In [ ]:

    

%matplotlib inline

import matplotlib
import matplotlib.pyplot as plt
import matplotlib.pylab as pylab
import numpy as np
import operator
import os
import unicodedata

from os import path
from tabulate import tabulate

pylab.rcParams['figure.figsize'] = 18, 12  # that's default image size for this interactive session


    

In [ ]:

    

experiments = [
    "bow_logreg",
    "wordvec_mlp_2_0",
    "wordvecpos_mlp_2_0"
]

directory = "../resources/results/results_semisupervised/"
verbs_file = "../resources/sensem/verbs"


    

In [ ]:

    

with open(verbs_file, "r") as f:
    verbs = unicodedata.normalize("NFC", f.read().decode("utf-8")).strip().split()

accuracies = {e: list() for e in experiments}
most_common_precision = {e: list() for e in experiments}
less_common_recall = {e: list() for e in experiments}

for vidx, verb in enumerate(verbs):
    vidx = "{:03}".format(vidx)

    if not path.isdir(path.join(directory, vidx)):
        continue

    for experiment in experiments:
        accuracy_file = path.join(directory, vidx, experiment, "accuracy")
        mcp_file = path.join(directory, vidx, experiment, "test_accuracy")
        lcr_file = path.join(directory, vidx, experiment, "less_common_recall")

        with open(accuracy_file, "r") as f:
            accuracies[experiment].append(
                np.mean([float(acc.strip()) for acc in f.readlines()])
            )
            
        with open(mcp_file, "r") as f:
            most_common_precision[experiment].append(
                np.mean([float(mcp.strip()) for mcp in f.readlines()])
            )
            
        with open(lcr_file, "r") as f:
            less_common_recall[experiment].append(
                np.mean([float(lcr.strip()) for lcr in f.readlines()])
            )


    

In [ ]:

    

best_experiments_accuracy =\
    sorted([(e, np.mean(acc)) for e, acc in accuracies.items()], key=operator.itemgetter(1), reverse=True)
best_experiments_mcp =\
    sorted([(e, np.mean(mcp)) for e, mcp in most_common_precision.iteritems()], key=operator.itemgetter(1), reverse=True)
best_experiments_lcr =\
    sorted([(e, np.mean(lcr)) for e, lcr in less_common_recall.iteritems()], key=operator.itemgetter(1), reverse=True)

def get_boxplot_accuracies():
    boxplot_accuracies = []

    for experiment, accuracy in best_experiments_accuracy:
        boxplot_accuracies.append(accuracies[experiment])

    return boxplot_accuracies

def get_boxplot_mcp():
    boxplot_mcp = []

    for experiment, mcp in best_experiments_mcp:
        boxplot_mcp.append(most_common_precision[experiment])

    return boxplot_mcp

def get_boxplot_lcr():
    boxplot_lcr = []

    for experiment, lcr in best_experiments_lcr:
        boxplot_lcr.append(less_common_recall[experiment])

    return boxplot_lcr


    

In [ ]:

    

plt.boxplot(get_boxplot_accuracies()[:10], labels=zip(*best_experiments_accuracy)[0][:10])

for experiment, value in best_experiments_accuracy[:10]:
    print "{}: {:.03f}".format(experiment, value)


    

In [ ]:

    

plt.boxplot(get_boxplot_mcp()[:10], labels=zip(*best_experiments_mcp)[0][:10])

for experiment, value in best_experiments_mcp[:10]:
    print "{}: {:.03f}".format(experiment, value)


    

In [ ]:

    

plt.boxplot(get_boxplot_lcr()[:10], labels=zip(*best_experiments_lcr)[0][:10])

for experiment, value in best_experiments_lcr[:10]:
    print "{}: {:.03f}".format(experiment, value)


    

In [ ]:

    

mean_accuracies = dict(best_experiments_accuracy)
mean_mcp = dict(best_experiments_mcp)
mean_lcr = dict(best_experiments_lcr)

header = ["Experiment", "Accuracy Mean", "MFC Precision Mean", "LFC Recall Mean"]
table = []

for experiment in experiments:
    table.append([
            experiment,
            mean_accuracies[experiment],
            mean_mcp[experiment],
            mean_lcr[experiment]
        ])

print tabulate(table, header, floatfmt=".3f")


    

In [ ]: