Adding baseline policy evaluator

.gitattributes

@@ -1 +0,0 @@
-*.csv filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -2,6 +2,7 @@ FROM pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime
 
 RUN apt-get update
 RUN apt-get install -y git
+RUN apt-get install texlive-latex-base texlive-fonts-recommended texlive-fonts-extra texlive-bibtex-extra
 
 COPY requirements.txt /tmp/requirements.txt
 

Result-Reports/Policies.md

@@ -159,7 +159,7 @@ Test data: 01-01-2023 until 08-10–2023
 
 
 TODO:
-- [ ] diffusion model oefening generative models vragen
+- [x] diffusion model oefening generative models vragen -> geen lab hierop
 - [ ] Non autoregressive models policy testen (Non Linear eerst) -> als dit al slect, niet verder kijken, wel vermelden
 - [ ] Policy in test set -> over charge cycles (stop trading electricity)
 

src/data/dataset.py

@@ -25,12 +25,19 @@ class NrvDataset(Dataset):
         self.sequence_length = sequence_length
         self.predict_sequence_length = predict_sequence_length
 
-        self.samples_to_skip = self.skip_samples(dataframe=dataframe)
+        self.samples_to_skip = self.skip_samples(dataframe=dataframe, full_day_skip=self.full_day_skip)
         total_indices = set(
             range(len(dataframe) - self.sequence_length - self.predict_sequence_length)
         )
         self.valid_indices = sorted(list(total_indices - set(self.samples_to_skip)))
 
+        # full day indices
+        full_day_skipped_samples = self.skip_samples(dataframe=dataframe, full_day_skip=True)
+        full_day_total_indices = set(
+            range(len(dataframe) - self.sequence_length - self.predict_sequence_length)
+        )
+        self.full_day_valid_indices = sorted(list(full_day_total_indices - set(full_day_skipped_samples)))
+
         self.history_features = []
         if self.data_config.LOAD_HISTORY:
             self.history_features.append("total_load")
@@ -73,7 +80,7 @@ class NrvDataset(Dataset):
 
         self.history_features, self.forecast_features = self.preprocess_data(dataframe)
 
-    def skip_samples(self, dataframe):
+    def skip_samples(self, dataframe, full_day_skip):
         nan_rows = dataframe[dataframe.isnull().any(axis=1)]
         nan_indices = nan_rows.index
         skip_indices = [
@@ -91,7 +98,7 @@ class NrvDataset(Dataset):
 
         # add indices that are not the start of a day (00:15) to the skip indices (use datetime column)
         # get indices of all 00:15 timestamps
-        if self.full_day_skip:
+        if full_day_skip:
             start_of_day_indices = dataframe[
                 dataframe["datetime"].dt.time != pd.Timestamp("00:00:00").time()
             ].index

src/policies/PolicyEvaluator.py

@@ -0,0 +1,249 @@
+from clearml import Task
+from tqdm import tqdm
+from src.policies.simple_baseline import BaselinePolicy
+import pandas as pd
+import numpy as np
+import torch
+import plotly.express as px
+
+from src.utils.imbalance_price_calculator import ImbalancePriceCalculator
+
+
+class PolicyEvaluator:
+    def __init__(self, baseline_policy: BaselinePolicy, task: Task = None):
+        self.baseline_policy = baseline_policy
+
+        self.ipc = ImbalancePriceCalculator(data_path="")
+        self.dates = baseline_policy.test_data["DateTime"].dt.date.unique()
+        self.dates = pd.to_datetime(self.dates)
+
+        ### Load Imbalance Prices ###
+        imbalance_prices = pd.read_csv("data/imbalance_prices.csv", sep=";")
+        imbalance_prices["DateTime"] = pd.to_datetime(
+            imbalance_prices["DateTime"], utc=True
+        )
+        self.imbalance_prices = imbalance_prices.sort_values(by=["DateTime"])
+
+        self.penalties = [0, 100, 300, 500, 800, 1000, 1500]
+        self.profits = []
+
+        self.task = task
+
+    def get_imbanlance_prices_for_date(self, date):
+        imbalance_prices_day = self.imbalance_prices[
+            self.imbalance_prices["DateTime"].dt.date == date
+        ]
+        return imbalance_prices_day["Positive imbalance price"].values
+
+    def evaluate_for_date(
+        self,
+        date,
+        idx_samples,
+        test_loader,
+        charge_thresholds=np.arange(-100, 250, 25),
+        discharge_thresholds=np.arange(-100, 250, 25),
+    ):
+        idx = test_loader.dataset.get_idx_for_date(date.date())
+
+        print("Evaluated for idx: ", idx)
+        (initial, samples) = idx_samples[idx]
+
+        if len(initial.shape) == 2:
+            initial = initial.cpu().numpy()[0][-1]
+        else:
+            initial = initial.cpu().numpy()[-1]
+        samples = samples.cpu().numpy()
+
+        initial = np.repeat(initial, samples.shape[0])
+        combined = np.concatenate((initial.reshape(-1, 1), samples), axis=1)
+
+        reconstructed_imbalance_prices = (
+            self.ipc.get_imbalance_prices_2023_for_date_vectorized(date, combined)
+        )
+        reconstructed_imbalance_prices = torch.tensor(
+            reconstructed_imbalance_prices, device="cuda"
+        )
+
+        real_imbalance_prices = self.get_imbanlance_prices_for_date(date.date())
+
+        for penalty in self.penalties:
+            found_charge_thresholds, found_discharge_thresholds = (
+                self.baseline_policy.get_optimal_thresholds(
+                    reconstructed_imbalance_prices,
+                    charge_thresholds,
+                    discharge_thresholds,
+                    penalty,
+                )
+            )
+
+            predicted_charge_threshold = found_charge_thresholds.mean(axis=0)
+            predicted_discharge_threshold = found_discharge_thresholds.mean(axis=0)
+
+            ### Determine Profits and Charge Cycles ###
+            simulated_profit, simulated_charge_cycles = self.baseline_policy.simulate(
+                torch.tensor([[real_imbalance_prices]]),
+                torch.tensor([predicted_charge_threshold]),
+                torch.tensor([predicted_discharge_threshold]),
+            )
+            self.profits.append(
+                [
+                    date,
+                    penalty,
+                    simulated_profit[0][0].item(),
+                    simulated_charge_cycles[0][0].item(),
+                    predicted_charge_threshold.item(),
+                    predicted_discharge_threshold.item(),
+                ]
+            )
+
+    def evaluate_test_set(self, idx_samples, test_loader):
+        self.profits = []
+        try:
+            for date in tqdm(self.dates):
+                self.evaluate_for_date(date, idx_samples, test_loader)
+        except KeyboardInterrupt:
+            print("Interrupted")
+            raise KeyboardInterrupt
+
+        except Exception as e:
+            print(e)
+            pass
+
+        self.profits = pd.DataFrame(
+            self.profits,
+            columns=[
+                "Date",
+                "Penalty",
+                "Profit",
+                "Charge Cycles",
+                "Charge Threshold",
+                "Discharge Threshold",
+            ],
+        )
+
+        print("Profits calculated")
+        print(self.profits.head())
+
+    def plot_profits_table(self):
+        # Check if task or penalties are not set
+        if (
+            self.task is None
+            or not hasattr(self, "penalties")
+            or not hasattr(self, "profits")
+        ):
+            print("Task, penalties, or profits not defined.")
+            return
+
+        if self.profits.empty:
+            print("Profits DataFrame is empty.")
+            return
+
+        # Aggregate profits and charge cycles by penalty, calculating totals and per-year values
+        aggregated = self.profits.groupby("Penalty").agg(
+            Total_Profit=("Profit", "sum"),
+            Total_Charge_Cycles=("Charge Cycles", "sum"),
+            Num_Days=("Date", "nunique"),
+        )
+        aggregated["Profit_Per_Year"] = (
+            aggregated["Total_Profit"] / aggregated["Num_Days"] * 365
+        )
+        aggregated["Charge_Cycles_Per_Year"] = (
+            aggregated["Total_Charge_Cycles"] / aggregated["Num_Days"] * 365
+        )
+
+        # Reset index to make 'Penalty' a column again and drop unnecessary columns
+        final_df = aggregated.reset_index().drop(
+            columns=["Total_Profit", "Total_Charge_Cycles", "Num_Days"]
+        )
+
+        # Rename columns to match expected output
+        final_df.columns = ["Penalty", "Total Profit", "Total Charge Cycles"]
+
+        # Profits till 400
+        profits_till_400 = self.get_profits_till_400()
+
+        # aggregate the final_df and profits_till_400 with columns: Penalty, total profit, total charge cycles, profit till 400, total charge cycles
+        final_df = final_df.merge(profits_till_400, on="Penalty")
+
+        # Log the final results table
+        self.task.get_logger().report_table(
+            "Policy Results", "Policy Results", iteration=0, table_plot=final_df
+        )
+
+    def plot_thresholds_per_day(self):
+        if self.task is None:
+            return
+
+        fig = px.line(
+            self.profits[self.profits["Penalty"] == 0],
+            x="Date",
+            y=["Charge Threshold", "Discharge Threshold"],
+            title="Charge and Discharge Thresholds per Day",
+        )
+
+        fig.update_layout(
+            width=1000,
+            height=600,
+            title_x=0.5,
+        )
+
+        self.task.get_logger().report_plotly(
+            "Thresholds per Day", "Thresholds per Day", iteration=0, figure=fig
+        )
+
+    def get_profits_as_scalars(self):
+        aggregated = self.profits.groupby("Penalty").agg(
+            Total_Profit=("Profit", "sum"),
+            Total_Charge_Cycles=("Charge Cycles", "sum"),
+            Num_Days=("Date", "nunique"),
+        )
+        aggregated["Profit_Per_Year"] = (
+            aggregated["Total_Profit"] / aggregated["Num_Days"] * 365
+        )
+        aggregated["Charge_Cycles_Per_Year"] = (
+            aggregated["Total_Charge_Cycles"] / aggregated["Num_Days"] * 365
+        )
+
+        # Reset index to make 'Penalty' a column again and drop unnecessary columns
+        final_df = aggregated.reset_index().drop(
+            columns=["Total_Profit", "Total_Charge_Cycles", "Num_Days"]
+        )
+
+        # Rename columns to match expected output
+        final_df.columns = ["Penalty", "Total Profit", "Total Charge Cycles"]
+        return final_df
+
+    def get_profits_till_400(self):
+        # calculates profits until 400 charge cycles per year are reached
+        number_of_days = len(self.profits["Date"].unique())
+        usable_charge_cycles = (400 / 365) * number_of_days
+
+        # now sum the profit until the usable charge cycles are reached
+        penalty_profits = {}
+        penalty_charge_cycles = {}
+
+        for index, row in self.profits.iterrows():
+            penalty = row["Penalty"]
+            profit = row["Profit"]
+            charge_cycles = row["Charge Cycles"]
+
+            if penalty not in penalty_profits:
+                penalty_profits[penalty] = 0
+                penalty_charge_cycles[penalty] = 0
+
+            if penalty_charge_cycles[penalty] < usable_charge_cycles:
+                penalty_profits[penalty] += profit
+                penalty_charge_cycles[penalty] += charge_cycles
+
+        df = pd.DataFrame(
+            list(
+                zip(
+                    penalty_profits.keys(),
+                    penalty_profits.values(),
+                    penalty_charge_cycles.values(),
+                )
+            ),
+            columns=["Penalty", "Profit_till_400", "Cycles_till_400"],
+        )
+
+        return df

src/policies/baselines/BaselinePolicyEvaluator.py

@@ -0,0 +1,73 @@
+from clearml import Task
+from src.policies.simple_baseline import BaselinePolicy
+from src.policies.PolicyEvaluator import PolicyEvaluator
+import numpy as np
+import pandas as pd
+from tqdm import tqdm
+import torch
+
+
+class BaselinePolicyEvaluator(PolicyEvaluator):
+    def __init__(self, baseline_policy: BaselinePolicy, task: Task = None):
+        super(baseline_policy, task)
+        self.dates = baseline_policy.train_data["DateTime"].dt.date.unique()
+        self.dates = pd.to_datetime(self.dates)
+        self.penalties = [0, 100, 300, 500, 800, 1000, 1500]
+        self.profits = []
+
+    def determine_thresholds_for_date(self, date):
+        charge_thresholds = np.arange(-100, 250, 25)
+        discharge_thresholds = np.arange(-100, 250, 25)
+
+        real_imbalance_prices = self.get_imbanlance_prices_for_date(date.date())
+
+        for penalty in self.penalties:
+            found_charge_thresholds, found_discharge_thresholds = (
+                self.baseline_policy.get_optimal_thresholds(
+                    torch.tensor([real_imbalance_prices]),
+                    charge_thresholds,
+                    discharge_thresholds,
+                    penalty,
+                )
+            )
+
+            best_charge_threshold = found_charge_thresholds.item()
+            best_discharge_threshold = found_discharge_thresholds.item()
+
+            simulated_profit, simulated_charge_cycles = self.baseline_policy.simulate(
+                torch.tensor([[real_imbalance_prices]]),
+                torch.tensor([best_charge_threshold]),
+                torch.tensor([best_discharge_threshold]),
+            )
+
+            self.profits.append(
+                [
+                    date,
+                    penalty,
+                    simulated_profit[0][0].item(),
+                    simulated_charge_cycles[0][0].item(),
+                    best_charge_threshold.item(),
+                    best_discharge_threshold.item(),
+                ]
+            )
+
+    def determine_best_thresholds(self):
+        self.profits = []
+        try:
+            for date in tqdm(self.dates):
+                self.determine_thresholds_for_date(date)
+        except Exception as e:
+            print(e)
+            pass
+
+        self.profits = pd.DataFrame(
+            self.profits,
+            columns=[
+                "Date",
+                "Penalty",
+                "Profit",
+                "Charge Cycles",
+                "Charge Threshold",
+                "Discharge Threshold",
+            ],
+        )

src/policies/baselines/global_threshold_baseline.py

@@ -0,0 +1,16 @@
+from src.utils.clearml import ClearMLHelper
+
+#### ClearML ####
+clearml_helper = ClearMLHelper(project_name="Thesis/NrvForecast")
+task = clearml_helper.get_task(task_name="Global Thresholds Baselien")
+task.execute_remotely(queue_name="default", exit_process=True)
+
+from src.policies.baselines.BaselinePolicyEvaluator import BaselinePolicyEvaluator
+from src.policies.simple_baseline import BaselinePolicy, Battery
+
+### Policy Evaluator ###
+battery = Battery(2, 1)
+baseline_policy = BaselinePolicy(battery, data_path="")
+policy_evaluator = BaselinePolicyEvaluator(baseline_policy, task)
+
+policy_evaluator.determine_best_thresholds()

src/policies/policy_executer.py

@@ -8,7 +8,8 @@ import pandas as pd
 import datetime
 from tqdm import tqdm
 from src.utils.imbalance_price_calculator import ImbalancePriceCalculator
-import time
+import seaborn as sns
+import matplotlib.pyplot as plt
 import plotly.express as px
 
 ### import functions ###
@@ -16,7 +17,7 @@ from src.trainers.quantile_trainer import auto_regressive as quantile_auto_regre
 from src.trainers.diffusion_trainer import sample_diffusion
 from src.utils.clearml import ClearMLHelper
 
-# argparse to parse task id and model type
+### Arguments ###
 parser = argparse.ArgumentParser()
 parser.add_argument('--task_id', type=str, default=None)
 parser.add_argument('--model_type', type=str, default=None)
@@ -27,6 +28,7 @@ assert args.task_id is not None, "Please specify task id"
 assert args.model_type is not None, "Please specify model type"
 assert args.model_name is not None, "Please specify model name"
 
+### Baseline Policy ###
 battery = Battery(2, 1)
 baseline_policy = BaselinePolicy(battery, data_path="")
 
@@ -163,20 +165,17 @@ def get_next_day_profits_for_date(model, data_processor, test_loader, date, ipc,
     return predicted_nrv_profits_cycles, baseline_profits_cycles, _charge_thresholds, _discharge_thresholds
 
 def next_day_test_set(model, data_processor, test_loader, ipc, predict_NRV: callable):
-    penalties = [0, 10, 50, 150, 300, 500, 600, 800, 1000, 1500, 2000, 2500]
+    penalties = [0, 50, 250, 500, 1000, 1500]
     predicted_nrv_profits_cycles = {i: [0, 0] for i in penalties}
     baseline_profits_cycles = {i: [0, 0] for i in penalties}
 
     charge_thresholds = {}
     discharge_thresholds = {}
 
-    # get all dates in test set
     dates = baseline_policy.test_data["DateTime"].dt.date.unique()
-
-    # dates back to datetime
     dates = pd.to_datetime(dates)
 
-    for date in tqdm(dates[:10]):
+    for date in tqdm(dates):
         try:
             new_predicted_nrv_profits_cycles, new_baseline_profits_cycles, new_charge_thresholds, new_discharge_thresholds = get_next_day_profits_for_date(model, data_processor, test_loader, date, ipc, predict_NRV, penalties)
             
@@ -191,8 +190,7 @@ def next_day_test_set(model, data_processor, test_loader, ipc, predict_NRV: call
                 baseline_profits_cycles[penalty][1] += new_baseline_profits_cycles[penalty][1]
 
         except Exception as e:
-            # print(f"Error for date {date}")
+            print(f"Error for date {date}")
-            raise e
 
     return predicted_nrv_profits_cycles, baseline_profits_cycles, charge_thresholds, discharge_thresholds
 
@@ -222,9 +220,6 @@ def main():
     # the charge_thresholds is a dictionary with date as key. The values of the dictionary is another dictionary with keys as penalties and values as the charge thresholds
     # create density plot that shows a density plot of the charge thresholds for each penalty (use seaborn displot) (One plot with a different color for each penalty)
 
-    import seaborn as sns
-    import matplotlib.pyplot as plt
-
     charge_thresholds_for_penalty = {}
     for d in charge_thresholds.values():
         for penalty, thresholds in d.items():
@@ -239,47 +234,73 @@ def main():
                 discharge_thresholds_for_penalty[penalty] = []
             discharge_thresholds_for_penalty[penalty].extend(thresholds)
 
+    def plot_threshold_distribution(thresholds: dict, title: str):
+        data_to_plot = []
+        for penalty, values in thresholds.items():
+            for value in values:
+                data_to_plot.append({'Penalty': penalty, 'Value': value.item()})
+        df = pd.DataFrame(data_to_plot)
+        palette = sns.color_palette("bright", len(thresholds.keys()))
+        fig = sns.displot(data=df, x="Value", hue="Penalty", kind="kde", palette=palette)
+        plt.title('Density of Charge Thresholds by Penalty')
+        plt.xlabel('Charge Threshold')
+        plt.ylabel('Density')
+        plt.legend(title='Penalty')
+        task.get_logger().report_matplotlib_figure(
+            "Policy Results", 
+            title, 
+            iteration=0, 
+            figure=fig
+        )
+        plt.close()
+
     ### Plot charge thresholds distribution ###
-    data_to_plot = []
+    plot_threshold_distribution(charge_thresholds_for_penalty, "Charge Thresholds")
-    for penalty, values in charge_thresholds_for_penalty.items():
-        for value in values:
-            data_to_plot.append({'Penalty': penalty, 'Value': value.item()})
-    df = pd.DataFrame(data_to_plot)
-    print(df.head())
-    palette = sns.color_palette("bright", len(charge_thresholds.keys()))
-    fig = sns.displot(data=df, x="Value", hue="Penalty", kind="kde", palette=palette)
-    plt.title('Density of Charge Thresholds by Penalty')
-    plt.xlabel('Charge Threshold')
-    plt.ylabel('Density')
-    plt.legend(title='Penalty')
-    task.get_logger().report_matplotlib_figure(
-        "Policy Results", 
-        "Charge Thresholds", 
-        iteration=0, 
-        figure=fig
-    )
-    plt.close()
 
     ### Plot discharge thresholds distribution ###
-    data_to_plot = []
+    plot_threshold_distribution(discharge_thresholds_for_penalty, "Discharge Thresholds")
-    for penalty, values in discharge_thresholds_for_penalty.items():
-        for value in values:
-            data_to_plot.append({'Penalty': penalty, 'Value': value.item()})
-    df = pd.DataFrame(data_to_plot)
-    palette = sns.color_palette("bright", len(discharge_thresholds.keys()))
-    fig = sns.displot(data=df, x="Value", hue="Penalty", kind="kde", palette=palette)
-    plt.title('Density of Charge Thresholds by Penalty')
-    plt.xlabel('Charge Threshold')
-    plt.ylabel('Density')
-    plt.legend(title='Penalty')
-    task.get_logger().report_matplotlib_figure(
-        "Policy Results", 
-        "Discharge Thresholds", 
-        iteration=0, 
-        figure=fig
-    )
-    plt.close()
 
+    def plot_thresholds_per_day(thresholds: dict, title: str):
+        # plot mean charge threshold per day (per penalty (other color))
+        data_to_plot = []
+        for date, values in thresholds.items():
+            for penalty, value in values.items():
+                mean_val = value.mean().item()
+                std_val = value.std().item()  # Calculate standard deviation
+                data_to_plot.append({'Date': date, 'Penalty': penalty, 'Mean': mean_val, 'StdDev': std_val})
+                print(f"Date: {date}, Penalty: {penalty}, Mean: {mean_val}, StdDev: {std_val}")
+        df = pd.DataFrame(data_to_plot)
+        df["Date"] = pd.to_datetime(df["Date"])
+
+        fig = px.line(
+            df,
+            x="Date",
+            y="Mean",
+            color="Penalty",
+            title=title,
+            labels={"Mean": "Threshold", "Date": "Date"},
+            markers=True,  # Adds markers to the lines
+            hover_data=["Penalty"],  # Adds additional hover information
+        )
+
+        fig.update_layout(
+            width=1000,  # Set the width of the figure
+            height=600,  # Set the height of the figure
+            title_x=0.5,  # Center the title horizontally
+        )
+
+        task.get_logger().report_plotly(
+            "Thresholds per Day", 
+            title, 
+            iteration=0, 
+            figure=fig
+        )
+
+    ### Plot mean charge thresholds per day ###
+    plot_thresholds_per_day(charge_thresholds, "Mean Charge Thresholds per Day")
+
+    ### Plot mean discharge thresholds per day ###
+    plot_thresholds_per_day(discharge_thresholds, "Mean Discharge Thresholds per Day")
 
 
     # create dataframe with columns "name", "penalty", "profit", "cycles"

src/trainers/autoregressive_trainer.py

@@ -33,67 +33,29 @@ class AutoRegressiveTrainer(Trainer):
         self.model.output_size = 1
 
     def debug_plots(self, task, train: bool, data_loader, sample_indices, epoch):
-        num_samples = len(sample_indices)
+        for actual_idx, idx in sample_indices.items():
-        rows = num_samples  # One row per sample since we only want one column
+            auto_regressive_output = self.auto_regressive(data_loader.dataset, [idx]*1000)
-
-        #  check if self has get_plot_error
-        if hasattr(self, "get_plot_error"):
-            cols = 2
-            print("Using get_plot_error")
-        else:
-            cols = 1
-            print("Using get_plot")
-
-        fig = make_subplots(
-            rows=rows,
-            cols=cols,
-            subplot_titles=[f"Sample {i+1}" for i in range(num_samples)],
-        )
-
-        for i, idx in enumerate(sample_indices):
-            auto_regressive_output = self.auto_regressive(data_loader.dataset, [idx])
             if len(auto_regressive_output) == 3:
                 initial, predictions, target = auto_regressive_output
             else:
-                initial, predictions, _, target = auto_regressive_output
+                initial, _, predictions, target = auto_regressive_output
 
-            initial = initial.squeeze(0)
+            
-            predictions = predictions.squeeze(0)
+            # keep one initial
-            target = target.squeeze(0)
+            initial = initial[0]
+            target = target[0]
 
-            sub_fig = self.get_plot(initial, target, predictions, show_legend=(i == 0))
+            predictions = predictions
 
-            row = i + 1
+            fig = self.get_plot(initial, target, predictions, show_legend=(0 == 0))
-            col = 1
 
-            for trace in sub_fig.data:
+            task.get_logger().report_matplotlib_figure(
-                fig.add_trace(trace, row=row, col=col)
+                title="Training" if train else "Testing",
-
+                series=f'Sample {actual_idx}',
-            if cols == 2:
+                iteration=epoch,
-                error_sub_fig = self.get_plot_error(
+                figure=fig,
-                    target, predictions
-                )
-                for trace in error_sub_fig.data:
-                    fig.add_trace(trace, row=row, col=col + 1)
-
-            loss = self.criterion(
-                predictions.to(self.device), target.to(self.device)
-            ).item()
-
-            fig["layout"]["annotations"][i].update(
-                text=f"{self.criterion.__class__.__name__}: {loss:.6f}"
             )
 
-        # y axis same for all plots
-        # fig.update_yaxes(range=[-1, 1], col=1)
-
-        fig.update_layout(height=1000 * rows)
-        task.get_logger().report_plotly(
-            title=f"{'Training' if train else 'Test'} Samples",
-            series="full_day",
-            iteration=epoch,
-            figure=fig,
-        )
 
     def auto_regressive(self, data_loader, idx, sequence_length: int = 96):
         self.model.eval()

src/trainers/diffusion_trainer.py

@@ -1,6 +1,7 @@
 from clearml import Task
 import torch
 import torch.nn as nn
+from src.policies.PolicyEvaluator import PolicyEvaluator
 from torchinfo import summary
 from src.losses.crps_metric import crps_from_samples
 from src.data.preprocessing import DataProcessor
@@ -13,14 +14,20 @@ import seaborn as sns
 import matplotlib.patches as mpatches
 
 
-def sample_diffusion(model: DiffusionModel, n: int, inputs: torch.tensor, noise_steps=1000, beta_start=1e-4, beta_end=0.02, ts_length=96):
+def sample_diffusion(
+    model: DiffusionModel,
+    n: int,
+    inputs: torch.tensor,
+    noise_steps=1000,
+    beta_start=1e-4,
+    beta_end=0.02,
+    ts_length=96,
+):
     device = next(model.parameters()).device
     beta = torch.linspace(beta_start, beta_end, noise_steps).to(device)
-    alpha = 1. - beta
+    alpha = 1.0 - beta
     alpha_hat = torch.cumprod(alpha, dim=0)
 
-    # inputs: (num_features) -> (batch_size, num_features)
-    # inputs: (time_steps, num_features) -> (batch_size, time_steps, num_features)
     if len(inputs.shape) == 2:
         inputs = inputs.repeat(n, 1)
     elif len(inputs.shape) == 3:
@@ -41,53 +48,74 @@ def sample_diffusion(model: DiffusionModel, n: int, inputs: torch.tensor, noise_
             else:
                 noise = torch.zeros_like(x)
 
-            x = 1/torch.sqrt(_alpha) * (x-((1-_alpha) / (torch.sqrt(1 - _alpha_hat))) * predicted_noise) + torch.sqrt(_beta) * noise
+            x = (
+                1
+                / torch.sqrt(_alpha)
+                * (x - ((1 - _alpha) / (torch.sqrt(1 - _alpha_hat))) * predicted_noise)
+                + torch.sqrt(_beta) * noise
+            )
+    x = torch.clamp(x, -1.0, 1.0)
     return x
 
 
-
 class DiffusionTrainer:
-    def __init__(self, model: nn.Module, data_processor: DataProcessor, device: torch.device):
+    def __init__(
+        self,
+        model: nn.Module,
+        data_processor: DataProcessor,
+        device: torch.device,
+        policy_evaluator: PolicyEvaluator = None,
+    ):
         self.model = model
         self.device = device
 
-        self.noise_steps = 20
+        self.noise_steps = 30
-        self.beta_start = 1e-4
+        self.beta_start = 0.0001
         self.beta_end = 0.02
         self.ts_length = 96
-        
+
         self.data_processor = data_processor
 
-        self.beta = torch.linspace(self.beta_start, self.beta_end, self.noise_steps).to(self.device)
+        self.beta = torch.linspace(self.beta_start, self.beta_end, self.noise_steps).to(
-        self.alpha = 1. - self.beta
+            self.device
+        )
+        self.alpha = 1.0 - self.beta
         self.alpha_hat = torch.cumprod(self.alpha, dim=0)
 
         self.best_score = None
+        self.policy_evaluator = policy_evaluator
 
     def noise_time_series(self, x: torch.tensor, t: int):
-        """ Add noise to time series
+        """Add noise to time series
         Args:
             x (torch.tensor): shape (batch_size, time_steps)
             t (int): index of time step
         """
         sqrt_alpha_hat = torch.sqrt(self.alpha_hat[t])[:, None]
-        sqrt_one_minus_alpha_hat = torch.sqrt(1. - self.alpha_hat[t])[:, None]
+        sqrt_one_minus_alpha_hat = torch.sqrt(1.0 - self.alpha_hat[t])[:, None]
         noise = torch.randn_like(x)
         return sqrt_alpha_hat * x + sqrt_one_minus_alpha_hat * noise, noise
-    
+
     def sample_timesteps(self, n: int):
-        """ Sample timesteps for noise
+        """Sample timesteps for noise
         Args:
             n (int): number of samples
         """
         return torch.randint(low=1, high=self.noise_steps, size=(n,))
 
-    
     def sample(self, model: DiffusionModel, n: int, inputs: torch.tensor):
-        x = sample_diffusion(model, n, inputs, self.noise_steps, self.beta_start, self.beta_end, self.ts_length)
+        x = sample_diffusion(
+            model,
+            n,
+            inputs,
+            self.noise_steps,
+            self.beta_start,
+            self.beta_end,
+            self.ts_length,
+        )
         model.train()
         return x
-    
+
     def random_samples(self, train: bool = True, num_samples: int = 10):
         train_loader, test_loader = self.data_processor.get_dataloaders(
             predict_sequence_length=96
@@ -98,9 +126,20 @@ class DiffusionTrainer:
         else:
             loader = test_loader
 
-        indices = np.random.randint(0, len(loader.dataset) - 1, size=num_samples)
+        # set seed
+        np.random.seed(42)
+
+        actual_indices = np.random.choice(
+            loader.dataset.full_day_valid_indices, num_samples, replace=False
+        )
+        indices = {}
+        for i in actual_indices:
+            indices[i] = loader.dataset.valid_indices.index(i)
+
+        print(actual_indices)
+
         return indices
-    
+
     def init_clearml_task(self, task):
         task.add_tags(self.model.__class__.__name__)
         task.add_tags(self.__class__.__name__)
@@ -110,13 +149,24 @@ class DiffusionTrainer:
 
         if self.data_processor.lstm:
             inputDim = self.data_processor.get_input_size()
-            other_input_data = torch.randn(1024, inputDim[1], self.model.other_inputs_dim).to(self.device)
+            other_input_data = torch.randn(
+                1024, inputDim[1], self.model.other_inputs_dim
+            ).to(self.device)
         else:
-            other_input_data = torch.randn(1024, self.model.other_inputs_dim).to(self.device)
+            other_input_data = torch.randn(1024, self.model.other_inputs_dim).to(
-        task.set_configuration_object("model", str(summary(self.model, input_data=[input_data, time_steps, other_input_data])))
+                self.device
+            )
+        task.set_configuration_object(
+            "model",
+            str(
+                summary(
+                    self.model, input_data=[input_data, time_steps, other_input_data]
+                )
+            ),
+        )
 
         self.data_processor = task.connect(self.data_processor, name="data_processor")
-        
+
     def train(self, epochs: int, learning_rate: float, task: Task = None):
         self.best_score = None
         optimizer = torch.optim.Adam(self.model.parameters(), lr=learning_rate)
@@ -150,7 +200,7 @@ class DiffusionTrainer:
                 optimizer.zero_grad()
                 loss.backward()
                 optimizer.step()
-            
+
             running_loss /= len(train_loader.dataset)
 
             if epoch % 40 == 0 and epoch != 0:
@@ -159,21 +209,24 @@ class DiffusionTrainer:
             if task:
                 task.get_logger().report_scalar(
                     title=criterion.__class__.__name__,
-                    series='train',
+                    series="train",
                     iteration=epoch,
                     value=loss.item(),
                 )
 
                 if epoch % 150 == 0 and epoch != 0:
-                    self.debug_plots(task, True, train_loader, train_sample_indices, epoch)
+                    self.debug_plots(
-                    self.debug_plots(task, False, test_loader, test_sample_indices, epoch)
+                        task, True, train_loader, train_sample_indices, epoch
+                    )
+                    self.debug_plots(
+                        task, False, test_loader, test_sample_indices, epoch
+                    )
 
         if task:
             task.close()
 
-        
     def debug_plots(self, task, training: bool, data_loader, sample_indices, epoch):
-        for i, idx in enumerate(sample_indices):
+        for actual_idx, idx in sample_indices.items():
             features, target, _ = data_loader.dataset[idx]
 
             features = features.to(self.device)
@@ -182,61 +235,115 @@ class DiffusionTrainer:
             self.model.eval()
             with torch.no_grad():
                 samples = self.sample(self.model, 100, features).cpu().numpy()
-               
+                samples = self.data_processor.inverse_transform(samples)
-            ci_99_upper = np.quantile(samples, 0.99, axis=0)
+                target = self.data_processor.inverse_transform(target)
-            ci_99_lower = np.quantile(samples, 0.01, axis=0)
 
-            ci_95_upper = np.quantile(samples, 0.95, axis=0)
+            ci_99_upper = np.quantile(samples, 0.995, axis=0)
-            ci_95_lower = np.quantile(samples, 0.05, axis=0)
+            ci_99_lower = np.quantile(samples, 0.005, axis=0)
 
-            ci_90_upper = np.quantile(samples, 0.9, axis=0)
+            ci_95_upper = np.quantile(samples, 0.975, axis=0)
-            ci_90_lower = np.quantile(samples, 0.1, axis=0)
+            ci_95_lower = np.quantile(samples, 0.025, axis=0)
 
-            ci_50_upper = np.quantile(samples, 0.5, axis=0)
+            ci_90_upper = np.quantile(samples, 0.95, axis=0)
-            ci_50_lower = np.quantile(samples, 0.5, axis=0)
+            ci_90_lower = np.quantile(samples, 0.05, axis=0)
+
+            ci_50_lower = np.quantile(samples, 0.25, axis=0)
+            ci_50_upper = np.quantile(samples, 0.75, axis=0)
 
             sns.set_theme()
             time_steps = np.arange(0, 96)
 
             fig, ax = plt.subplots(figsize=(20, 10))
-            ax.plot(time_steps, samples.mean(axis=0), label="Mean of NRV samples", linewidth=3)
+            ax.plot(
+                time_steps,
+                samples.mean(axis=0),
+                label="Mean of NRV samples",
+                linewidth=3,
+            )
             # ax.fill_between(time_steps, ci_lower, ci_upper, color='b', alpha=0.2, label='Full Interval')
 
-            ax.fill_between(time_steps, ci_99_lower, ci_99_upper, color='b', alpha=0.2, label='99% Interval')
+            ax.fill_between(
-            ax.fill_between(time_steps, ci_95_lower, ci_95_upper, color='b', alpha=0.2, label='95% Interval')
+                time_steps,
-            ax.fill_between(time_steps, ci_90_lower, ci_90_upper, color='b', alpha=0.2, label='90% Interval')
+                ci_99_lower,
-            ax.fill_between(time_steps, ci_50_lower, ci_50_upper, color='b', alpha=0.2, label='50% Interval')
+                ci_99_upper,
+                color="b",
+                alpha=0.2,
+                label="99% Interval",
+            )
+            ax.fill_between(
+                time_steps,
+                ci_95_lower,
+                ci_95_upper,
+                color="b",
+                alpha=0.2,
+                label="95% Interval",
+            )
+            ax.fill_between(
+                time_steps,
+                ci_90_lower,
+                ci_90_upper,
+                color="b",
+                alpha=0.2,
+                label="90% Interval",
+            )
+            ax.fill_between(
+                time_steps,
+                ci_50_lower,
+                ci_50_upper,
+                color="b",
+                alpha=0.2,
+                label="50% Interval",
+            )
 
             ax.plot(target, label="Real NRV", linewidth=3)
             # full_interval_patch = mpatches.Patch(color='b', alpha=0.2, label='Full Interval')
-            ci_99_patch = mpatches.Patch(color='b', alpha=0.3, label='99% Interval')
+            ci_99_patch = mpatches.Patch(color="b", alpha=0.3, label="99% Interval")
-            ci_95_patch = mpatches.Patch(color='b', alpha=0.4, label='95% Interval')
+            ci_95_patch = mpatches.Patch(color="b", alpha=0.4, label="95% Interval")
-            ci_90_patch = mpatches.Patch(color='b', alpha=0.5, label='90% Interval')
+            ci_90_patch = mpatches.Patch(color="b", alpha=0.5, label="90% Interval")
-            ci_50_patch = mpatches.Patch(color='b', alpha=0.6, label='50% Interval')
+            ci_50_patch = mpatches.Patch(color="b", alpha=0.6, label="50% Interval")
 
-
+            ax.legend(
-            ax.legend(handles=[ci_99_patch, ci_95_patch, ci_90_patch, ci_50_patch, ax.lines[0], ax.lines[1]])
+                handles=[
+                    ci_99_patch,
+                    ci_95_patch,
+                    ci_90_patch,
+                    ci_50_patch,
+                    ax.lines[0],
+                    ax.lines[1],
+                ]
+            )
 
             task.get_logger().report_matplotlib_figure(
                 title="Training" if training else "Testing",
-                series=f'Sample {i}',
+                series=f"Sample {actual_idx}",
                 iteration=epoch,
                 figure=fig,
             )
 
             plt.close()
 
-    def test(self, data_loader: torch.utils.data.DataLoader, epoch: int, task: Task = None):
+    def test(
+        self, data_loader: torch.utils.data.DataLoader, epoch: int, task: Task = None
+    ):
         all_crps = []
-        for inputs, targets, _ in data_loader:
+        generated_samples = {}
+        for inputs, targets, idx_batch in data_loader:
             inputs, targets = inputs.to(self.device), targets.to(self.device)
-            
+            print(inputs.shape, targets.shape)
+
             number_of_samples = 100
             sample = self.sample(self.model, number_of_samples, inputs)
 
             # reduce samples from (batch_size*number_of_samples, time_steps) to (batch_size, number_of_samples, time_steps)
             samples_batched = sample.reshape(inputs.shape[0], number_of_samples, 96)
 
+            # add samples to generated_samples generated_samples[idx.item()] = (initial, samples)
+            for i, (idx, samples) in enumerate(zip(idx_batch, samples_batched)):
+                generated_samples[idx.item()] = (
+                    self.data_processor.inverse_transform(inputs[i][:96]),
+                    self.data_processor.inverse_transform(samples),
+                )
+
             # calculate crps
             crps = crps_from_samples(samples_batched, targets)
             crps_mean = crps.mean(axis=1)
@@ -252,16 +359,38 @@ class DiffusionTrainer:
 
         if task:
             task.get_logger().report_scalar(
-                title="CRPS",
+                title="CRPS", series="test", value=mean_crps, iteration=epoch
-                series='test',
-                value=mean_crps,
-                iteration=epoch
             )
 
+        if self.policy_evaluator:
+            _, test_loader = self.data_processor.get_dataloaders(
+                predict_sequence_length=self.ts_length, full_day_skip=True
+            )
+
+            self.policy_evaluator.evaluate_test_set(generated_samples, test_loader)
+
+            df = self.policy_evaluator.get_profits_as_scalars()
+
+            for idx, row in df.iterrows():
+                task.get_logger().report_scalar(
+                    title="Profit",
+                    series=f"penalty_{row['Penalty']}",
+                    value=row["Total Profit"],
+                    iteration=epoch,
+                )
+
+            df = self.policy_evaluator.get_profits_till_400()
+            for idx, row in df.iterrows():
+                task.get_logger().report_scalar(
+                    title="Profit_till_400",
+                    series=f"penalty_{row['Penalty']}",
+                    value=row["Profit_till_400"],
+                    iteration=epoch,
+                )
+
     def save_checkpoint(self, val_loss, task, iteration: int):
         torch.save(self.model, "checkpoint.pt")
         task.update_output_model(
             model_path="checkpoint.pt", iteration=iteration, auto_delete_file=False
         )
         self.best_score = val_loss
-        

src/trainers/quantile_trainer.py

@@ -1,5 +1,6 @@
 import torch
 from tqdm import tqdm
+from src.policies.PolicyEvaluator import PolicyEvaluator
 from src.losses.crps_metric import crps_from_samples
 from src.trainers.trainer import Trainer
 from src.trainers.autoregressive_trainer import AutoRegressiveTrainer
@@ -10,6 +11,9 @@ import plotly.graph_objects as go
 import numpy as np
 import matplotlib.pyplot as plt
 from scipy.interpolate import CubicSpline
+import matplotlib.pyplot as plt
+import seaborn as sns
+import matplotlib.patches as mpatches
 
 
 def sample_from_dist(quantiles, preds):
@@ -28,10 +32,11 @@ def sample_from_dist(quantiles, preds):
 
     # random probabilities of (1000, 1)
     import random
+
     probs = np.array([random.random() for _ in range(1000)])
 
     spline = CubicSpline(quantiles, preds, axis=1)
-    
+
     samples = spline(probs)
 
     # get the diagonal
@@ -39,6 +44,7 @@ def sample_from_dist(quantiles, preds):
 
     return samples
 
+
 def auto_regressive(dataset, model, quantiles, idx_batch, sequence_length: int = 96):
     device = next(model.parameters()).device
     prev_features, targets = dataset.get_batch(idx_batch)
@@ -62,7 +68,7 @@ def auto_regressive(dataset, model, quantiles, idx_batch, sequence_length: int =
     predictions_full = new_predictions_full.unsqueeze(1)
 
     for i in range(sequence_length - 1):
-        if len(list(prev_features.shape)) == 2: 
+        if len(list(prev_features.shape)) == 2:
             new_features = torch.cat(
                 (prev_features[:, 1:96], samples), dim=1
             )  # (batch_size, 96)
@@ -99,9 +105,7 @@ def auto_regressive(dataset, model, quantiles, idx_batch, sequence_length: int =
         )  # (batch_size, sequence_length)
 
         with torch.no_grad():
-            new_predictions_full = model(
+            new_predictions_full = model(prev_features)  # (batch_size, quantiles)
-                prev_features
-            )  # (batch_size, quantiles)
         predictions_full = torch.cat(
             (predictions_full, new_predictions_full.unsqueeze(1)), dim=1
         )  # (batch_size, sequence_length, quantiles)
@@ -120,6 +124,7 @@ def auto_regressive(dataset, model, quantiles, idx_batch, sequence_length: int =
         target_full.unsqueeze(-1),
     )
 
+
 class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
     def __init__(
         self,
@@ -129,10 +134,13 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
         data_processor: DataProcessor,
         quantiles: list,
         device: torch.device,
+        policy_evaluator: PolicyEvaluator = None,
         debug: bool = True,
     ):
 
         self.quantiles = quantiles
+        self.test_set_samples = {}
+        self.policy_evaluator = policy_evaluator
 
         criterion = PinballLoss(quantiles=quantiles)
         super().__init__(
@@ -147,6 +155,7 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
 
     def calculate_crps_from_samples(self, task, dataloader, epoch: int):
         crps_from_samples_metric = []
+        generated_samples = {}
 
         with torch.no_grad():
             total_samples = len(dataloader.dataset) - 96
@@ -155,24 +164,59 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
 
                 if len(idx_batch) == 0:
                     continue
-                
+
                 for idx in tqdm(idx_batch):
                     computed_idx_batch = [idx] * 100
-                    _, _, samples, targets = self.auto_regressive(
+                    initial, _, samples, targets = self.auto_regressive(
                         dataloader.dataset, idx_batch=computed_idx_batch
                     )
+
+                    generated_samples[idx.item()] = (
+                        self.data_processor.inverse_transform(initial),
+                        self.data_processor.inverse_transform(samples),
+                    )
+
                     samples = samples.unsqueeze(0)
                     targets = targets.squeeze(-1)
                     targets = targets[0].unsqueeze(0)
-                    
+
                     crps = crps_from_samples(samples, targets)
 
                     crps_from_samples_metric.append(crps[0].mean().item())
 
         task.get_logger().report_scalar(
-            title="CRPS_from_samples", series="test", value=np.mean(crps_from_samples_metric), iteration=epoch
+            title="CRPS_from_samples",
+            series="test",
+            value=np.mean(crps_from_samples_metric),
+            iteration=epoch,
         )
 
+        # using the policy evaluator, evaluate the policy with the generated samples
+        if self.policy_evaluator is not None:
+            _, test_loader = self.data_processor.get_dataloaders(
+                predict_sequence_length=self.model.output_size, full_day_skip=True
+            )
+            self.policy_evaluator.evaluate_test_set(generated_samples, test_loader)
+            df = self.policy_evaluator.get_profits_as_scalars()
+
+            # for each row, report the profits
+            for idx, row in df.iterrows():
+                task.get_logger().report_scalar(
+                    title="Profit",
+                    series=f"penalty_{row['Penalty']}",
+                    value=row["Total Profit"],
+                    iteration=epoch,
+                )
+
+            df = self.policy_evaluator.get_profits_till_400()
+            for idx, row in df.iterrows():
+                task.get_logger().report_scalar(
+                    title="Profit_till_400",
+                    series=f"penalty_{row['Penalty']}",
+                    value=row["Profit_till_400"],
+                    iteration=epoch,
+                )
+
     def log_final_metrics(self, task, dataloader, train: bool = True):
         metrics = {metric.__class__.__name__: 0.0 for metric in self.metrics_to_track}
         transformed_metrics = {
@@ -192,19 +236,25 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
 
                 if train == False:
                     for idx in tqdm(idx_batch):
-                        computed_idx_batch = [idx] * 100
+                        computed_idx_batch = [idx] * 250
-                        _, outputs, samples, targets = self.auto_regressive(
+                        initial, outputs, samples, targets = self.auto_regressive(
                             dataloader.dataset, idx_batch=computed_idx_batch
                         )
+
+                        # save the samples for the idx, these will be used for evaluating the policy
+                        self.test_set_samples[idx.item()] = (
+                            self.data_processor.inverse_transform(initial),
+                            self.data_processor.inverse_transform(samples),
+                        )
+
                         samples = samples.unsqueeze(0)
                         targets = targets.squeeze(-1)
                         targets = targets[0].unsqueeze(0)
-                        
+
                         crps = crps_from_samples(samples, targets)
 
                         crps_from_samples_metric.append(crps[0].mean().item())
 
-
                 _, outputs, samples, targets = self.auto_regressive(
                     dataloader.dataset, idx_batch=idx_batch
                 )
@@ -258,39 +308,39 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
 
             if train == False:
                 task.get_logger().report_single_value(
-                    name="test_CRPS_from_samples_transformed", value=np.mean(crps_from_samples_metric)
+                    name="test_CRPS_from_samples_transformed",
+                    value=np.mean(crps_from_samples_metric),
                 )
 
-    def get_plot_error(
+    # def get_plot_error(
-        self,
+    #     self,
-        next_day,
+    #     next_day,
-        predictions,
+    #     predictions,
-    ):
+    # ):
-        metric = PinballLoss(quantiles=self.quantiles)
+    #     metric = PinballLoss(quantiles=self.quantiles)
-        fig = go.Figure()
+    #     fig = go.Figure()
 
-        next_day_np = next_day.view(-1).cpu().numpy()
+    #     next_day_np = next_day.view(-1).cpu().numpy()
-        predictions_np = predictions.cpu().numpy()
+    #     predictions_np = predictions.cpu().numpy()
 
-        if True:
+    #     if True:
-            next_day_np = self.data_processor.inverse_transform(next_day_np)
+    #         next_day_np = self.data_processor.inverse_transform(next_day_np)
-            predictions_np = self.data_processor.inverse_transform(predictions_np)
+    #         predictions_np = self.data_processor.inverse_transform(predictions_np)
 
-        # for each time step, calculate the error using the metric
+    #     # for each time step, calculate the error using the metric
-        errors = []
+    #     errors = []
-        for i in range(96):
+    #     for i in range(96):
 
-            target_tensor = torch.tensor(next_day_np[i]).unsqueeze(0)
+    #         target_tensor = torch.tensor(next_day_np[i]).unsqueeze(0)
-            prediction_tensor = torch.tensor(predictions_np[i]).unsqueeze(0)
+    #         prediction_tensor = torch.tensor(predictions_np[i]).unsqueeze(0)
 
-            errors.append(metric(prediction_tensor, target_tensor))
+    #         errors.append(metric(prediction_tensor, target_tensor))
 
-        # plot the error 
+    #     # plot the error
-        fig.add_trace(go.Scatter(x=np.arange(96), y=errors, name=metric.__class__.__name__))
+    #     fig.add_trace(go.Scatter(x=np.arange(96), y=errors, name=metric.__class__.__name__))
-        fig.update_layout(title=f"Error of {metric.__class__.__name__} for each time step")
+    #     fig.update_layout(title=f"Error of {metric.__class__.__name__} for each time step")
-
-        return fig
 
+    #     return fig
 
     def get_plot(
         self,
@@ -312,33 +362,114 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
             next_day_np = self.data_processor.inverse_transform(next_day_np)
             predictions_np = self.data_processor.inverse_transform(predictions_np)
 
+        ci_99_upper = np.quantile(predictions_np, 0.995, axis=0)
+        ci_99_lower = np.quantile(predictions_np, 0.005, axis=0)
+
+        ci_95_upper = np.quantile(predictions_np, 0.975, axis=0)
+        ci_95_lower = np.quantile(predictions_np, 0.025, axis=0)
+
+        ci_90_upper = np.quantile(predictions_np, 0.95, axis=0)
+        ci_90_lower = np.quantile(predictions_np, 0.05, axis=0)
+
+        ci_50_lower = np.quantile(predictions_np, 0.25, axis=0)
+        ci_50_upper = np.quantile(predictions_np, 0.75, axis=0)
+
         # Add traces for current and next day
-        fig.add_trace(go.Scatter(x=np.arange(96), y=current_day_np, name="Current Day"))
+        # fig.add_trace(go.Scatter(x=np.arange(96), y=current_day_np, name="Current Day"))
-        fig.add_trace(go.Scatter(x=96 + np.arange(96), y=next_day_np, name="Next Day"))
+        # fig.add_trace(go.Scatter(x=96 + np.arange(96), y=next_day_np, name="Next Day"))
 
-        for i, q in enumerate(self.quantiles):
+        # for i, q in enumerate(self.quantiles):
-            fig.add_trace(
+        #     fig.add_trace(
-                go.Scatter(
+        #         go.Scatter(
-                    x=96 + np.arange(96),
+        #             x=96 + np.arange(96),
-                    y=predictions_np[:, i],
+        #             y=predictions_np[:, i],
-                    name=f"Prediction (Q={q})",
+        #             name=f"Prediction (Q={q})",
-                    line=dict(dash="dash"),
+        #             line=dict(dash="dash"),
-                )
+        #         )
-            )
+        #     )
 
-        # Update the layout
+        # # Update the layout
-        fig.update_layout(
+        # fig.update_layout(
-            title="Predictions and Quantiles of the Linear Model",
+        #     title="Predictions and Quantiles of the Linear Model",
-            showlegend=show_legend,
+        #     showlegend=show_legend,
+        # )
+
+        sns.set_theme()
+        time_steps = np.arange(0, 96)
+
+        fig, ax = plt.subplots(figsize=(20, 10))
+        ax.plot(
+            time_steps,
+            predictions_np.mean(axis=0),
+            label="Mean of NRV samples",
+            linewidth=3,
+        )
+        # ax.fill_between(time_steps, ci_lower, ci_upper, color='b', alpha=0.2, label='Full Interval')
+
+        ax.fill_between(
+            time_steps,
+            ci_99_lower,
+            ci_99_upper,
+            color="b",
+            alpha=0.2,
+            label="99% Interval",
+        )
+        ax.fill_between(
+            time_steps,
+            ci_95_lower,
+            ci_95_upper,
+            color="b",
+            alpha=0.2,
+            label="95% Interval",
+        )
+        ax.fill_between(
+            time_steps,
+            ci_90_lower,
+            ci_90_upper,
+            color="b",
+            alpha=0.2,
+            label="90% Interval",
+        )
+        ax.fill_between(
+            time_steps,
+            ci_50_lower,
+            ci_50_upper,
+            color="b",
+            alpha=0.2,
+            label="50% Interval",
         )
 
+        ax.plot(next_day_np, label="Real NRV", linewidth=3)
+        # full_interval_patch = mpatches.Patch(color='b', alpha=0.2, label='Full Interval')
+        ci_99_patch = mpatches.Patch(color="b", alpha=0.3, label="99% Interval")
+        ci_95_patch = mpatches.Patch(color="b", alpha=0.4, label="95% Interval")
+        ci_90_patch = mpatches.Patch(color="b", alpha=0.5, label="90% Interval")
+        ci_50_patch = mpatches.Patch(color="b", alpha=0.6, label="50% Interval")
+
+        ax.legend(
+            handles=[
+                ci_99_patch,
+                ci_95_patch,
+                ci_90_patch,
+                ci_50_patch,
+                ax.lines[0],
+                ax.lines[1],
+            ]
+        )
         return fig
 
     def auto_regressive(self, dataset, idx_batch, sequence_length: int = 96):
-        return auto_regressive(dataset, self.model, self.quantiles, idx_batch, sequence_length)
+        return auto_regressive(
+            dataset, self.model, self.quantiles, idx_batch, sequence_length
+        )
 
     def plot_quantile_percentages(
-        self, task, data_loader, train: bool = True, iteration: int = None, full_day: bool = False
+        self,
+        task,
+        data_loader,
+        train: bool = True,
+        iteration: int = None,
+        full_day: bool = False,
     ):
         quantiles = self.quantiles
         total = 0
@@ -368,20 +499,18 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
 
                 else:
                     inputs = inputs.to(self.device)
-                    outputs = self.model(inputs).cpu().numpy() # (batch_size, num_quantiles)
+                    outputs = (
-                    targets = targets.squeeze(-1).cpu().numpy() # (batch_size, 1)
+                        self.model(inputs).cpu().numpy()
+                    )  # (batch_size, num_quantiles)
+                    targets = targets.squeeze(-1).cpu().numpy()  # (batch_size, 1)
 
                 for i, q in enumerate(quantiles):
-                    quantile_counter[q] += np.sum(
+                    quantile_counter[q] += np.sum(targets < outputs[:, i])
-                       targets < outputs[:, i]
-                    )
 
                 total += len(targets)
 
         # to numpy array of length len(quantiles)
-        percentages = np.array(
+        percentages = np.array([quantile_counter[q] / total for q in quantiles])
-            [quantile_counter[q] / total for q in quantiles]
-        )
 
         bar_width = 0.35
         index = np.arange(len(quantiles))
@@ -389,9 +518,7 @@ class AutoRegressiveQuantileTrainer(AutoRegressiveTrainer):
         # Plotting the bars
         fig, ax = plt.subplots(figsize=(15, 10))
 
-        bar1 = ax.bar(
+        bar1 = ax.bar(index, quantiles, bar_width, label="Ideal", color="brown")
-            index, quantiles, bar_width, label="Ideal", color="brown"
-        )
         bar2 = ax.bar(
             index + bar_width, percentages, bar_width, label="NN model", color="blue"
         )
@@ -441,7 +568,6 @@ class NonAutoRegressiveQuantileRegression(Trainer):
     ):
         self.quantiles = quantiles
 
-
         criterion = NonAutoRegressivePinballLoss(quantiles=quantiles)
         super().__init__(
             model=model,

src/trainers/trainer.py

@@ -86,7 +86,7 @@ class Trainer:
 
     def random_samples(self, train: bool = True, num_samples: int = 10):
         train_loader, test_loader = self.data_processor.get_dataloaders(
-            predict_sequence_length=self.model.output_size
+            predict_sequence_length=96
         )
 
         if train:
@@ -94,7 +94,14 @@ class Trainer:
         else:
             loader = test_loader
 
-        indices = np.random.randint(0, len(loader.dataset) - 1, size=num_samples)
+        np.random.seed(42)
+        actual_indices = np.random.choice(loader.dataset.full_day_valid_indices, num_samples, replace=False)
+        indices = {}
+        for i in actual_indices:
+            indices[i] = loader.dataset.valid_indices.index(i)
+
+        print(actual_indices)
+        
         return indices
 
     def train(self, epochs: int, remotely: bool = False, task: Task = None):
@@ -107,8 +114,8 @@ class Trainer:
                 predict_sequence_length=self.model.output_size
             )
 
-            train_samples = self.random_samples(train=True)
+            train_samples = self.random_samples(train=True, num_samples=5)
-            test_samples = self.random_samples(train=False)
+            test_samples = self.random_samples(train=False, num_samples=5)
 
             self.init_clearml_task(task)
 
@@ -189,7 +196,7 @@ class Trainer:
 
             if task:
                 self.finish_training(task=task)
-                task.close()
+                # task.close()
         except Exception:
             if task:
                 task.close()

src/training_scripts/autoregressive_quantiles.py

@@ -1,3 +1,14 @@
+from src.utils.clearml import ClearMLHelper
+
+#### ClearML ####
+clearml_helper = ClearMLHelper(project_name="Thesis/NrvForecast")
+task = clearml_helper.get_task(
+    task_name="Autoregressive Quantile Regression: Non Linear"
+)
+task.execute_remotely(queue_name="default", exit_process=True)
+
+from src.policies.PolicyEvaluator import PolicyEvaluator
+from src.policies.simple_baseline import BaselinePolicy, Battery
 from src.models.lstm_model import GRUModel
 from src.data import DataProcessor, DataConfig
 from src.trainers.quantile_trainer import AutoRegressiveQuantileTrainer
@@ -11,11 +22,6 @@ import torch.nn as nn
 from src.models.time_embedding_layer import TimeEmbedding
 
 
-#### ClearML ####
-clearml_helper = ClearMLHelper(project_name="Thesis/NrvForecast")
-task = clearml_helper.get_task(task_name="Autoregressive Quantile Regression: Linear + Quarter + DoW + Load + Wind + Net")
-
-
 #### Data Processor ####
 data_config = DataConfig()
 
@@ -32,7 +38,6 @@ data_config.DAY_OF_WEEK = True
 data_config.NOMINAL_NET_POSITION = True
 
 
-
 data_config = task.connect(data_config, name="data_features")
 
 data_processor = DataProcessor(data_config, path="", lstm=False)
@@ -58,22 +63,35 @@ else:
 
 model_parameters = {
     "learning_rate": 0.0001,
-    "hidden_size": 512,
+    "hidden_size": 256,
-    "num_layers": 2,
+    "num_layers": 4,
     "dropout": 0.2,
-    "time_feature_embedding": 4,
+    "time_feature_embedding": 16,
 }
 
 model_parameters = task.connect(model_parameters, name="model_parameters")
 
-time_embedding = TimeEmbedding(data_processor.get_time_feature_size(), model_parameters["time_feature_embedding"])
+time_embedding = TimeEmbedding(
+    data_processor.get_time_feature_size(), model_parameters["time_feature_embedding"]
+)
 # lstm_model = GRUModel(time_embedding.output_dim(inputDim), len(quantiles), hidden_size=model_parameters["hidden_size"], num_layers=model_parameters["num_layers"], dropout=model_parameters["dropout"])
-# non_linear_model = NonLinearRegression(time_embedding.output_dim(inputDim), len(quantiles), hiddenSize=model_parameters["hidden_size"], numLayers=model_parameters["num_layers"], dropout=model_parameters["dropout"])
+non_linear_model = NonLinearRegression(
-linear_model = LinearRegression(time_embedding.output_dim(inputDim), len(quantiles))
+    time_embedding.output_dim(inputDim),
+    len(quantiles),
+    hiddenSize=model_parameters["hidden_size"],
+    numLayers=model_parameters["num_layers"],
+    dropout=model_parameters["dropout"],
+)
+# linear_model = LinearRegression(time_embedding.output_dim(inputDim), len(quantiles))
 
-model = nn.Sequential(time_embedding, linear_model)
+model = nn.Sequential(time_embedding, non_linear_model)
 optimizer = torch.optim.Adam(model.parameters(), lr=model_parameters["learning_rate"])
 
+### Policy Evaluator ###
+battery = Battery(2, 1)
+baseline_policy = BaselinePolicy(battery, data_path="")
+policy_evaluator = PolicyEvaluator(baseline_policy, task)
+
 #### Trainer ####
 trainer = AutoRegressiveQuantileTrainer(
     model,
@@ -82,6 +100,7 @@ trainer = AutoRegressiveQuantileTrainer(
     data_processor,
     quantiles,
     "cuda",
+    policy_evaluator=policy_evaluator,
     debug=False,
 )
 
@@ -91,3 +110,15 @@ trainer.add_metrics_to_track(
 trainer.early_stopping(patience=30)
 trainer.plot_every(5)
 trainer.train(task=task, epochs=epochs, remotely=True)
+
+### Policy Evaluation ###
+idx_samples = trainer.test_set_samples
+_, test_loader = trainer.data_processor.get_dataloaders(
+    predict_sequence_length=trainer.model.output_size, full_day_skip=True
+)
+
+policy_evaluator.evaluate_test_set(idx_samples, test_loader)
+policy_evaluator.plot_profits_table()
+policy_evaluator.plot_thresholds_per_day()
+
+task.close()

src/training_scripts/diffusion_training.py

@@ -1,25 +1,17 @@
-from clearml import Task
-from src.data import DataProcessor, DataConfig
-from src.trainers.trainer import Trainer
 from src.utils.clearml import ClearMLHelper
-from src.models import *
-from src.losses import *
-import torch
-import numpy as np
-from torch.nn import MSELoss, L1Loss
-from datetime import datetime
-import torch.nn as nn
-from src.models.time_embedding_layer import TimeEmbedding
-from src.models.diffusion_model import GRUDiffusionModel, SimpleDiffusionModel
-from src.trainers.diffusion_trainer import DiffusionTrainer
-
 
 clearml_helper = ClearMLHelper(project_name="Thesis/NrvForecast")
 task = clearml_helper.get_task(task_name="Diffusion Training")
-
-# execute remotely
 task.execute_remotely(queue_name="default", exit_process=True)
-print("Running remotely")
+
+from src.models import *
+from src.losses import *
+from src.models.time_embedding_layer import TimeEmbedding
+from src.models.diffusion_model import GRUDiffusionModel, SimpleDiffusionModel
+from src.trainers.diffusion_trainer import DiffusionTrainer
+from src.data import DataProcessor, DataConfig
+from src.policies.simple_baseline import BaselinePolicy, Battery
+from src.policies.PolicyEvaluator import PolicyEvaluator
 
 #### Data Processor ####
 data_config = DataConfig()
@@ -38,7 +30,7 @@ data_config.NOMINAL_NET_POSITION = True
 data_config = task.connect(data_config, name="data_features")
 
 data_processor = DataProcessor(data_config, path="", lstm=False)
-data_processor.set_batch_size(128)
+data_processor.set_batch_size(64)
 data_processor.set_full_day_skip(True)
 
 inputDim = data_processor.get_input_size()
@@ -47,18 +39,28 @@ print("Input dim: ", inputDim)
 model_parameters = {
     "epochs": 5000,
     "learning_rate": 0.0001,
-    "hidden_sizes": [512, 512, 512],
+    "hidden_sizes": [128, 128],
-    "time_dim": 64,
+    "time_dim": 8,
 }
 
 model_parameters = task.connect(model_parameters, name="model_parameters")
 
 #### Model ####
-# model = SimpleDiffusionModel(96, model_parameters["hidden_sizes"], other_inputs_dim=inputDim[1], time_dim=model_parameters["time_dim"])
+model = SimpleDiffusionModel(
-model = GRUDiffusionModel(96, model_parameters["hidden_sizes"], other_inputs_dim=inputDim[2], time_dim=model_parameters["time_dim"], gru_hidden_size=256)
+    96,
+    model_parameters["hidden_sizes"],
+    other_inputs_dim=inputDim[1],
+    time_dim=model_parameters["time_dim"],
+)
+# model = GRUDiffusionModel(96, model_parameters["hidden_sizes"], other_inputs_dim=inputDim[2], time_dim=model_parameters["time_dim"], gru_hidden_size=128)
 
-print("Starting training ...")
+### Policy Evaluator ###
+battery = Battery(2, 1)
+baseline_policy = BaselinePolicy(battery, data_path="")
+policy_evaluator = PolicyEvaluator(baseline_policy, task)
 
 #### Trainer ####
-trainer = DiffusionTrainer(model, data_processor, "cuda")
+trainer = DiffusionTrainer(
-trainer.train(model_parameters["epochs"], model_parameters["learning_rate"], task)
+    model, data_processor, "cuda", policy_evaluator=policy_evaluator
+)
+trainer.train(model_parameters["epochs"], model_parameters["learning_rate"], task)