Merge pull request #57 from sgbaird/plotting_utils

add plotting uitls examples to refreshers

.pre-commit-config.yaml

@@ -9,6 +9,7 @@ repos:
     args: ['--maxkb=10000']
   - id: check-ast
   - id: check-json
+    exclude: ^\.devcontainer/devcontainer\.json$
   - id: check-merge-conflict
   - id: check-xml
   - id: check-yaml

oryx-build-commands.txt

@@ -0,0 +1,2 @@
+PlatformWithVersion=Python
+BuildCommands=conda env create --file environment.yml --prefix ./venv --quiet

scripts/refreshers/plot_honegumi.py

@@ -0,0 +1,127 @@
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+
+ax_client = None
+batch_size = 2
+
+# single objective, single observation -------------------------------------------------
+
+objectives = ax_client.objective_names
+df = ax_client.get_trials_data_frame()[objectives]
+
+trace = np.minimum.accumulate(df)  # change if maximizing
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=150)
+ax.scatter(df.index, df, ec="k", fc="none", label="Observed")
+ax.plot(df.index, trace, color="#0033FF", lw=2, label="Best to Trial")
+ax.set_xlabel("Trial Number")
+ax.set_ylabel(objectives[0])
+ax.legend()
+plt.show()
+
+# single objective, batch observations -------------------------------------------------
+
+objectives = ax_client.objective_names
+df = ax_client.get_trials_data_frame()[objectives]
+df.index = df.index // batch_size
+
+trace = np.minimum.accumulate(df.groupby(df.index).min())  # change if maximizing
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=150)
+ax.scatter(df.index, df, ec="k", fc="none", label="Observed")
+ax.plot(df.index.unique(), trace, color="#0033FF", lw=2, label="Best to Trial")
+ax.set_xlabel("Trial Number")
+ax.set_ylabel(objectives[0])
+ax.legend()
+plt.show()
+
+# multi objective, single or batch observations ----------------------------------------
+
+objectives = ax_client.objective_names
+df = ax_client.get_trials_data_frame()[objectives]
+
+pareto = ax_client.get_pareto_optimal_parameters()
+pareto = pd.DataFrame([p[1][0] for p in pareto.values()]).sort_values(objectives[0])
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=200)
+ax.scatter(df[objectives[0]], df[objectives[1]], fc="None", ec="k", label="Observed")
+ax.plot(
+    pareto[objectives[0]],
+    pareto[objectives[1]],
+    color="#0033FF",
+    lw=2,
+    label="Pareto Front",
+)
+ax.set_xlabel(objectives[0])
+ax.set_ylabel(objectives[1])
+ax.legend()
+plt.show()
+
+# multi task, single objective, single observation -------------------------------------
+
+task = "A"  # specify task results to plot
+
+objectives = ax_client.objective_names
+objectives.extend(["task"])
+df = ax_client.get_trials_data_frame()[objectives]
+df = df[df.task == task].drop(columns=["task"])
+
+trace = np.minimum.accumulate(df)  # change if maximizing
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=150)
+ax.scatter(df.index, df, ec="k", fc="none", label="Observed")
+ax.plot(df.index, trace, color="#0033FF", lw=2, label="Best to Trial")
+ax.set_xlabel("Trial Number")
+ax.set_ylabel(objectives[0])
+ax.legend()
+ax.set_title(f"Task {task}")
+plt.show()
+
+# multi task, single objective, batch observations -------------------------------------
+
+task = "A"  # specify task results to plot
+
+objectives = ax_client.objective_names
+objectives.extend(["task"])
+df = ax_client.get_trials_data_frame()[objectives]
+df = df[df.task == task].drop(columns=["task"])
+df.index = df.index // batch_size
+
+trace = np.minimum.accumulate(df.groupby(df.index).min())  # change if maximizing
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=150)
+ax.scatter(df.index, df, ec="k", fc="none", label="Observed")
+ax.plot(df.index.unique, trace, color="#0033FF", lw=2, label="Best to Trial")
+ax.set_xlabel("Trial Number")
+ax.set_ylabel(objectives[0])
+ax.legend()
+plt.show()
+
+# multi task, multi objective, single or batch observations ----------------------------
+
+# !!! not tested yet !!!
+
+task = "A"  # specify task results to plot
+
+objectives = ax_client.objective_names
+objectives.extend(["task"])
+df = ax_client.get_trials_data_frame()[objectives]
+df = df[df.task == task].drop(columns=["task"])
+
+pareto = ax_client.get_pareto_optimal_parameters()
+pareto = pd.DataFrame([p[1][0] for p in pareto.values()]).sort_values(objectives[0])
+
+fig, ax = plt.subplots(figsize=(6, 4), dpi=200)
+ax.scatter(df[objectives[0]], df[objectives[1]], fc="None", ec="k", label="Observed")
+ax.plot(
+    pareto[objectives[0]],
+    pareto[objectives[1]],
+    color="#0033FF",
+    lw=2,
+    label="Pareto Front",
+)
+ax.set_xlabel(objectives[0])
+ax.set_ylabel(objectives[1])
+ax.legend()
+plt.show()