plot-results.py

#!/usr/bin/env python
# coding: utf-8

# In[2]:


import os
import pandas as pd
import plotly.io as pio
import plotly.express as px
import plotly.graph_objs as go
from itertools import combinations 
from plotly.subplots import make_subplots
import plotly.io as pio

# render figures in notebook
pio.renderers.default = "notebook_connected"

# templates figures
px.defaults.template = "simple_white"
pio.templates.default = "simple_white"

# layout for all figures
LAYOUT_FIGURES = dict(
    autosize=False,
    width = 500,
    height = 500,
    xaxis = dict(
      constrain="domain",
      mirror=True,
      showexponent="all",
      exponentformat="power"
    ),
    yaxis = dict(
      scaleanchor = "x",
      scaleratio = 1,
      mirror=True,
      showexponent="all",
      exponentformat="power"
    ),
    title = dict(
      y = 0.9,
      x = 0.5,
      xanchor = 'center',
      yanchor = 'top'
    )
)


# # Auxiliary Functions

# In[3]:


def create_folder(path):
    """Creates a folder if it does not exist
    
    Parameters
    ----------
    path : str
        Path of the new folder
    
    Examples
    --------
    
    >>> create_folder('./results')
    """
    if not os.path.exists(path):
        os.makedirs(path)


# In[4]:


def create_figure(df, model):
    """Creates a scatter figure showing the time taken by each tool to verify each property of a model
    
    Parameters
    ----------
    df : pandas.Dataframe
        Dataframe containing the results of the experiments
    model : string
        model to be plotted
    
    Returns
    -------
    plotly.graph_objects.Figure
        Scatter figure
    
    Examples
    --------
    
    >>> import os
    >>> import pandas as pd
    >>> csv_file = os.path.join("results", "output.csv")
    >>> df = pd.read_csv(csv_file)
    >>> fig = create_figure(df, 'philo10')
    """
    model_df = df[df.model == model]

    figure = px.scatter(model_df, 
                        x="formula", y="time",
                        title=model, 
                        color="tool", 
                        symbol_sequence=['x'])

    figure.update_layout(yaxis_title="time (s)", title=LAYOUT_FIGURES['title'])
    return figure


# In[21]:


def get_axis_title(experiment, show_strategy=True):
    """Get the axis title of a figure depending on the experiment being plotted
    
    Parameters
    ----------
    experiment : str
        String with the experiment information
    show_strategy : bool, optional
        Flag to show the information related to the strategy used by the tool
    
    Returns
    -------
    str
        axis title
        
    Examples
    --------
    
    >>> get_axis_title('pmc-sog_otfL_couv99-default_1_1', True)
    pmc-sog (Lace, strategy: couv99-default, # cores: 1)
    """
    information = experiment.split('_')
    tool_name = information[0]
    
    info = []
    library_dic = {
        'otfL': 'Lace',
        'otfP': 'Pthreads',
        'otfC': 'Cthreads',
        'otf': 'Hybrid'
    }
    
    if (len(information) == 5):
        info.append(library_dic[information[1]])

    if (show_strategy):
        info.append('strategy: {}'.format(information[-3]))

    nb_nodes = int(information[-2])
    if (nb_nodes > 1):
        info.append('# nodes: {}'.format(nb_nodes))

    info.append('# cores: {}'.format(information[-1]))

    title = '{} ({})'.format(tool_name, ', '.join(info))
    
    return title


# In[22]:


def create_log_figure(table, table_errors, model, tool_x, tool_y, show_strategy=True):
    """Creates a Scatter figure in logarithmic scale comparing the performance of two tools
    
    Parameters
    ----------
    table : pandas.Dataframe
        Dataframe with the times of each experiment
    table_errors : pandas.Dataframe
        Dataframe with the errors of each experiment
    model : string
        Model to be analyzed
    tool_x : string
        Tool to be compared and plotted on the x-axis
    tool_y : string
        Tool to be compared and plotted on the y-axis
    show_strategy : bool
        Flag to show the stretagy used by the tools
        
    Returns
    -------
    plotly.graph_objects.Figure
        Scatter figure
        
    Examples
    --------
    >>> import os
    >>> import pandas as pd
    >>> csv_file = os.path.join("results", "output.csv")
    >>> df = pd.read_csv(csv_file)
    >>> table = df.set_index(['model', 'formula', 'tool'], drop=True).unstack('tool')
    >>> fig = create_log_figure(table['time'], table['error'], 'philo10', 'pmc-sog_otfL_couv99-default_1_8', 'pmc-sog_otfP_couv99-default_1_8')
    """
    try:
        min_values = table.loc[model].min()
        max_values = table.loc[model].max()

        min_value = min(min_values[tool_x], min_values[tool_y])/2.
        max_value = max(max_values[tool_x], max_values[tool_y])

        figure = px.scatter(table.loc[model],
                            title=model,
                            x=tool_x, y=tool_y, 
                            log_x=True, log_y=True,
                            range_x=[min_value, max_value],
                            range_y=[min_value, max_value],
                            color="property",
                            hover_data=[
                                ['formula #{}'.format(i) for i in table.loc[model].index],
                                table_errors.loc[model, tool_x],
                                table_errors.loc[model, tool_y]
                                ],
                            color_discrete_map={"T": "green", "F": "red", "U": "black"},
                            symbol_sequence=["circle-open"])

        line = go.Scatter(x=[min_value, max_value], 
                            y=[min_value, max_value],
                            mode='lines', showlegend=False,
                            line=dict(color='black', width=1))

        figure.add_traces(line)
        figure.update_layout(LAYOUT_FIGURES, 
                                xaxis_title=get_axis_title(tool_x, show_strategy),
                                yaxis_title=get_axis_title(tool_y, show_strategy))
    
        return figure
    except Exception as e:
        print("Error when ploting model: {} - tool_x: {} - tool_y: {}".format(model, tool_x, tool_y))
        print(e)


# In[31]:


# Experiment filters

def versus_dfs(experiments):
    """Selects only experiments using DFS strategy"""
    exp1, exp2 = experiments
    strategy_exp1= exp1.split('_')[1]
    strategy_exp2= exp2.split('_')[1]
    
    return strategy_exp1 == 'dfs' or strategy_exp2 == 'dfs'

def versus_sequential(experiments):
    """Selects only experiments run sequentially """
    exp1, exp2 = experiments
    nodes_exp1, threads_exp1 = exp1.split('_')[-2:]
    nodes_exp2, threads_exp2 = exp2.split('_')[-2:]

    return (nodes_exp1 == '1' and nodes_exp2 == '1') and             (threads_exp1 == '1' or threads_exp2 == '1')

def same_tool(experiments, tool):
    """Selects only experiments comparing the same tool"""
    exp1, exp2 = experiments
    tool_exp1= exp1.split('_')[0]
    tool_exp2= exp2.split('_')[0]
    return tool_exp1.startswith(tool) and tool_exp2.startswith(tool)

def same_number_threads(experiments):
    """Selects only experiments comparing the same number of processes and cores"""
    exp1, exp2 = experiments
    nodes_exp1, threads_exp1 = exp1.split('_')[-2:]
    nodes_exp2, threads_exp2 = exp2.split('_')[-2:]
    return (nodes_exp1 == nodes_exp2) and (threads_exp1 == threads_exp2) 

def same_thread_library(experiments):
    """Selects only experiments comparing the same parallelization library"""
    exp1, exp2 = experiments
    library_exp1 = exp1.split('_')[1]
    library_exp2 = exp2.split('_')[1]
    return library_exp1 == library_exp2

def same_strategy(experiments):
    """Selects only experiments comparing the same strategy"""
    exp1, exp2 = experiments
    strategy_exp1 = exp1.split('_')[2]
    strategy_exp2 = exp2.split('_')[2]
    return strategy_exp1 == strategy_exp2

def only_couvreur_strategy(experiments):
    """Selects only experiments comparing couvreur emptiness check algorithm"""
    exp1, exp2 = experiments
    strategy_exp1 = exp1.split('_')[2]
    strategy_exp2 = exp2.split('_')[2]
    return strategy_exp1.startswith('couv99') and strategy_exp2.startswith('couv99')

def compare_threads_library(experiments):
    """Compares parallization libraries used in pmc-sog. 
    
    It selects experiments where the tool is only pmc-sog and the strategy, number of threads, 
    number of processus are the same.
    """
    return same_tool(experiments, 'pmc-sog') and             same_strategy(experiments) and             same_number_threads(experiments) and             not same_thread_library(experiments)

def compare_couvreur_strategies(experiments):
    """Compares couvreurs strategies used in pmc-sog. 
    
    It selects experiments where the tool is only pmc-sog, the strategy is couvreur, and 
    the parallelization library, number of threads, number of processus are the same.
    """
    return only_couvreur_strategy(experiments) and             same_thread_library(experiments) and             same_number_threads(experiments)

def same_distributed_number_threads(experiments):
    """Selects only experiments where the multiplication of theirs nodes with cores are the same."""
    exp1, exp2 = experiments
    nodes_exp1, threads_exp1 = exp1.split('_')[-2:]
    nodes_exp2, threads_exp2 = exp2.split('_')[-2:]
    return (int(nodes_exp1) * int(threads_exp1)) == (int(nodes_exp2) * int(threads_exp2))

def compare_tools(experiments):
    """Compares pmc-sog and pnml2lts-mc using the DFS algorithm. 
    
    It selects experiments where the tools are not the same, the exploration algorithm is DFS and 
    the number of processus and cores are the same.
    """
    return same_distributed_number_threads(experiments) and             not (same_tool(experiments, 'pmc-sog') or same_tool(experiments,'pnml2lts-mc')) and             versus_dfs(experiments)

def compare_multithreading(experiments):
    """Compares the sequential and multi-core version of pmc-sog. 
    
    It selects experiments where the tools is pmc-sog, the parallelization library, the emptiness check 
    strategy are the same. Here the number of processus and cores are different.
    """
    return same_tool(experiments, 'pmc-sog') and             same_thread_library(experiments) and             same_strategy(experiments) and             versus_sequential(experiments)

def against_hybrid(experiments):
    """Selects only experiments comparing with hybrid mode"""
    exp1, exp2 = experiments
    library_exp1 = exp1.split('_')[1]
    library_exp2 = exp2.split('_')[1]
    return (library_exp1 == 'otf') or (library_exp2 == 'otf')


def compare_distributed(experiments):
    """Compares the hybrid version of pmc-sog"""
    return same_tool(experiments, 'pmc-sog') and         same_strategy(experiments) and         same_distributed_number_threads(experiments) and         against_hybrid(experiments)

# Plots to be created
plots = {
    'compare_thread_library': compare_threads_library,
    'compare_couvreur_algorithm': compare_couvreur_strategies,
    'compare_tools': compare_tools,
    'compare_multicore': compare_multithreading,
    'compare_distributed': compare_distributed
}


# # Load Data

# In[8]:


# Root folder
PROJECT_FOLDER = os.path.abspath(os.pardir)

# csv file with the output
csv_file = os.path.join(PROJECT_FOLDER, "results", "output.csv")

# Output folder
OUTPUT_FOLDER = os.path.join(PROJECT_FOLDER,"results", "figures")
create_folder(OUTPUT_FOLDER)


# In[9]:


# read data
df = pd.read_csv(csv_file)

# merge the information related to the experiment (# nodes, # threads, strategy) to the tool column
df['tool'] = df[['tool', 'strategy', 'num_nodes', 'num_threads']].astype(str).apply('_'.join, axis=1)
df = df.drop(columns=['strategy', 'num_nodes', 'num_threads'])

df.head()


# In[10]:


# ground truth for properties
p_df = pd.read_csv(csv_file)
p_df =p_df[
    (p_df.tool=='pnml2lts-mc') & 
    (p_df.strategy == 'ndfs') & 
    (p_df.num_nodes == 1) & 
    (p_df.num_threads == 16)]

# only property column is needed
p_df = p_df.drop(columns=['tool', 'strategy', 'num_nodes', 'num_threads', 'time', 'error'])
p_df.fillna('U', inplace=True)
p_df.set_index(['model', 'formula'], inplace=True)
p_df.sort_index(inplace=True)

p_df.head()


# In[11]:


# table with times, verification output and error for each experiment
table = df.set_index(['model', 'formula', 'tool'], drop=True).unstack('tool')
table.head()


# # Preprocessing of data

# In[12]:


ZERO = 10e-5
TIMEOUT = 10 * 60 # 5 minutes = 300 seconds


# In[13]:


# table with times for each experiment
table_time = table['time'].copy()

# replace non finished experiments with a dummy value, e.g. timeout
table_time.fillna(TIMEOUT, inplace=True)

# replace 0.00 time for 10^(-5), we cannot plot log(0)
table_time.replace(0.0, ZERO, inplace=True)

# add verification output to the table
table_time = pd.concat([table_time, p_df], axis=1)

table_time.head()


# In[14]:


# table with verification output for each experiment
table_property = table['property'].copy()

# replace non finished experiments with a dummy value
table_property.fillna('U', inplace=True)

# add ground truth to the table
table_property = pd.concat([table_property, p_df], axis=1)

table_property.head()


# In[15]:


# table with error for each experiment
table_error = table['error'].copy()

table_error.head()


# # Examples

# In[19]:


fig = create_figure(df, "philo10")
fig.show()


# In[23]:


fig = create_log_figure(table_time, table_error, "philo10", "pmc-sog_otf_couv99-default_2_8", "pnml2lts-mc_dfs_1_16")
fig.show()


# # Generate Figures

# In[24]:


# models
models = df.model.unique()

# tools 
tools = df.tool.unique()


# In[52]:


# create all the figures formula vs time

folder = os.path.join(OUTPUT_FOLDER, 'time-plots')
create_folder(folder)

for model in models:
    try:
        fig = create_figure(df, model)
        
        # save figures in html and pdf
        fig.write_html(os.path.join(folder, model + '.html'))
        fig.write_image(os.path.join(folder, model + '.pdf'))
    except KeyError:
        print("Error: {} was not plotted".format(model))


# In[ ]:


# create all the log figures

tools_pairs = [sorted(t) for t in (combinations(tools, 2))]

for plot, filter_method in plots.items():
    axes = list(filter(filter_method, tools_pairs))
    
    for model in models:
        folder = os.path.join(OUTPUT_FOLDER, plot, model)
        create_folder(folder)
        
        for axe in axes:
            try:
                show_strategy = plot == 'compare_couvreur_algorithm'
                fig = create_log_figure(table_time, table_error, model, axe[0], axe[1], show_strategy)
                
                # save figures in html and pdf
                figure_name = os.path.join(folder, '{}-{}-VS-{}-log'.format(model, axe[0], axe[1]))
                fig.write_html(figure_name + '.html')
                fig.write_image(figure_name + '.pdf')
            except KeyError:
                print("Error: {} was not plotted".format(model))


# In[ ]: