roboquant 2.3.2__tar.gz → 2.3.4__tar.gz

{roboquant-2.3.2 → roboquant-2.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: roboquant
-Version: 2.3.2
+Version: 2.3.4
 Summary: A fast algorithmic trading platform with support for AI based strategies
 Keywords: algorithmic,trading,machine learning,AI,investment,finance,crypto,stocks,exchange,forex
 Author: roboquant team
@@ -77,7 +77,7 @@ python3 -m pip install --upgrade roboquant
 The core of roboquant limits the number of dependencies. But you can install roboquant including one or more of the optional dependencies if you require certain additional functionality:
 
 ```shell
-# AI based strategies using Torch and SB3
+# AI based strategies using PyTorch and SB3
 python3 -m pip install --upgrade "roboquant[ai]"
 
 # Integration with Interactive Brokers

{roboquant-2.3.2 → roboquant-2.3.4}/README.md

@@ -35,7 +35,7 @@ python3 -m pip install --upgrade roboquant
 The core of roboquant limits the number of dependencies. But you can install roboquant including one or more of the optional dependencies if you require certain additional functionality:
 
 ```shell
-# AI based strategies using Torch and SB3
+# AI based strategies using PyTorch and SB3
 python3 -m pip install --upgrade "roboquant[ai]"
 
 # Integration with Interactive Brokers

{roboquant-2.3.2 → roboquant-2.3.4}/pyproject.toml

@@ -19,7 +19,7 @@ module-name = "roboquant"
 
 [project]
 name = "roboquant"
-version = "2.3.2"
+version = "2.3.4"
 authors = [{ name = "roboquant team", email = "info@roboquant.org" }]
 description = "A fast algorithmic trading platform with support for AI based strategies"
 readme = "README.md"

{roboquant-2.3.2 → roboquant-2.3.4}/roboquant/ai/features.py

@@ -16,9 +16,16 @@ FloatArray = NDArray[np.float32]
 
 
 class Feature(Generic[T]):
-    """Base class for different types of features.
+    """Base class for different types of features. Features are the data building blocks for
+    roboquant machine learning models and can be used to extract relevant information.
+
     The ones included by default are either based either an `Event` or an `Account`.
-    Typically Event features are used for input and Account features are used for reward/label/output."""
+    Typically Event features are used for input data and Account features are used for
+    reward/label/output data.
+
+    Features can be combined, sliced, nested, normalized and cached to create more
+    complex features.
+    """
 
     @abstractmethod
     def calc(self, value: T) -> FloatArray:
@@ -33,7 +40,8 @@ class Feature(Generic[T]):
         """Return the size of this feature"""
 
     def reset(self):
-        """Reset the state of the feature"""
+        """Reset the state of the feature. This is called at the start of a new epoch.
+        If the feature has no state, this can be left empty."""
 
     def _shape(self):
         """return the shape of this feature as a tuple"""
@@ -244,7 +252,7 @@ class ReturnFeature(Feature[T]):
 class LongReturnsFeature(Feature[T]):
     def __init__(self, feature: Feature[T], period: int) -> None:
         super().__init__()
-        self.history = deque(maxlen=period)
+        self.history: deque[FloatArray] = deque(maxlen=period)
         self.feature: Feature = feature
 
     def calc(self, value: T) -> FloatArray:
@@ -275,7 +283,7 @@ class MaxReturnFeature(Feature[T]):
     def __init__(self, feature: Feature[T], period: int) -> None:
         super().__init__()
         assert feature.size() == 1
-        self.history = deque(maxlen=period)
+        self.history: deque[FloatArray] = deque(maxlen=period)
         self.feature: Feature = feature
 
     def calc(self, value: T) -> FloatArray:
@@ -304,7 +312,7 @@ class MinReturnFeature(Feature[T]):
 
     def __init__(self, feature: Feature[T], period: int) -> None:
         super().__init__()
-        self.history = deque(maxlen=period)
+        self.history: deque[FloatArray] = deque(maxlen=period)
         self.feature: Feature = feature
 
     def calc(self, value: T) -> FloatArray:
@@ -429,6 +437,7 @@ class DayOfMonthFeature(Feature[Event]):
     """Calculate a day of month where the first day of the month is 0 and the last day is 30.
     Result can be one-hot encoded or not, depending on the `one_hot_encoded` parameter.
     If `one_hot_encoded` is True, the result will be a 30-element array else the result will be a single value.
+
     For example, if the event time is the 15th of the month, the result will be:
     - one-hot encoded: [0.0, 0.0, ..., 1.0, 0.0, ..., 0.0] (1.0 at index 14)
     - not one-hot encoded: [14.0]
@@ -637,11 +646,14 @@ class QuoteFeature(Feature[Event]):
 
 
 class CacheFeature(Feature[Event]):
-    """Cache the results of an event feature from a previous run. This can speed up the learning process considerable, but
-    this requires that:
+    """Cache the results of an event feature from a previous run. This can speed up the learning process
+    considerable, but this requires that:
 
     - the feed to have always an increasing time value (monotonic)
-    - the feature has to produce the same output at a given time.
+    - the underlying feature has to produce the same output at a given time (deterministic)
+
+    A reset doesn't clear the cache, but it will reset the underlying feature. If you want to clear
+    the cache, use the `clear()` method.
     """
 
     def __init__(self, feature: Feature[Event], validate=False) -> None:
@@ -678,7 +690,10 @@ class CacheFeature(Feature[Event]):
 
 
 class VolumeFeature(Feature[Event]):
-    """Extract the volume for one or more assets in the event"""
+    """Extract the trading volume for one or more assets in the event.
+    Make sure that the data avaialble in the event contains adjusted volume data (for example for stock splits),
+    otherwise the results will be incorrect.
+    """
 
     def __init__(self, *assets: Asset, volume_type: str = "DEFAULT") -> None:
         super().__init__()

{roboquant-2.3.2 → roboquant-2.3.4}/roboquant/ai/strategies.py

@@ -19,7 +19,10 @@ logger = logging.getLogger(__name__)
 
 
 class FeatureStrategy(Strategy):
-    """Abstract base class for strategies that use event features as input."""
+    """Abstract base class for strategies that use event features as input.
+    Typically these features are used to create input data for machine learning models,
+    which can then be used to generate signals.
+    """
 
     def __init__(self, input_feature: Feature[Event], history: int, dtype="float32"):
         super().__init__()
@@ -104,9 +107,13 @@ class TimeSeriesStrategy(FeatureStrategy):
     """Strategy using a sequenced neural network to predict future time series values.
     The input and label features are calculated from events.
     The model is expected to be a PyTorch module that takes the input features and outputs predictions.
+    An example of a model could be a recurrent neural network (RNN) or a long short-term memory (LSTM) network.
+
     The strategy can be used to generate buy/sell signals based on the model's predictions.
     This strategy is suitable for time series forecasting tasks where the model learns from historical data
-    to predict future values. Right now it only supports a single asset, but this can be extended in the future.
+    to predict future values.
+
+    Right now this class only supports a single asset, but this can be extended.
     """
 
     def __init__(
@@ -140,7 +147,7 @@ class TimeSeriesStrategy(FeatureStrategy):
             else:
                 p = output.item()
 
-            logger.info("Prediction p=%s time=%s", p, dt)
+            logger.info("Prediction asset=%s p=%s time=%s", self.asset.symbol, p, dt)
             if p >= self.buy_pct:
                 return [Signal.buy(self.asset)]
             if p <= self.sell_pct:

roboquant-2.3.4/roboquant/crypto/__init__.py

@@ -0,0 +1,4 @@
+from roboquant.crypto.cryptobroker import CryptoBroker
+from roboquant.crypto.cryptofeed import CryptoFeed
+
+__all__ = ["CryptoFeed", "CryptoBroker"]

{roboquant-2.3.2 → roboquant-2.3.4}/roboquant/signal.py

@@ -22,8 +22,7 @@ class SignalType(Flag):
     """Indicates that this signal can be used for both entering and exiting a position size"""
 
     def __repr__(self):
-        return self.name.split(".")[-1]  # type: ignore
-
+        return self.name.split(".")[-1] if self.name else super().__repr__()
 
 @dataclass(slots=True, frozen=True)
 class Signal:

roboquant-2.3.2/roboquant/crypto/__init__.py

@@ -1,3 +0,0 @@
-from roboquant.crypto import CryptoFeed, CryptoBroker
-
-__all__ = ["CryptoFeed", "CryptoBroker"]