mirror of
https://github.com/freqtrade/freqtrade.git
synced 2025-12-03 02:23:05 +00:00
chore: improve naming, don't duplicate column list
This commit is contained in:
@@ -15,7 +15,7 @@ from freqtrade.exceptions import DependencyException
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
ADDED_COLUMNS = [
|
||||
ORDERFLOW_ADDED_COLUMNS = [
|
||||
"trades",
|
||||
"orderflow",
|
||||
"imbalances",
|
||||
@@ -36,7 +36,7 @@ def _init_dataframe_with_trades_columns(dataframe: pd.DataFrame):
|
||||
:param dataframe: Dataframe to populate
|
||||
"""
|
||||
# Initialize columns with appropriate dtypes
|
||||
for column in ADDED_COLUMNS:
|
||||
for column in ORDERFLOW_ADDED_COLUMNS:
|
||||
dataframe[column] = np.nan
|
||||
|
||||
# Set columns to object type
|
||||
@@ -115,7 +115,7 @@ def populate_dataframe_with_trades(
|
||||
cache_idx = cached_grouped_trades.index[
|
||||
cached_grouped_trades["date"] == candle_start
|
||||
][0]
|
||||
for col in ADDED_COLUMNS:
|
||||
for col in ORDERFLOW_ADDED_COLUMNS:
|
||||
dataframe.at[index, col] = cached_grouped_trades.at[cache_idx, col]
|
||||
continue
|
||||
|
||||
|
||||
@@ -4,7 +4,10 @@ import pytest
|
||||
|
||||
from freqtrade.constants import DEFAULT_TRADES_COLUMNS
|
||||
from freqtrade.data.converter import populate_dataframe_with_trades
|
||||
from freqtrade.data.converter.orderflow import trades_to_volumeprofile_with_total_delta_bid_ask
|
||||
from freqtrade.data.converter.orderflow import (
|
||||
ORDERFLOW_ADDED_COLUMNS,
|
||||
trades_to_volumeprofile_with_total_delta_bid_ask,
|
||||
)
|
||||
from freqtrade.data.converter.trade_converter import trades_list_to_df
|
||||
from freqtrade.data.dataprovider import DataProvider
|
||||
from tests.strategy.strats.strategy_test_v3 import StrategyTestV3
|
||||
@@ -505,21 +508,8 @@ def test_analyze_with_orderflow(
|
||||
assert "open" in df.columns
|
||||
assert spy.call_count == 0
|
||||
|
||||
expected_cols = [
|
||||
"trades",
|
||||
"orderflow",
|
||||
"imbalances",
|
||||
"stacked_imbalances_bid",
|
||||
"stacked_imbalances_ask",
|
||||
"max_delta",
|
||||
"min_delta",
|
||||
"bid",
|
||||
"ask",
|
||||
"delta",
|
||||
"total_trades",
|
||||
]
|
||||
# Not expected to run - shouldn't have added orderflow columns
|
||||
for col in expected_cols:
|
||||
for col in ORDERFLOW_ADDED_COLUMNS:
|
||||
assert col not in df.columns, f"Column {col} found in df.columns"
|
||||
|
||||
default_conf_usdt["exchange"]["use_public_trades"] = True
|
||||
@@ -539,7 +529,7 @@ def test_analyze_with_orderflow(
|
||||
assert "open" in df1.columns
|
||||
assert spy.call_count == 5
|
||||
|
||||
for col in expected_cols:
|
||||
for col in ORDERFLOW_ADDED_COLUMNS:
|
||||
assert col in df1.columns, f"Column {col} not found in df.columns"
|
||||
|
||||
if col not in ("stacked_imbalances_bid", "stacked_imbalances_ask"):
|
||||
@@ -560,7 +550,7 @@ def test_analyze_with_orderflow(
|
||||
assert len(df2) == len(ohlcv_history)
|
||||
assert "open" in df2.columns
|
||||
assert spy.call_count == 0
|
||||
for col in expected_cols:
|
||||
for col in ORDERFLOW_ADDED_COLUMNS:
|
||||
assert col in df2.columns, f"Round2: Column {col} not found in df.columns"
|
||||
|
||||
if col not in ("stacked_imbalances_bid", "stacked_imbalances_ask"):
|
||||
|
||||
Reference in New Issue
Block a user