add linear and log linear regression for spack builds

Signed-off-by: vsoch <vsoch@users.noreply.github.com>

experiments/simulation-spack-build/data/extract.py

@@ -0,0 +1,194 @@
+#!/usr/bin/env python3
+
+import dataset
+
+# import pickle
+import os
+
+# from glob import glob
+import pandas
+import yaml
+import json
+import random
+import numpy
+from jinja2 import Template
+
+from sklearn import linear_model
+import matplotlib.pyplot as plt
+from sklearn.preprocessing import OneHotEncoder
+
+here = os.path.abspath(os.path.dirname(__file__))
+
+# Nice selection of colors for cluster names (rainbow...)
+import matplotlib.colors as mcolors
+
+
+db_file = os.path.join(here, "raw", "db")
+if not os.path.exists(db_file):
+    raise ValueError(f"Database {db_file} does not exist.")
+
+labels_file = os.path.join(here, "raw", "query_result_2024-04-02T02_48_24.138231Z.csv")
+if not os.path.exists(labels_file):
+    raise ValueError(f"Labels {labels_file} does not exist.")
+
+# cluster template
+cluster_template = os.path.join(here, "templates", "cluster-nodes.json")
+subsystem_template = os.path.join(here, "templates", "subsystem.json")
+
+
+def load_yaml(filename):
+    with open(filename, "r") as file:
+        data = yaml.safe_load(file)
+    return data
+
+
+def get_db(filename: str) -> dataset.Database:
+    return dataset.connect(
+        url=f"sqlite:///{filename}", on_connect_statements=["PRAGMA foreign_keys = ON;"]
+    )
+
+
+def write_file(content, filename):
+    """
+    Write json to file
+    """
+    with open(filename, "w") as outfile:
+        outfile.write(content)
+
+
+def write_yaml(obj, filename):
+    with open(filename, "w") as file:
+        yaml.dump(obj, file)
+
+
+def write_json(obj, filename):
+    with open(filename, "w") as fd:
+        fd.write(json.dumps(obj, indent=4))
+
+
+def read_json(filename):
+    with open(filename, "r") as fd:
+        content = json.loads(fd.read())
+    return content
+
+
+def read_file(filename):
+    """
+    Read content from file
+    """
+    with open(filename, "r") as fd:
+        content = fd.read()
+    return content
+
+
+def main():
+    db = get_db(db_file)
+
+    # Read in labels
+    builds = pandas.DataFrame(db["builds"])
+    labels = pandas.read_csv(labels_file)
+
+    colors = list(mcolors.CSS4_COLORS)
+    random.shuffle(colors)
+
+    # Filter to successful builds, and try to create distribution of build durations
+    success_labels = labels[labels.status == "success"]
+    success = builds[builds.job_id.isin(success_labels.job_id)]
+    packages = success.pkg_name.unique()
+    data = {}
+    for package in packages:
+        subset = success[success.pkg_name == package]
+        data[package] = {
+            "build_count": subset.shape[0],
+            "compilers": list(subset.compiler_name.unique()),
+            "job_size": list(subset.job_size.unique()),
+            "arch": list(subset.arch.unique()),
+            "compiler_versions": list(subset.compiler_version.unique()),
+            "durations": {
+                "min": subset.duration.min(),
+                "max": subset.duration.max(),
+                "mean": success.duration.mean(),
+            },
+        }
+
+    write_json(data, "package-summaries.json")
+
+    # Keep these columns as is
+    columns = ["cpu_mean", "cpu_max", "cpu_min", "mem_mean", "mem_max", "mem_min"]
+
+    # Markdown file to preview all
+    content = "# Linear Models for Package Build Times\n\n"
+    log_content = "# Log Linear Models for Package Build Times\n\n"
+    for package in packages:
+        subset = success[success.pkg_name == package]
+
+        df = subset[columns]
+        df.index = subset.job_id
+
+        if df.shape[0] < 100:
+            continue
+
+        if not colors:
+            colors = list(mcolors.CSS4_COLORS)
+
+        # Let's try a linear model that predicts build time from features
+        encoder_one_hot = OneHotEncoder(sparse_output=False)
+        for column in [
+            "job_size",
+            "compiler_name",
+            "compiler_version",
+            "arch",
+            "pkg_version",
+        ]:
+            encoded_x = pandas.DataFrame(
+                encoder_one_hot.fit_transform(subset[[column]])
+            )
+            encoded_x.columns = [
+                f"{column}_{x}" for x in list(encoder_one_hot.categories_[0])
+            ]
+            encoded_x.index = df.index
+            df = pandas.merge(df, encoded_x, right_index=True, left_index=True)
+
+        regr = linear_model.LinearRegression()
+        y = subset.duration.tolist()
+        regr.fit(df.values, y)
+        pred = regr.predict(df.values)
+
+        plt.scatter(pred, y, color=colors.pop(0))
+        plt.title(f"Linear model: build times for {package} with {df.shape[0]} builds")
+        plt.xlabel("Actual build time (seconds)", fontsize=14)
+        plt.ylabel("Predicted build time (seconds)", fontsize=14)
+        path = os.path.join(here, "img", f"{package}-build-time-linear-regression.png")
+        plt.savefig(path)
+        relpath = os.path.basename(path)
+        content += f"\n## {package}\n\n![img/{relpath}](img/{relpath})\n"
+        plt.close()
+
+        # Do the same, but take log
+        regr = linear_model.LinearRegression()
+
+        # We shouldn't have any zeros
+        y = numpy.log(subset.duration.tolist())
+        regr.fit(df.values, y)
+        pred = [numpy.exp(x) for x in regr.predict(df.values)]
+
+        plt.scatter(pred, subset.duration.tolist(), color=colors.pop(0))
+        plt.title(
+            f"Log linear model: build times for {package} with {df.shape[0]} builds"
+        )
+        plt.xlabel("Actual build time (seconds)", fontsize=14)
+        plt.ylabel("Predicted build time (seconds)", fontsize=14)
+        path = os.path.join(
+            here, "img", f"{package}-build-time-log-linear-regression.png"
+        )
+        plt.savefig(path)
+        relpath = os.path.basename(path)
+        log_content += f"\n## {package}\n\n![img/{relpath}](img/{relpath})\n"
+        plt.close()
+
+    write_file(content, "regression-model.md")
+    write_file(log_content, "regression-log-model.md")
+
+
+if __name__ == "__main__":
+    main()