Explore Training Curves

A basic notebook to generate plots from training logs. See Logger class in utils.py.

Python Notebook by Patrick Coady: Learning Artificial Intelligence

Summary

Hypothesis, conditions, and results explored in this notebook ...



In [1]:

    
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
%matplotlib inline
plt.rcParams["figure.figsize"] = (10,5)
plt.style.use('ggplot')



In [2]:

    
def df_plot(dfs, x, ys, ylim=None, legend_loc='best'):
    """ Plot y vs. x curves from pandas dataframe(s)
    
    Args:
        dfs: list of pandas dataframes
        x: str column label for x variable
        y: list of str column labels for y variable(s)
        ylim: tuple to override automatic y-axis limits
        legend_loc: str to override automatic legend placement:
            'upper left', 'lower left', 'lower right' , 'right' ,
            'center left', 'center right', 'lower center',
            'upper center', and 'center'
    """
    if ylim is not None:
        plt.ylim(ylim)
    for df, name in dfs:
        name = name.split('_')[1]
        for y in ys:
            plt.plot(df[x], df[y], linewidth=2,
                     label=name + ' ' + y.replace('_', ''))
    plt.xlabel(x.replace('_', ''))
    plt.legend(loc=legend_loc)
    plt.show()



In [3]:

    
# ENTER LIST OF LOG FILENAMES HERE:
filepaths = ['log-files/Hopper-v1/Jul-18_14:30:37/log.csv',
             'log-files/Hopper-v1/Jul-18_14:51:02/log.csv']
dataframes = []
names = []
for filepath in filepaths:
    names.append(filepath.split('/')[2])
    dataframes.append(pd.read_csv(filepath))
data = list(zip(dataframes, names))

Plots



In [4]:

    
df_plot(data, '_Iteration', ['_MeanReward'])
df_plot(data, '_Iteration', ['KL'])
df_plot(data, '_Iteration', ['ExplainedVarOld'])
df_plot(data, '_Iteration', ['PolicyLoss'], ylim=(-0.05, 0))



In [5]:

    
df_plot(data, '_Iteration', ['_mean_discrew'])
df_plot(data, '_Iteration', ['_std_discrew'])
df_plot(data, '_Iteration', ['_mean_adv'])
df_plot(data, '_Iteration', ['_std_adv'])



In [6]:

    
df_plot(data, '_Iteration', ['_mean_obs'])
df_plot(data, '_Iteration', ['_min_obs'])
df_plot(data, '_Iteration', ['_max_obs'])



In [7]:

    
df_plot(data, '_Iteration', ['_mean_act'])
df_plot(data, '_Iteration', ['_min_act'])
df_plot(data, '_Iteration', ['_max_act'])
df_plot(data, '_Iteration', ['_std_act'])



In [8]:

    
df_plot(data, '_Iteration', ['PolicyEntropy'])



In [ ]: