add cross-experiment plots

mixnet/cmd/queuesim_plot_crossexp.py

@@ -0,0 +1,20 @@
+import argparse
+
+import pandas as pd
+
+from queuesim.paramset import ExperimentID, SessionID
+from queuesim.plot import draw_cross_experiment_plots, draw_plots
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Draw cross-experiment plots",
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+    )
+    parser.add_argument(
+        "--csv-paths", type=str, required=True, help="Experiment result CSV paths"
+    )
+    parser.add_argument("--outdir", type=str, required=True, help="output directory")
+    args = parser.parse_args()
+
+    dfs = [pd.read_csv(path) for path in args.csv_paths.split(",")]
+    draw_cross_experiment_plots(dfs, args.outdir)

mixnet/queuesim/plot.py

@@ -57,15 +57,6 @@ def draw_plots(
         )
 
 
-def __param_set_legend(row):
-    legend = ""
-    for i, param in enumerate(PARAM_SET):
-        if i > 0:
-            legend += ", "
-        legend += f"{param}:{row[param]}"
-    return legend
-
-
 def __overview_by_queue_type(
     df: pd.DataFrame,
     exp_id: ExperimentID,
@@ -97,14 +88,24 @@ def __impact_of_param_by_queue_type(
     )
 
 
-def __draw_plot_by_queue_type(df: pd.DataFrame, title: str, out_path: str):
+def __draw_plot_by_queue_type(
+    df: pd.DataFrame, title: str, out_path: str, legend_columns: list[str] = PARAM_SET
+):
     # Add a column that will be used as a legend
-    df["parameter_set"] = df.apply(__param_set_legend, axis=1)
+    def __create_legend_value(row):
+        legend = ""
+        for i, param in enumerate(legend_columns):
+            if i > 0:
+                legend += ", "
+            legend += f"{param}:{row[param]}"
+        return legend
+
+    df["legend"] = df.apply(__create_legend_value, axis=1)
 
     # Prepare DataFrame in long format for seaborn boxplot
     long_format_df = pd.melt(
         df,
-        id_vars=["queue_type", "parameter_set"],
+        id_vars=["queue_type", "legend"],
         value_vars=BOXPLOT_VALUE_VARS,
         var_name="dtime_metric",
         value_name="dtime",
@@ -112,7 +113,7 @@ def __draw_plot_by_queue_type(df: pd.DataFrame, title: str, out_path: str):
 
     # Plotting
     plt.figure(figsize=(15, 10))
-    sns.boxplot(data=long_format_df, x="queue_type", y="dtime", hue="parameter_set")
+    sns.boxplot(data=long_format_df, x="queue_type", y="dtime", hue="legend")
     plt.title(title)
     plt.xlabel("Queue Type")
     plt.ylabel("Dissemination Time")
@@ -131,3 +132,55 @@ def __draw_plot_by_queue_type(df: pd.DataFrame, title: str, out_path: str):
     assert not os.path.exists(out_path)
     plt.savefig(out_path)
     plt.draw()
+    print(f"Saved plot to {out_path}")
+
+
+def draw_cross_experiment_plots(dfs: list[pd.DataFrame], outdir: str):
+    assert os.path.exists(outdir)
+    assert len(dfs) == len(ExperimentID)
+
+    # Common filtering conditions
+    common_conditions = {
+        "num_nodes": 80,
+        "peering_degree": 16,
+        "min_queue_size": 40,
+        "transmission_rate": 40,
+    }
+    # Define the filtering conditions for each DataFrame
+    conditions = [
+        {"num_senders": 1, "num_sent_msgs": 1},
+        {"num_senders": 1, "num_sent_msgs": 8},
+        {"num_senders": 8, "num_sent_msgs": 1},
+        {"num_senders": 8, "num_sent_msgs": 8},
+    ]
+
+    filtered_dfs: list[pd.DataFrame] = []
+    for exp_idx, (df, condition) in enumerate(zip(dfs, conditions)):
+        # Combine common and specific conditions
+        all_conditions = {**common_conditions, **condition}
+        # Filter the DataFrame
+        filtered_df = pd.DataFrame(
+            df[
+                (df["num_nodes"] == all_conditions["num_nodes"])
+                & (df["peering_degree"] == all_conditions["peering_degree"])
+                & (df["min_queue_size"] == all_conditions["min_queue_size"])
+                & (df["transmission_rate"] == all_conditions["transmission_rate"])
+                & (df["num_senders"] == all_conditions["num_senders"])
+                & (df["num_sent_msgs"] == all_conditions["num_sent_msgs"])
+            ]
+        )
+        filtered_df["experiment"] = ExperimentID(exp_idx + 1).value
+        filtered_dfs.append(
+            pd.DataFrame(
+                filtered_df[
+                    ["experiment"] + PARAM_SET + ["queue_type"] + BOXPLOT_VALUE_VARS
+                ]
+            )
+        )
+
+    __draw_plot_by_queue_type(
+        pd.concat(filtered_dfs),
+        "Comparisons by Experiment and Queue Type",
+        f"{outdir}/cross_experiment_plot.png",
+        legend_columns=["experiment"] + PARAM_SET,
+    )