tweaks

README.md

@@ -63,15 +63,17 @@ Allora Model Maker is a comprehensive machine learning framework designed for ti
 
 2. Create a conda environment:
     ```bash
-   conda create --name modelmaker python=3.9 && conda activate modelmaker
+    conda env create -f environment.yml
     ```
-
-3. Preinstall setuptools, cython and numpy
+    If you want to manually do it:
+    ```bash
+    conda create --name modelmaker python=3.9 && conda activate modelmaker
+    ```
+   Preinstall setuptools, cython and numpy
    ```bash
    pip install setuptools==72.1.0 Cython==3.0.11 numpy==1.24.3
    ```
-
-4. Install dependencies:
+   Install dependencies:
     ```bash
    pip install -r requirements.txt
     ```

data/tiingo_data_fetcher.py

@@ -6,9 +6,11 @@
 
 # Load the .env.local file if it exists, otherwise load .env
 if os.path.exists(".env.local"):
-    load_dotenv(dotenv_path=".env.local")
+    print("Loading .env.local file...")
+    load_dotenv(dotenv_path=".env.local", override=True)
 else:
-    load_dotenv()  # Defaults to loading .env
+    print("Loading .env file...")
+    load_dotenv(dotenv_path=".env")  # Defaults to loading .env
 
 # Retrieve the API keys from environment variables
 TIINGO_API_KEY = os.getenv("TIINGO_API_KEY")

environment.yml

@@ -0,0 +1,137 @@
+name: modelmaker
+channels:
+  - conda-forge
+  - defaults
+dependencies:
+  - appnope=0.1.4
+  - asttokens=2.4.1
+  - bzip2=1.0.8
+  - ca-certificates=2024.8.30
+  - comm=0.2.2
+  - debugpy=1.8.6
+  - decorator=5.1.1
+  - exceptiongroup=1.2.2
+  - executing=2.1.0
+  - importlib-metadata=8.5.0
+  - ipykernel=6.29.5
+  - ipython=8.18.1
+  - jedi=0.19.1
+  - jupyter_client=8.6.3
+  - jupyter_core=5.7.2
+  - krb5=1.21.3
+  - libcxx=19.1.0
+  - libedit=3.1.20191231
+  - libffi=3.4.2
+  - libsodium=1.0.20
+  - libsqlite=3.46.1
+  - libzlib=1.3.1
+  - matplotlib-inline=0.1.7
+  - ncurses=6.5
+  - nest-asyncio=1.6.0
+  - openssl=3.3.2
+  - packaging=24.1
+  - parso=0.8.4
+  - pexpect=4.9.0
+  - pickleshare=0.7.5
+  - pip=24.2
+  - platformdirs=4.3.6
+  - prompt-toolkit=3.0.48
+  - psutil=6.0.0
+  - ptyprocess=0.7.0
+  - pure_eval=0.2.3
+  - pygments=2.18.0
+  - python=3.9.20
+  - python_abi=3.9
+  - pyzmq=26.2.0
+  - readline=8.2
+  - six=1.16.0
+  - stack_data=0.6.2
+  - tk=8.6.13
+  - tornado=6.4.1
+  - traitlets=5.14.3
+  - typing_extensions=4.12.2
+  - wcwidth=0.2.13
+  - wheel=0.44.0
+  - xz=5.2.6
+  - zeromq=4.3.5
+  - zipp=3.20.2
+  - pip:
+      - aiohappyeyeballs==2.4.3
+      - aiohttp==3.10.8
+      - aiosignal==1.3.1
+      - annotated-types==0.7.0
+      - anyio==4.6.0
+      - astroid==3.2.4
+      - async-timeout==4.0.3
+      - attrs==24.2.0
+      - black==24.8.0
+      - certifi==2024.8.30
+      - cfgv==3.4.0
+      - charset-normalizer==3.3.2
+      - click==8.1.7
+      - cmdstanpy==1.2.4
+      - contourpy==1.3.0
+      - cycler==0.12.1
+      - cython==3.0.11
+      - dill==0.3.9
+      - distlib==0.3.8
+      - filelock==3.16.1
+      - fonttools==4.54.1
+      - frozenlist==1.4.1
+      - fsspec==2024.9.0
+      - holidays==0.57
+      - identify==2.6.1
+      - idna==3.10
+      - importlib-resources==6.4.5
+      - isort==5.13.2
+      - jinja2==3.1.4
+      - joblib==1.4.2
+      - kiwisolver==1.4.7
+      - markupsafe==2.1.5
+      - matplotlib==3.9.2
+      - mccabe==0.7.0
+      - mpmath==1.3.0
+      - multidict==6.1.0
+      - mypy-extensions==1.0.0
+      - networkx==3.2.1
+      - nodeenv==1.9.1
+      - numpy==1.24.3
+      - pandas==2.2.3
+      - pathspec==0.12.1
+      - patsy==0.5.6
+      - pillow==10.4.0
+      - pipdeptree==2.23.4
+      - plotly==5.24.0
+      - pre-commit==3.8.0
+      - prophet==1.1.5
+      - pydantic==2.9.2
+      - pydantic-core==2.23.4
+      - pylint==3.2.7
+      - pyparsing==3.1.4
+      - pystan==2.19.1.1
+      - python-dateutil==2.9.0.post0
+      - python-dotenv==1.0.1
+      - pytz==2024.2
+      - pyyaml==6.0.2
+      - requests==2.32.3
+      - schedule==1.2.2
+      - scikit-learn==1.5.1
+      - scipy==1.13.1
+      - seaborn==0.13.2
+      - setuptools==72.1.0
+      - sniffio==1.3.1
+      - stanio==0.5.1
+      - starlette==0.38.5
+      - statsmodels==0.14.2
+      - sympy==1.13.3
+      - tenacity==9.0.0
+      - threadpoolctl==3.5.0
+      - tomli==2.0.1
+      - tomlkit==0.13.2
+      - torch==2.4.1
+      - tqdm==4.66.5
+      - tzdata==2024.2
+      - urllib3==2.2.3
+      - virtualenv==20.26.6
+      - xgboost==2.1.1
+      - yarl==1.13.1

models/arima/model.py

@@ -40,17 +40,20 @@ def train(self, data: pd.DataFrame):
             print("Data is not stationary, applying differencing...")
             close_prices = differencing(close_prices)
 
-        # Perform grid search if enabled in the configuration
+        # Perform enhanced grid search if enabled in the configuration
         if self.config.use_grid_search:
+            print("Performing grid search for ARIMA parameters...")
             grid_search_arima(
                 self,
                 data=close_prices,
                 p_values=self.config.p_values,
                 d_values=self.config.d_values,
                 q_values=self.config.q_values,
             )
+        else:
+            print("Using default ARIMA parameters...")
 
-        # Fit ARIMA model with the configured best parameters
+        # Fit ARIMA model with the configured or best-found parameters
         self.model = ARIMA(close_prices, order=self.config.best_params)
         self.model = self.model.fit(method_kwargs={"maxiter": self.config.max_iter})
 
@@ -73,23 +76,19 @@ def inference(self, input_data: pd.DataFrame) -> pd.DataFrame:
         close_prices = resample_data(self, close_prices)
 
         # Forecast the number of steps equal to the length of the input data
-        # pylint: disable=no-member
         predictions = self.model.forecast(steps=len(close_prices))
 
         # If differencing was applied, reverse the differencing to get price predictions
         if self.config.best_params[1] > 0:  # If d > 0, reverse differencing
             predictions = reverse_differencing(close_prices, predictions)
 
-        # Optionally check for negative values and flag them instead of clipping
-        predictions[predictions < 0] = (
-            np.nan
-        )  # Replace unreasonable negative prices with NaN
+        # Replace unreasonable negative values with NaN
+        predictions[predictions < 0] = np.nan
 
-        # convert the date to string
+        # Convert the date to string and return results
         predictions = pd.Series(
             predictions.values, index=close_prices.index.astype(str)
         )
-
         return pd.DataFrame(
             {"date": predictions.index, "prediction": predictions.values.ravel()}
         )

models/lstm/configs.py

@@ -11,13 +11,13 @@ class LstmConfig:
     def __init__(self):
         # Model architecture parameters
         self.input_size = 1  # Input size (number of features)
-        self.hidden_size = 50  # Number of LSTM units per layer
+        self.hidden_size = 64  # Number of LSTM units per layer
         self.output_size = 1  # Output size (prediction dimension)
         self.num_layers = 2  # Number of stacked LSTM layers
         self.dropout = 0.5  # Dropout probability for regularization
 
         # Training parameters
-        self.learning_rate = 0.001  # Learning rate for the optimizer
+        self.learning_rate = 0.0001  # Learning rate for the optimizer
         self.batch_size = 32  # Batch size for training
         self.epochs = 100  # Number of training epochs
         self.early_stopping_patience = 10  # Early stopping patience in epochs

models/prophet/configs.py

@@ -10,12 +10,15 @@ class ProphetConfig:
 
     def __init__(self):
         # Prophet model configuration
-        self.growth = "linear"  # Options: 'linear', 'logistic'
-        self.changepoint_prior_scale = 0.05  # Regularization strength for changepoints
-        self.seasonality_mode = "additive"  # Options: 'additive', 'multiplicative'
-        self.yearly_seasonality = True  # Whether to include yearly seasonality
+        self.growth = "logistic"  # Options: 'linear', 'logistic'
+        self.cap = None  # Optional, for logistic growth. Can be dynamically calculated if None.
+        self.changepoint_prior_scale = 0.25  # Regularization strength for changepoints
+        self.seasonality_mode = (
+            "multiplicative"  # Options: 'additive', 'multiplicative'
+        )
+        self.yearly_seasonality = False  # Whether to include yearly seasonality
         self.weekly_seasonality = True  # Whether to include weekly seasonality
-        self.daily_seasonality = False  # Whether to include daily seasonality
+        self.daily_seasonality = True  # Whether to include daily seasonality
 
         # Forecast parameters
         self.periods = 365  # Default number of periods for future forecasts (trading days for stocks usually 252)

models/prophet/model.py

@@ -22,12 +22,30 @@ def __init__(self, model_name="prophet", config=ProphetConfig(), debug=False):
 
     def train(self, data: pd.DataFrame):
         df = data[["date", "close"]].copy()
+        df["date"] = pd.to_datetime(df["date"], errors="coerce")
+
+        # Drop rows with NaN values in 'date' or 'close'
+        df = df.dropna(subset=["date", "close"])
 
         if self.config.remove_timezone:
-            df["date"] = df["date"].dt.tz_localize(None)  # Remove timezone if present
+            df["date"] = df["date"].dt.tz_localize(None)
 
         df = df.rename(columns={"date": "ds", "close": "y"})
-        self.model.fit(df)
+
+        # Handle logistic growth: Set 'cap' and 'floor' values
+        if self.config.growth == "logistic":
+            max_y = df["y"].max()
+            df["cap"] = max_y * 1.1  # Set cap to 10% above max value
+            df["floor"] = 0  # Set a floor to prevent negative growth
+
+        if self.debug:
+            # Print data to check for NaNs or extreme values
+            print(df.isna().sum())
+            print(df.describe())
+
+        # Fit model
+        self.model.fit(df, iter=20000)
+
         self.save()
 
     def inference(self, input_data: pd.DataFrame) -> pd.DataFrame:
@@ -47,12 +65,24 @@ def inference(self, input_data: pd.DataFrame) -> pd.DataFrame:
         if self.config.remove_timezone:
             future["ds"] = pd.to_datetime(future["ds"]).dt.tz_localize(None)
 
+        # Handle logistic growth: Ensure 'cap' and 'floor' columns are present
+        if self.config.growth == "logistic":
+            if "cap" not in input_data.columns:
+                # Dynamically set 'cap' if missing (assuming similar logic as training)
+                future["cap"] = (
+                    input_data["close"].max() * 1.1
+                )  # Example logic for setting 'cap'
+
+            if "floor" not in input_data.columns:
+                # Set the 'floor' if it's missing (e.g., default to 0)
+                future["floor"] = 0
+
         if self.debug:
             print("Future DataFrame for ProphetModel (after renaming):")
             print(future)
 
         # Perform the forecast using Prophet
-        forecast = self.model.predict(future)
+        forecast = self.model.predict(future, False)
         if self.debug:
             print("Forecast Output:")
             print(forecast[["ds", "yhat"]])