0)
        pred_df = pd.DataFrame(yb.detach().numpy(), columns=dk.label_list)
        pred_df, _, _ = dk.label_pipeline.inverse_transform(pred_df)

        if self.freqai_info.get("DI_threshold", 0) > 0:
            dk.DI_values = dk.feature_pipeline["di"].di_values
        else:
            dk.DI_values = np.zeros(outliers.shape[0])
        dk.do_predict = outliers

        if x.shape[1] > 1:
            zeros_df = pd.DataFrame(np.zeros((x.shape[1] - len(pred_df), len(pred_df.columns))),
                                    columns=pred_df.columns)
            pred_df = pd.concat([zeros_df, pred_df], axis=0, ignore_index=True)
        return (pred_df, dk.do_predict)

1	from typing import Any, Dict, Tuple	1✔
2
3	import numpy as np	1✔
4	import numpy.typing as npt	1✔
5	import pandas as pd	1✔
6	import torch	1✔
7
8	from freqtrade.freqai.base_models.BasePyTorchRegressor import BasePyTorchRegressor	1✔
9	from freqtrade.freqai.data_kitchen import FreqaiDataKitchen	1✔
10	from freqtrade.freqai.torch.PyTorchDataConvertor import (DefaultPyTorchDataConvertor,	1✔
11	PyTorchDataConvertor)
12	from freqtrade.freqai.torch.PyTorchModelTrainer import PyTorchTransformerTrainer	1✔
13	from freqtrade.freqai.torch.PyTorchTransformerModel import PyTorchTransformerModel	1✔
14
15
16	class PyTorchTransformerRegressor(BasePyTorchRegressor):	1✔
17	"""
18	This class implements the fit method of IFreqaiModel.
19	in the fit method we initialize the model and trainer objects.
20	the only requirement from the model is to be aligned to PyTorchRegressor
21	predict method that expects the model to predict tensor of type float.
22	the trainer defines the training loop.
23
24	parameters are passed via `model_training_parameters` under the freqai
25	section in the config file. e.g:
26	{
27	...
28	"freqai": {
29	...
30	"conv_width": 30, // PyTorchTransformer is based on windowing
31	"feature_parameters": {
32	...
33	"include_shifted_candles": 0, // which removes the need for shifted candles
34	...
35	},
36	"model_training_parameters" : {
37	"learning_rate": 3e-4,
38	"trainer_kwargs": {
39	"n_steps": 5000,
40	"batch_size": 64,
41	"n_epochs": null
42	},
43	"model_kwargs": {
44	"hidden_dim": 512,
45	"dropout_percent": 0.2,
46	"n_layer": 1,
47	},
48	}
49	}
50	}
51	"""
52
53	@property	1✔
54	def data_convertor(self) -> PyTorchDataConvertor:	1✔
55	return DefaultPyTorchDataConvertor(target_tensor_type=torch.float)	1✔
56
57	def __init__(self, **kwargs) -> None:	1✔
58	super().__init__(**kwargs)	1✔
59	config = self.freqai_info.get("model_training_parameters", {})	1✔
60	self.learning_rate: float = config.get("learning_rate", 3e-4)	1✔
61	self.model_kwargs: Dict[str, Any] = config.get("model_kwargs", {})	1✔
62	self.trainer_kwargs: Dict[str, Any] = config.get("trainer_kwargs", {})	1✔
63
64	def fit(self, data_dictionary: Dict, dk: FreqaiDataKitchen, **kwargs) -> Any:	1✔
65	"""
66	User sets up the training and test data to fit their desired model here
67	:param data_dictionary: the dictionary holding all data for train, test,
68	labels, weights
69	:param dk: The datakitchen object for the current coin/model
70	"""
71
72	n_features = data_dictionary["train_features"].shape[-1]	1✔
73	n_labels = data_dictionary["train_labels"].shape[-1]	1✔
74	model = PyTorchTransformerModel(	1✔
75	input_dim=n_features,
76	output_dim=n_labels,
77	time_window=self.window_size,
78	**self.model_kwargs
79	)
80	model.to(self.device)	1✔
81	optimizer = torch.optim.AdamW(model.parameters(), lr=self.learning_rate)	1✔
82	criterion = torch.nn.MSELoss()	1✔
83	# check if continual_learning is activated, and retrieve the model to continue training
84	trainer = self.get_init_model(dk.pair)	1✔
85	if trainer is None:	1✔
86	trainer = PyTorchTransformerTrainer(	1✔
87	model=model,
88	optimizer=optimizer,
89	criterion=criterion,
90	device=self.device,
91	data_convertor=self.data_convertor,
92	window_size=self.window_size,
93	tb_logger=self.tb_logger,
94	**self.trainer_kwargs,
95	)
96	trainer.fit(data_dictionary, self.splits)	1✔
97	return trainer	1✔
98
99	def predict(	1✔
100	self, unfiltered_df: pd.DataFrame, dk: FreqaiDataKitchen, **kwargs
101	) -> Tuple[pd.DataFrame, npt.NDArray[np.int_]]:
102	"""
103	Filter the prediction features data and predict with it.
104	:param unfiltered_df: Full dataframe for the current backtest period.
105	:return:
106	:pred_df: dataframe containing the predictions
107	:do_predict: np.array of 1s and 0s to indicate places where freqai needed to remove
108	data (NaNs) or felt uncertain about data (PCA and DI index)
109	"""
110
111	dk.find_features(unfiltered_df)	1✔
112	dk.data_dictionary["prediction_features"], _ = dk.filter_features(	1✔
113	unfiltered_df, dk.training_features_list, training_filter=False
114	)
115
116	dk.data_dictionary["prediction_features"], outliers, _ = dk.feature_pipeline.transform(	1✔
117	dk.data_dictionary["prediction_features"], outlier_check=True)
118
119	x = self.data_convertor.convert_x(	1✔
120	dk.data_dictionary["prediction_features"],
121	device=self.device
122	)
123	# if user is asking for multiple predictions, slide the window
124	# along the tensor
125	x = x.unsqueeze(0)	1✔
126	# create empty torch tensor
127	self.model.model.eval()	1✔
128	yb = torch.empty(0).to(self.device)	1✔
129	if x.shape[1] > self.window_size:	1✔
130	ws = self.window_size	1✔
131	for i in range(0, x.shape[1] - ws):	1✔
132	xb = x[:, i:i + ws, :].to(self.device)	1✔
133	y = self.model.model(xb)	1✔
134	yb = torch.cat((yb, y), dim=1)	1✔
135	else:
136	yb = self.model.model(x)	×
137
138	yb = yb.cpu().squeeze(0)	1✔
139	pred_df = pd.DataFrame(yb.detach().numpy(), columns=dk.label_list)	1✔
140	pred_df, _, _ = dk.label_pipeline.inverse_transform(pred_df)	1✔
141
142	if self.freqai_info.get("DI_threshold", 0) > 0:	1✔
143	dk.DI_values = dk.feature_pipeline["di"].di_values	×
144	else:
145	dk.DI_values = np.zeros(outliers.shape[0])	1✔
146	dk.do_predict = outliers	1✔
147
148	if x.shape[1] > 1:	1✔
149	zeros_df = pd.DataFrame(np.zeros((x.shape[1] - len(pred_df), len(pred_df.columns))),	1✔
150	columns=pred_df.columns)
151	pred_df = pd.concat([zeros_df, pred_df], axis=0, ignore_index=True)	1✔
152	return (pred_df, dk.do_predict)	1✔

freqtrade / freqtrade / 9394559170

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous