Merge pull request freqtrade#11178 from TheJoeSchr/fix/orderflow_imbalance_list

Fix:orderflow returns a list for stacked imbalances

Author: xmatthias

docs/advanced-orderflow.md

@@ -70,8 +70,8 @@ dataframe["delta"] # Difference between ask and bid volume.
 dataframe["min_delta"] # Minimum delta within the candle
 dataframe["max_delta"] # Maximum delta within the candle
 dataframe["total_trades"] # Total number of trades
-dataframe["stacked_imbalances_bid"] # Price level of stacked bid imbalance 
-dataframe["stacked_imbalances_ask"] # Price level of stacked ask imbalance  
+dataframe["stacked_imbalances_bid"] # List of price levels of stacked bid imbalance range beginnings
+dataframe["stacked_imbalances_ask"] # List of price levels of stacked ask imbalance range beginnings
 ```
 
 You can access these columns in your strategy code for further analysis. Here's an example:

freqtrade/data/converter/orderflow.py

@@ -164,12 +164,12 @@ def populate_dataframe_with_trades(
                 dataframe.at[index, "imbalances"] = imbalances.to_dict(orient="index")
 
                 stacked_imbalance_range = config_orderflow["stacked_imbalance_range"]
-                dataframe.at[index, "stacked_imbalances_bid"] = stacked_imbalance_bid(
-                    imbalances, stacked_imbalance_range=stacked_imbalance_range
+                dataframe.at[index, "stacked_imbalances_bid"] = stacked_imbalance(
+                    imbalances, label="bid", stacked_imbalance_range=stacked_imbalance_range
                 )
 
-                dataframe.at[index, "stacked_imbalances_ask"] = stacked_imbalance_ask(
-                    imbalances, stacked_imbalance_range=stacked_imbalance_range
+                dataframe.at[index, "stacked_imbalances_ask"] = stacked_imbalance(
+                    imbalances, label="ask", stacked_imbalance_range=stacked_imbalance_range
                 )
 
                 bid = np.where(
@@ -256,34 +256,24 @@ def trades_orderflow_to_imbalances(df: pd.DataFrame, imbalance_ratio: int, imbal
     return dataframe
 
 
-def stacked_imbalance(
-    df: pd.DataFrame, label: str, stacked_imbalance_range: int, should_reverse: bool
-):
+def stacked_imbalance(df: pd.DataFrame, label: str, stacked_imbalance_range: int):
     """
     y * (y.groupby((y != y.shift()).cumsum()).cumcount() + 1)
     https://stackoverflow.com/questions/27626542/counting-consecutive-positive-values-in-python-pandas-array
     """
     imbalance = df[f"{label}_imbalance"]
     int_series = pd.Series(np.where(imbalance, 1, 0))
-    stacked = int_series * (
-        int_series.groupby((int_series != int_series.shift()).cumsum()).cumcount() + 1
-    )
-
-    max_stacked_imbalance_idx = stacked.index[stacked >= stacked_imbalance_range]
-    stacked_imbalance_price = np.nan
-    if not max_stacked_imbalance_idx.empty:
-        idx = (
-            max_stacked_imbalance_idx[0]
-            if not should_reverse
-            else np.flipud(max_stacked_imbalance_idx)[0]
-        )
-        stacked_imbalance_price = imbalance.index[idx]
-    return stacked_imbalance_price
-
-
-def stacked_imbalance_ask(df: pd.DataFrame, stacked_imbalance_range: int):
-    return stacked_imbalance(df, "ask", stacked_imbalance_range, should_reverse=True)
-
-
-def stacked_imbalance_bid(df: pd.DataFrame, stacked_imbalance_range: int):
-    return stacked_imbalance(df, "bid", stacked_imbalance_range, should_reverse=False)
+    # Group consecutive True values and get their counts
+    groups = (int_series != int_series.shift()).cumsum()
+    counts = int_series.groupby(groups).cumsum()
+
+    # Find indices where count meets or exceeds the range requirement
+    valid_indices = counts[counts >= stacked_imbalance_range].index
+
+    stacked_imbalance_prices = []
+    if not valid_indices.empty:
+        # Get all prices from valid indices from beginning of the range
+        stacked_imbalance_prices = [
+            imbalance.index.values[idx - (stacked_imbalance_range - 1)] for idx in valid_indices
+        ]
+    return stacked_imbalance_prices

tests/data/test_converter_orderflow.py

@@ -1,11 +1,11 @@
-import numpy as np
 import pandas as pd
 import pytest
 
 from freqtrade.constants import DEFAULT_TRADES_COLUMNS
 from freqtrade.data.converter import populate_dataframe_with_trades
 from freqtrade.data.converter.orderflow import (
     ORDERFLOW_ADDED_COLUMNS,
+    stacked_imbalance,
     timeframe_to_DateOffset,
     trades_to_volumeprofile_with_total_delta_bid_ask,
 )
@@ -185,24 +185,24 @@ def test_public_trades_mock_populate_dataframe_with_trades__check_orderflow(
     assert results["max_delta"] == 17.298
 
     # Assert that stacked imbalances are NaN (not applicable in this test)
-    assert np.isnan(results["stacked_imbalances_bid"])
-    assert np.isnan(results["stacked_imbalances_ask"])
+    assert results["stacked_imbalances_bid"] == []
+    assert results["stacked_imbalances_ask"] == []
 
     # Repeat assertions for the third from last row
     results = df.iloc[-2]
     assert pytest.approx(results["delta"]) == -20.862
     assert pytest.approx(results["min_delta"]) == -54.559999
     assert 82.842 == results["max_delta"]
-    assert 234.99 == results["stacked_imbalances_bid"]
-    assert 234.96 == results["stacked_imbalances_ask"]
+    assert results["stacked_imbalances_bid"] == [234.97]
+    assert results["stacked_imbalances_ask"] == [234.94]
 
     # Repeat assertions for the last row
     results = df.iloc[-1]
     assert pytest.approx(results["delta"]) == -49.302
     assert results["min_delta"] == -70.222
     assert pytest.approx(results["max_delta"]) == 11.213
-    assert np.isnan(results["stacked_imbalances_bid"])
-    assert np.isnan(results["stacked_imbalances_ask"])
+    assert results["stacked_imbalances_bid"] == []
+    assert results["stacked_imbalances_ask"] == []
 
 
 def test_public_trades_trades_mock_populate_dataframe_with_trades__check_trades(
@@ -358,7 +358,8 @@ def test_public_trades_binned_big_sample_list(public_trades_list):
     assert 197.512 == df["bid_amount"].iloc[0]  # total bid amount
     assert 88.98 == df["ask_amount"].iloc[0]  # total ask amount
     assert 26 == df["ask"].iloc[0]  # ask price
-    assert -108.532 == pytest.approx(df["delta"].iloc[0])  # delta (bid amount - ask amount)
+    # delta (bid amount - ask amount)
+    assert -108.532 == pytest.approx(df["delta"].iloc[0])
 
     assert 3 == df["bid"].iloc[-1]  # bid price
     assert 50.659 == df["bid_amount"].iloc[-1]  # total bid amount
@@ -567,6 +568,40 @@ def test_analyze_with_orderflow(
     assert isinstance(lastval_of2, dict)
 
 
+def test_stacked_imbalances_multiple_prices():
+    """Test that stacked imbalances correctly returns multiple price levels when present"""
+    # Test with empty result
+    df_no_stacks = pd.DataFrame(
+        {
+            "bid_imbalance": [False, False, True, False],
+            "ask_imbalance": [False, True, False, False],
+        },
+        index=[234.95, 234.96, 234.97, 234.98],
+    )
+    no_stacks = stacked_imbalance(df_no_stacks, "bid", stacked_imbalance_range=2)
+    assert no_stacks == []
+
+    # Create a sample DataFrame with known imbalances
+    df = pd.DataFrame(
+        {
+            "bid_imbalance": [True, True, True, False, False, True, True, False, True],
+            "ask_imbalance": [False, False, True, True, True, False, False, True, True],
+        },
+        index=[234.95, 234.96, 234.97, 234.98, 234.99, 235.00, 235.01, 235.02, 235.03],
+    )
+    # Test bid imbalances (should return prices in ascending order)
+    bid_prices = stacked_imbalance(df, "bid", stacked_imbalance_range=2)
+    assert bid_prices == [234.95, 234.96, 235.00]
+
+    # Test ask imbalances (should return prices in descending order)
+    ask_prices = stacked_imbalance(df, "ask", stacked_imbalance_range=2)
+    assert ask_prices == [234.97, 234.98, 235.02]
+
+    # Test with higher stacked_imbalance_range
+    bid_prices_higher = stacked_imbalance(df, "bid", stacked_imbalance_range=3)
+    assert bid_prices_higher == [234.95]
+
+
 def test_timeframe_to_DateOffset():
     assert timeframe_to_DateOffset("1s") == pd.DateOffset(seconds=1)
     assert timeframe_to_DateOffset("1m") == pd.DateOffset(minutes=1)