this model is so fucking ass

README.md

@@ -1,28 +1,5 @@
 # midterm ml&dm1
 
-## not cool
+## real
 
-currently this shit is a blatant copy. changing soon.
+not real
-
-## how to
-
-load this thing on google colab dont run on device its so cooked
-
-how to git? do not `push --force`. prefer pull requests. its like github
-
-## roadmap
-
-we will plan to do this in multiple different ways using scikit-learn and pytorch. tensorflow is ancient.
-
-> [!NOTE]
-> ~phase out tensorflow. dont use it.~
->
-> its 2 different platforms for 2 different types of people. use whatever ig
-
-maybe also fit more weather related stats? this dataset is taken from nasa so sources are there
-
-## collab
-
-seng tells me this project can be planned without code. basically someone find da waes and other people implements them
-
-thing is you click [here at scikit learn](https://scikit-learn.org/stable/supervised_learning.html) and tell me how many ways there are you can use to solve a problem. holy fucking shit. too many.

data_loader.py

@@ -32,9 +32,10 @@ def load_weather_data(file_path: str) -> pd.DataFrame:
 
     # Convert YEAR, MO, DY to datetime index
     if all(col in data.columns for col in ("YEAR", "MO", "DY")):
-        data["date"] = pd.to_datetime(
+        dt_conv_table = {
-            {k: data[v] for v, k in zip(["YEAR", "MO", "DY"], ["year", "month", "day"])}
+            k: data[v] for v, k in zip(["YEAR", "MO", "DY"], ["year", "month", "day"])
-        )
+        }
+        data["date"] = pd.to_datetime(dt_conv_table)
         data.drop(columns=["YEAR", "MO", "DY"], inplace=True)
         data.set_index("date", inplace=True)
 

main.py

@@ -1,4 +1,6 @@
 # import pandas as pd
+from matplotlib import pyplot as plt
+import numpy as np
 from sklearn.model_selection import train_test_split  # type: ignore[import]
 
 from data_loader import load_weather_data
@@ -6,6 +8,7 @@ from model_train import (
     cosine_lr_schedule,
     evaluate_model,
     predict_linear_regression,
+    run_autoregressive_prediction,
     train_linear_regression_gd,
 )
 from preprocessing import create_sequences, inverse_scale, scale_data
@@ -80,3 +83,9 @@ plot_scatter(y_test.reshape(-1, 1), predictions.reshape(-1, 1))
 plot_residual_scatter(y_test, predictions)
 plot_loss_curve_with_annotations(losses)
 plot_feature_importance(theta, feature_names=[f"QV2M_t-{i}" for i in range(seq_length)])
+
+
+# Step 8: Autoregressive Prediction
+autoregressive_predictions = run_autoregressive_prediction(
+    theta, X_test, y_test, scaler, date_test, data.shape[1]
+)

model_train.py

@@ -1,7 +1,10 @@
 # from itertools import islice
+from matplotlib import pyplot as plt
 import numpy as np
 from sklearn.metrics import mean_squared_error, r2_score  # type: ignore[import]
 
+from preprocessing import inverse_scale
+
 
 def train_linear_regression_gd(
     X_train,
@@ -94,3 +97,78 @@ def cosine_lr_schedule(epoch, max_lr=0.01, min_lr=0.001, total_epochs=1000):
     return (
         min_lr + (max_lr - min_lr) * (1 + math.cos(math.pi * epoch / total_epochs)) / 2
     )
+
+
+def autoregressive_prediction(model, initial_sequence, steps, scaler, n_features):
+    """
+    Perform autoregressive predictions using the model.
+
+    Args:
+        model: The trained model (theta for linear regression).
+        initial_sequence: The initial sequence to start predictions (from test data).
+        steps: Number of steps to predict.
+        scaler: Scaler used for inverse scaling.
+        n_features: Number of features in the data.
+
+    Returns:
+        predictions: List of predicted values.
+    """
+    sequence = initial_sequence.copy()  # Start with the initial sequence
+    predictions = []
+
+    for _ in range(steps):
+        # Predict the next value
+        next_prediction = predict_linear_regression(sequence.reshape(1, -1), model)
+        predictions.append(next_prediction[0])
+
+        # Update the sequence: Remove the oldest value and append the new prediction
+        sequence = np.roll(sequence, -1)  # Shift sequence left
+        sequence[-1] = next_prediction  # Add the new prediction
+
+    # Rescale predictions back to original scale
+    predictions_rescaled = inverse_scale(np.array(predictions), scaler, n_features)
+    return predictions_rescaled
+
+
+def run_autoregressive_prediction(
+    model, X_test, y_test, scaler, date_test, n_features, steps=None
+):
+    """
+    Run autoregressive prediction and visualize the results.
+
+    Args:
+        model: The trained model (theta for linear regression).
+        X_test: Test features.
+        y_test: Test labels.
+        scaler: Scaler used for inverse scaling.
+        date_test: Dates corresponding to the test set.
+        n_features: Number of features in the data.
+        steps: Number of steps to predict (default is the length of y_test).
+
+    Returns:
+        autoregressive_predictions: Rescaled autoregressive predictions.
+    """
+    if steps is None:
+        steps = len(y_test)  # Default to the length of the test set
+
+    # Use the first test sequence as the initial sequence
+    initial_sequence = X_test[0]
+
+    # Perform autoregressive prediction
+    autoregressive_predictions = autoregressive_prediction(
+        model, initial_sequence, steps, scaler, n_features
+    )
+
+    # Visualize the results
+    plt.figure(figsize=(12, 6))
+    plt.plot(date_test, inverse_scale(y_test, scaler, n_features), label="Actual")
+    plt.plot(date_test, autoregressive_predictions, label="Autoregressive Predictions")
+    plt.title("Autoregressive Predictions vs Actual")
+    plt.xlabel("Date")
+    plt.ylabel("Value")
+    plt.legend()
+    plt.grid(True)
+    plt.tight_layout()
+    plt.show()
+
+    return autoregressive_predictions

preprocessing.py

@@ -1,5 +1,6 @@
 import numpy as np
 from sklearn.preprocessing import MinMaxScaler  # type: ignore[import]
+
 # from sklearn.model_selection import train_test_split