update to 2.4.0 & add spinningup plotter

Author: gxywy

README_zh.md

@@ -62,7 +62,7 @@ custom_logger.update(fieldvalues=变量值(list), total_steps=当前训练步数
 rl_plotter --save --show
 ```
 
-## Example
+## 例子
 
 **1. 常用命令**
 

rl_plotter/__init__.py

@@ -1 +1 @@
-__all__ = ["logger", "log_utils", "plotter", "plot_utils"]
+__all__ = ["logger", "log_utils", "plotter", "plotter_spinup", "plot_utils"]

rl_plotter/plot_utils.py

@@ -3,6 +3,7 @@
 
 __author__ = 'MICROYU'
 
+from datetime import date
 import re
 import matplotlib.pyplot as plt
 import os.path as osp
@@ -162,7 +163,7 @@ def load_csv_results(dir, filename="monitor.csv"):
 	#df.headers = headers # HACK to preserve backwards compatibility
 	return df
 
-def load_results(root_dir_or_dirs="./", filename="monitor.csv", filters=[]):
+def load_results(root_dir_or_dirs="./", filename="monitor.csv", filters=['']):
 
 	if isinstance(root_dir_or_dirs, str):
 		rootdirs = [osp.expanduser(root_dir_or_dirs)]
@@ -368,21 +369,78 @@ def allequal(qs):
 			plt.legend(
 					[groups_results[key]['legend'] for key in groups_results.keys()],
 					['%s (%i)'%(key.replace('without', 'w/o').replace('_', '-'), groups_results[key]['num']) for key in groups_results.keys()] if average_group else groups_results.keys(),
-					loc=9 if legend_outside else legend_loc, bbox_to_anchor = (0.5,-0.1) if legend_outside else (1,1) if legend_outside else None, borderpad=legend_borderpad, labelspacing=legend_labelspacing,ncol= len(groups_results.keys()) if legend_outside else 1)
+					loc=9 if legend_outside else legend_loc, bbox_to_anchor = (0.5,-0.1) if legend_outside else (1,1) if legend_outside else None, borderpad=legend_borderpad, labelspacing=legend_labelspacing, ncol=len(groups_results.keys()) if legend_outside else 1)
 		else:
 			plt.legend(
 					[groups_results[key]['legend'] for key in groups_results.keys()],
 					['%s'%(key.replace('without', 'w/o').replace('_', '-')) for key in groups_results.keys()] if average_group else groups_results.keys(),
-					loc=9 if legend_outside else legend_loc, bbox_to_anchor = (0.5,-0.1) if legend_outside else (1,1) if legend_outside else None, borderpad=legend_borderpad, labelspacing=legend_labelspacing,ncol= len(groups_results.keys()) if legend_outside else 1)
+					loc=9 if legend_outside else legend_loc, bbox_to_anchor = (0.5,-0.1) if legend_outside else (1,1) if legend_outside else None, borderpad=legend_borderpad, labelspacing=legend_labelspacing, ncol=len(groups_results.keys()) if legend_outside else 1)
 	# add title
 	plt.title(title)
 	# add xlabels
 	if xlabel is not None: plt.xlabel(xlabel)
 
 
+def plot_data(data, xaxis='total_steps', value="mean_score", condition="Condition1", smooth=1, 
+        legend_outside=False,
+        legend_loc=0,
+        legend_borderpad=1.0,
+        legend_labelspacing=1.0,
+        font_scale=1.5,
+        **kwargs):
+    import seaborn as sns
+    if smooth > 1:
+        """
+        smooth data with moving window average.
+        that is,
+            smoothed_y[t] = average(y[t-k], y[t-k+1], ..., y[t+k-1], y[t+k])
+        where the "smooth" param is width of that window (2k+1)
+        """
+        y = np.ones(smooth)
+        for datum in data:
+            x = np.asarray(datum[value])
+            z = np.ones(len(x))
+            smoothed_x = np.convolve(x,y,'same') / np.convolve(z,y,'same')
+            datum[value] = smoothed_x
+    if isinstance(data, list):
+        data = pandas.concat(data, ignore_index=True)
+    
+    data.sort_values(by='Condition1', axis=0)
+
+    sns.set(style="darkgrid", font_scale=font_scale)
+    sns.lineplot(data=data, x=xaxis, y=value, hue=condition, ci='sd', **kwargs)
+    handles, labels = plt.gca().get_legend_handles_labels()
+
+    plt.legend(
+        handles[1:],
+        ['%s'%(key.replace('without', 'w/o').replace('_', '-')) for key in labels[1:]],
+        loc=9 if legend_outside else legend_loc, bbox_to_anchor = (0.5,-0.1) if legend_outside else (1,1) if legend_outside else None, borderpad=legend_borderpad, labelspacing=legend_labelspacing, ncol=len(labels)-1 if legend_outside else 1)
+
+    xscale = np.max(np.asarray(data[xaxis])) > 5e3
+    if xscale:
+        # Just some formatting niceness: x-axis scale in scientific notation if max x is large
+        plt.ticklabel_format(style='sci', axis='x', scilimits=(0,0))
+
+    plt.tight_layout(pad=0.5)
+
+
 if __name__ == "__main__":
-	allresults = load_results("./logs", filename="evaluator.csv")
-	for result in allresults:
-		print(result["dirname"])
-	plot_results(allresults, average_group=False, smooth_radius=0)
-	#plt.savefig('figure', dpi=400)
+	# allresults = load_results("./logs", filename="evaluator.csv")
+	# for result in allresults:
+	# 	print(result["dirname"])
+	# plot_results(allresults, average_group=False, smooth_radius=0)
+	# plt.show()
+
+    # allresults = load_results("./", filename="evaluator.csv")
+    # datas = []
+    # for result in allresults:
+    #     result['data'].insert(len(result['data'].columns),'Condition1', default_split_fn(result))
+    #     datas.append(result['data'])
+    # plt.figure()
+    # fig = plt.gcf()
+    # fig.set_size_inches((16, 9), forward=False)
+
+    # plot_data(data=datas)
+    # plt.show()
+    pass
+    

rl_plotter/plotter.py

@@ -11,8 +11,8 @@
 
 def main():
 	parser = argparse.ArgumentParser(description='plotter')
-	parser.add_argument('--fig_length', type=int, default=8, 
-						help='matplotlib figure length (default: 8)')
+	parser.add_argument('--fig_length', type=int, default=6, 
+						help='matplotlib figure length (default: 6)')
 	parser.add_argument('--fig_width', type=int, default=6, 
 						help='matplotlib figure width (default: 6)')
 	parser.add_argument('--style', default='seaborn',
@@ -68,7 +68,7 @@ def main():
 						help='log dir (default: ./)')
 	parser.add_argument('--filters', default=[''], nargs='+',
 						help='filters of dirname')
-	parser.add_argument('--filename', default='evaluator',
+	parser.add_argument('--filename', default='evaluator.csv',
 						help='csv filename')
 	parser.add_argument('--show', action='store_true',
 						help='show figure')

rl_plotter/plotter_spinup.py

@@ -0,0 +1,112 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+__author__ = 'MICROYU'
+
+import argparse
+import matplotlib.pyplot as plt
+import matplotlib.ticker as mticker
+from rl_plotter import plot_utils as pu
+
+
+def main():
+	parser = argparse.ArgumentParser(description='rl-plotter')
+	parser.add_argument('--log_dir', default='./',
+						help='log dir (default: ./)')
+	parser.add_argument('--filters', default=[''], nargs='+',
+						help='filters of dirname')
+	parser.add_argument('--filename', default='evaluator.csv',
+						help='csv filename')
+	parser.add_argument('--show', action='store_true',
+						help='show figure')
+	parser.add_argument('--save', action='store_true',
+					help='save figure')
+	parser.add_argument('--dpi', type=int, default=400,
+						help='figure dpi (default: 400)')
+	parser.add_argument('--fig_length', type=int, default=6, 
+						help='matplotlib figure length (default: 6)')
+	parser.add_argument('--fig_width', type=int, default=6, 
+						help='matplotlib figure width (default: 6)')
+
+	parser.add_argument('--title', default=None,
+						help='matplotlib figure title (default: None)')
+	parser.add_argument('--xlabel', default=None,
+						help='matplotlib figure xlabel')
+	parser.add_argument('--ylabel', default=None,
+						help='matplotlib figure ylabel')
+	parser.add_argument('--xkey', default='total_steps',
+						help='x-axis key in csv file (default: l)')
+	parser.add_argument('--ykey', default=['mean_score'], nargs='+',
+						help='y-axis key in csv file (support multi) (default: r)')
+	parser.add_argument('--smooth', type=int, default=1,
+						help='smooth radius of y axis (default: 1)')
+	parser.add_argument('--xlim', type=int, default=None,
+						help='x-axis limitation (default: None)')
+	
+	parser.add_argument('--legend_loc', type=int, default=0,
+						help='location of legend')
+	parser.add_argument('--legend_outside', action='store_true',
+						help='place the legend outside of the figure')
+	parser.add_argument('--borderpad', type=float, default=0.5,
+						help='borderpad of legend (default: 0.5)')
+	parser.add_argument('--labelspacing', type=float, default=0.5,
+						help='labelspacing of legend (default: 0.5)')
+	parser.add_argument('--font_scale', type=float, default=1,
+						help='font_scale of seaborn (default: 1)')
+	args = parser.parse_args()
+	
+	if args.xlabel is None:
+		args.xlabel = 'Timesteps'
+
+	if args.ylabel is None:
+		args.ylabel = 'Episode Reward'
+
+	if '.' not in args.filename:
+		args.filename = args.filename + '.csv'
+	
+	# OpenAI baseline's monitor
+	if args.filename == 'monitor.csv':
+		args.xkey = 'l'
+		args.ykey = ['r']
+	
+	# OpenAI spinup's progress
+	if args.filename == 'progress.txt' or args.filename == 'progress.csv':
+		args.xkey = 'TotalEnvInteracts'
+		args.ykey = ['AverageTestEpRet']
+	
+	# rl-plotter's evaluator
+	if args.filename == 'evaluator.csv':
+		args.xkey = 'total_steps'
+		args.ykey = ['mean_score']
+
+	if args.save is False:
+		args.show = True
+
+	allresults = pu.load_results(args.log_dir, filename=args.filename, filters=args.filters)
+	datas = []
+	for result in allresults:
+		result['data'].insert(len(result['data'].columns),'Condition1', pu.default_split_fn(result))
+		datas.append(result['data'])
+	pu.plot_data(data=datas, xaxis=args.xkey, value=args.ykey[0], smooth=args.smooth, 
+		legend_outside=args.legend_outside,
+		legend_loc=args.legend_loc,
+		legend_borderpad=args.borderpad,
+		legend_labelspacing=args.labelspacing,
+		font_scale=args.font_scale)
+	plt.title(args.title)
+	plt.xlabel(args.xlabel)
+	plt.ylabel(args.ylabel)
+	fig = plt.gcf()
+	fig.set_size_inches((args.fig_length, args.fig_width), forward=False)
+
+	if args.xlim is not None:
+		plt.xlim((0, args.xlim))
+
+	if args.save:
+		plt.savefig(args.log_dir + 'figure', dpi=args.dpi, bbox_inches='tight')
+	if args.show:
+		plt.show()
+
+
+if __name__ == "__main__":
+	main()

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
     name="rl_plotter",
-    version="2.3.6",
+    version="2.4.0",
     author="Xiaoyu Gong",
     author_email="[email protected]",
     description="A plotter for reinforcement learning (RL)",
@@ -19,7 +19,7 @@
         "Operating System :: OS Independent",
     ],
     entry_points = {
-        'console_scripts': ['rl_plotter=rl_plotter.plotter:main'],
+        'console_scripts': ['rl_plotter=rl_plotter.plotter:main','rl_plotter_spinup=rl_plotter.plotter_spinup:main'],
     },
     python_requires='>=3.0',
 )