Add the script to parse tuning log

deep_speech_2/tools/parse_tuning_log.py

@@ -0,0 +1,102 @@
+"""Parse the log for tuning and plot error surface."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import re
+import numpy as np
+import argparse
+import functools
+import _init_paths
+from utils.utility import add_arguments, print_arguments
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+
+parser = argparse.ArgumentParser(description=__doc__)
+add_arg = functools.partial(add_arguments, argparser=parser)
+add_arg("log_path", str, '', "log path for parsing")
+add_arg("fig_name", str, 'error_surface.png', "name of output figure")
+args = parser.parse_args()
+
+
+def plot_error_surface(num_alphas, alphas, betas, error_rate_type, err_ave):
+    fig = plt.figure(figsize=(8, 6))
+    ax = Axes3D(fig)
+
+    num_betas = len(alphas) // num_alphas
+    alphas_2d = np.reshape(alphas, (num_alphas, num_betas))
+    betas_2d = np.reshape(betas, (num_alphas, num_betas))
+    err_ave_2d = np.reshape(err_ave, (num_alphas, num_betas))
+
+    ax.plot_surface(
+        alphas_2d,
+        betas_2d,
+        err_ave_2d,
+        rstride=1,
+        cstride=1,
+        alpha=0.8,
+        cmap='rainbow')
+    z_label = 'WER' if error_rate_type == 'wer' else 'CER'
+    ax.set_xlabel('alpha', fontsize=12)
+    ax.set_ylabel('beta', fontsize=12)
+    ax.set_zlabel(z_label, fontsize=12)
+    plt.savefig(args.fig_name)
+    plt.show()
+
+
+def parse_log():
+    if not os.path.isfile(args.log_path):
+        raise IOError("Invaid model path: %s" % args.log_path)
+
+    error_rate_type = None
+    num_alphas, num_betas = 0, 0
+    alphas, betas, err_ave = [], [], []
+
+    err_rate_pat = re.compile(
+        '\(alpha, beta\) = '
+        '\([-+]?\d+(?:\.\d+)?, [-+]?\d+(?:\.\d+)?\), \[[wcer]')
+    num_pat = re.compile(r'[-+]?\d+(?:\.\d+)?')
+
+    with open(args.log_path, "r") as log_file:
+        line = log_file.readline()
+        while line:
+            if err_rate_pat.match(line) is not None:
+                triple = num_pat.findall(line)
+                alphas.append(float(triple[0]))
+                betas.append(float(triple[1]))
+                err_ave.append(float(triple[2]))
+            elif line.find("error_rate_type:") != -1:
+                error_rate_type = line.strip().split()[1]
+            elif line.find("num_alphas:") != -1:
+                num_alphas = int(line.strip().split()[1])
+            elif line.find("num_betas:") != -1:
+                num_betas = int(line.strip().split()[1])
+            line = log_file.readline()
+
+    if error_rate_type == None:
+        raise ValueError("Illegal log format, cannot find error_rate_type")
+
+    if num_alphas <= 0:
+        raise ValueError("Illegal log format, invalid num_alphas")
+
+    if num_betas <= 0:
+        raise ValueError("Illegal log format, invalid num_betas")
+
+    if alphas == []:
+        raise ValueError("Illegal log format, cannot find grid search result")
+
+    if num_alphas * num_betas != len(alphas):
+        raise ValueError("Illegal log format, data's shape mismatches")
+
+    return num_alphas, alphas, betas, error_rate_type, err_ave,
+
+
+def main():
+    print_arguments(args)
+    num_alphas, alphas, betas, error_rate_type, err_ave = parse_log()
+    plot_error_surface(num_alphas, alphas, betas, error_rate_type, err_ave)
+
+
+if __name__ == '__main__':
+    main()