def return_summary_statistics_for_rp(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1,
nt='rp',
nm='eth',
cluster_sizes_of_choice={
'1': 1,
'3': 3,
'6': 6
}):
if not d:
d = {'nm': nm, 'nt': nt, 'wl': wl, 't': range(10, 30 + 1)}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
df = {}
s = {}
for k, v in cluster_sizes_of_choice.iteritems():
df[k] = rfd(df_filtered, d={'nn': v})[measurement_of_interest]
s[k] = df[k].describe()
for x in s.itervalues():
x['range'] = x['max'] - x['min']
v = pd.DataFrame(s).reset_index()
set_display_format_for_floats(format_='{:.2g}'.format)
return v.to_latex(index=False)
def return_summary_statistics_for_vms(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1):
if not d:
d = {'nt': 'vm', 'wl': wl, 'nn': nn, 't': range(10, 30 + 1)}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
df_1gb = rfd(df_filtered, d={'ram': '1GB'})[measurement_of_interest]
df_2gb = rfd(df_filtered, d={'ram': '2GB'})[measurement_of_interest]
df_4gb = rfd(df_filtered, d={'ram': '4GB'})[measurement_of_interest]
s = df_1gb.describe()
t = df_2gb.describe()
u = df_4gb.describe()
for x in [s, t, u]:
x['range'] = x['max'] - x['min']
v = pd.DataFrame(dict(ram1GB=s, ram2GB=t, ram4GB=u)).reset_index()
return v.to_latex(index=False)
def get_max_absolute_deviation_from_ref(
csv_file='combined_results_revised.csv',
wl='a',
list_of_number_of_nodes=None,
csv_reference_data_file='abramova_results.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
d_ref=None,
t_range=None,
node_type='rp',
nm='eth'):
# Initialize Variables----------------------------
if not list_of_number_of_nodes:
list_of_number_of_nodes = [1, 3, 6]
if not t_range:
t_range = range(10, 30 + 1)
if not d:
d = {'nt': node_type, 'wl': wl, 'nn': 1, 't': t_range, 'nm': nm}
if not d_ref:
d_ref = d
# ------------------------------------------------
# Read from CSV File -----------------------------
df = pd.read_csv(csv_file)
df_ref = pd.read_csv(csv_reference_data_file)
# ------------------------------------------------
# For each node, get the max's and min's
bound_dict = {}
for node_count in list_of_number_of_nodes:
d['nn'] = node_count # adjust the filter(s)
d_ref['nn'] = node_count
df_flt = rfd(df, d)
df_ref_flt = rfd(df_ref, d_ref)
df_summary = df_flt.describe()
df_ref_summary = df_ref_flt.describe()
max_ = df_summary[measurement_of_interest]['max']
min_ = df_summary[measurement_of_interest]['min']
max_ref = df_ref_summary[measurement_of_interest]['max']
min_ref = df_ref_summary[measurement_of_interest]['min']
bound_dict[node_count] = max(
abs(max_ - min_ref),
abs(min_ - max_ref)) # assumes one mean is always greater
bound = max(bound_dict.values())
return bound
def generate_bound_statements_rp_wired_and_wireless(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1,
nt='rp',
nm='eth'):
s = 'The median value of the corresponding wired experiment will serve as the reference in this paragraph. '
d_wlan = {'nm': 'wlan', 'nt': nt, 'wl': wl, 't': range(10, 30 + 1)}
d_eth = {'nm': 'eth', 'nt': nt, 'wl': wl, 't': range(10, 30 + 1)}
ref_values = {
'a': {
1: 58430,
3: 65650,
6: 87310
},
'c': {
1: 88000,
3: 90210,
6: 118090
},
'e': {
1: 223180,
3: 330820,
6: 404660
}
}
df = pd.read_csv(csv_file)
df_filtered_eth = rfd(df, d=d_eth)
df_filtered_wlan = rfd(df, d=d_wlan)
for node in [1, 3, 6]:
df_eth = rfd(df_filtered_eth, d={'nn': node})[measurement_of_interest]
df_wlan = rfd(df_filtered_wlan, d={'nn':
node})[measurement_of_interest]
summary_eth = df_eth.describe()
summary_wlan = df_wlan.describe()
s += generate_bound_statement(
max_=summary_wlan['max'],
min_=summary_wlan['min'],
series='a node network of {}'.format(node),
ref=summary_eth['50%'])
return s
def return_summary_statistics(
csv_file='results/exp0_cstress/cassandra_stress_results_compression_strategy.csv',
measurement_of_interest='op/s',
op='reads_only',
d=None,
compression_strategy='LZ4Compressor',
network_type='wired'):
if not d:
d = {
'compression': compression_strategy,
'op': op,
'network_type': network_type
}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
# df = {}
# s = {}
# df = rfd(df_filtered, d={'nn': v})[measurement_of_interest]
s = df_filtered[measurement_of_interest].describe()
for x in [s]:
x['range'] = x['max'] - x['min']
v = pd.DataFrame(s).reset_index()
set_display_format_for_floats(format_='{:.5g}'.format)
return v.to_latex(index=False)
def return_general_summary_table(
main_csv_file='combined_results_revised.csv',
reference_csv_file='abramova_results.csv',
trial_list=None,
measurement_of_interest='[OVERALL] RunTime(ms)',
desired_index=None,
desired_columns=None,
desired_summary_function=np.median):
if not trial_list:
trial_list = range(10, 30 + 1)
if not desired_index:
desired_index = ['nn', 'wl', 'dbs']
if not desired_columns:
desired_columns = ['nt', 'nm', 'ram']
df = pd.read_csv(main_csv_file)
df = rfd(df=df, d={'t': trial_list})
if reference_csv_file:
df = df.append(pd.read_csv(reference_csv_file))
table = pd.pivot_table(df,
values=measurement_of_interest,
index=desired_index,
columns=desired_columns,
aggfunc=desired_summary_function)
return table
def anova_for_variation_in_ram_1(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None):
if not d:
d = {'nt': 'vm', 'wl': 'a', 'nn': 1}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
df_1gb = rfd(df_filtered, d={'ram': '1GB'})[measurement_of_interest]
df_2gb = rfd(df_filtered, d={'ram': '2GB'})[measurement_of_interest]
df_4gb = rfd(df_filtered, d={'ram': '4GB'})[measurement_of_interest]
results = return_anova_summary_table(df_1gb, df_2gb, df_4gb)
return results
def return_entire_speedup_table(
workload,
comparison_description,
measurement_of_interest='[OVERALL] RunTime(ms)',
csv_file=None):
df = pd.read_csv(csv_file)
dd = []
d = [None, None]
df_filtered = [None, None]
speedup_dictionary = {}
for cluster_size in range(1, 6 + 1) + [[1, 2, 3, 4, 5, 6]]:
if comparison_description == 'rp_v_vm':
nm = ['nodal', 'eth']
nt = ['vm', 'rp']
elif comparison_description == 'wlan_v_eth':
nm = ['eth', 'wlan']
nt = ['rp', 'rp']
else:
nm = 'error'
nt = 'error'
for i in [0, 1]:
d[i] = {
'nt': nt[i],
'wl': workload,
'ram': '1GB',
'nm': nm[i],
'nn': cluster_size,
't': range(10, 30 + 1)
}
df_filtered[i] = rfd(df=df, d=d[i])
x = df_filtered[0][measurement_of_interest]
y = df_filtered[1][measurement_of_interest]
speedup_dictionary = return_speedup_stats(x, y)
if cluster_size == [1, 2, 3, 4, 5, 6]:
speedup_dictionary['cluster_size'] = 'OVERALL'
else:
speedup_dictionary['cluster_size'] = cluster_size
dd.append(speedup_dictionary)
table_in_dataframe_format = pd.DataFrame(dd)
table_in_dataframe_format = table_in_dataframe_format.transpose()
return table_in_dataframe_format.to_latex()
def generate_bound_statements_rp_wired(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1,
nt='rp',
nm='eth'):
s = ''
d = None
if not d:
d = {'nm': nm, 'nt': nt, 'wl': wl, 't': range(10, 30 + 1)}
ref_values = {
'a': {
1: 58430,
3: 65650,
6: 87310
},
'c': {
1: 88000,
3: 90210,
6: 118090
},
'e': {
1: 223180,
3: 330820,
6: 404660
}
}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
for node in [1, 3, 6]:
df = rfd(df_filtered, d={'nn': node})[measurement_of_interest]
summary = df.describe()
s += generate_bound_statement(
max_=summary['max'],
min_=summary['min'],
series='a node network of {}'.format(node),
ref=ref_values[wl][node])
return s
def anova_for_variation_in_ram(csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1):
if not d:
d = {'nt': 'vm', 'wl': wl, 'nn': nn, 't': range(10, 30 + 1)}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
df_1gb = rfd(df_filtered, d={'ram': '1GB'})[measurement_of_interest]
df_2gb = rfd(df_filtered, d={'ram': '2GB'})[measurement_of_interest]
df_4gb = rfd(df_filtered, d={'ram': '4GB'})[measurement_of_interest]
return return_embedded_latex_tables(
latex_table_as_string=return_anova_summary_table(
df_1gb, df_2gb, df_4gb),
label='ram_variance_analysis_workload_' + wl + '_' + str(nn) + '_node',
caption='ANOVA Summary Table for '
'Workload {}, {} Node Cluster'.format(wl.capitalize(), nn))
def return_summary_statistics_tabular(
workload='a',
nt='vm',
ram='1GB',
nm='nodal',
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)'):
ss = []
for cluster_size in [1, 2, 3, 4, 5, 6, [1, 2, 3, 4, 5, 6]]:
d = {
'nt': nt,
'wl': workload,
'ram': ram,
'nm': nm,
'nn': cluster_size,
't': range(10, 30 + 1)
}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)[measurement_of_interest]
if not df_filtered.empty:
s = df_filtered.describe()
s['range'] = s['max'] - s['min']
if cluster_size == [1, 2, 3, 4, 5, 6]:
s['cluster_size'] = "Overall"
else:
s['cluster_size'] = cluster_size
ss.append(s)
v = pd.DataFrame(ss).reset_index()
w = pd.pivot_table(v, columns=['cluster_size'])
return w.to_latex(index=True)
def return_max_min_range_for_all_levels_of_ram(
csv_file='combined_results_revised.csv',
measurement_of_interest='[OVERALL] RunTime(ms)',
d=None,
wl='a',
nn=1):
if not d:
d = {'nt': 'vm', 'wl': wl, 'nn': nn, 't': range(10, 30 + 1)}
df = pd.read_csv(csv_file)
df_filtered = rfd(df, d=d)
df_Xgb = df_filtered[measurement_of_interest]
s = df_Xgb.describe()
for x in [s]:
x['range'] = x['max'] - x['min']
return s['max'], s['min'], s['range']
def speedup_analysis_tables(csv_file,
comparison_description,
workload,
measurement_of_interest='[OVERALL] RunTime(ms)'):
s = ':::something went wrong generating the speedup analysis tables:::'
df = pd.read_csv(csv_file)
label = '{}_{}'.format(comparison_description, workload)
reference_statement = '' # initialize
if comparison_description == 'ram_v_ram':
reference_statement = 'See Table \\ref{{{}}}.'.format(
'table:{}'.format(label))
dd = {}
dd['slope'] = []
dd['intercept'] = []
dd['r_value'] = []
dd['p_value'] = []
dd['std_err'] = []
dd['cluster_size'] = []
for cluster_size in range(1, 6 + 1):
d = {
'nt': 'vm',
'wl': workload,
'nn': cluster_size,
't': range(10, 30 + 1)
}
df_filtered = rfd(df=df, d=d)
df_filtered['ram_in_gb'] = df_filtered['ram'].map(
convert_ram_text_to_gb)
if not df_filtered.empty:
x = df_filtered['ram_in_gb']
y = df_filtered[measurement_of_interest]
slope, intercept, r_value, p_value, std_err = stats.linregress(
x, y)
dd['slope'].append(slope)
dd['intercept'].append(intercept)
dd['r_value'].append(r_value)
dd['p_value'].append(p_value)
dd['std_err'].append(std_err)
dd['cluster_size'].append(cluster_size)
dd = pd.DataFrame(dd)
s = return_embedded_latex_tables(
latex_table_as_string=dd.to_latex(index=False,
columns=[
'cluster_size', 'slope',
'intercept', 'r_value',
'p_value', 'std_err'
]),
caption='Linear Regression over amount of RAM',
label=label)
elif comparison_description in ['rp_only', 'wlan_only']:
reference_statement = 'See Table \\ref{{{}}}.'.format(
'table:{}'.format(label))
dd = {}
dd['slope'] = []
dd['intercept'] = []
dd['r_value'] = []
dd['p_value'] = []
dd['std_err'] = []
if comparison_description == 'rp_only':
nm = 'eth'
elif comparison_description == 'wlan_only':
nm = 'wlan'
else:
nm = 'error'
d = {
'nt': 'rp',
'wl': workload,
'ram': '1GB',
'nm': nm,
't': range(10, 30 + 1)
}
df_filtered = rfd(df=df, d=d)
x = df_filtered['nn']
y = df_filtered[measurement_of_interest]
slope, intercept, r_value, p_value, std_err = stats.linregress(x, y)
dd['slope'].append(slope)
dd['intercept'].append(intercept)
dd['r_value'].append(r_value)
dd['p_value'].append(p_value)
dd['std_err'].append(std_err)
dd = pd.DataFrame(dd)
s = return_embedded_latex_tables(
latex_table_as_string=dd.to_latex(index=False,
columns=[
'slope', 'intercept',
'r_value', 'p_value',
'std_err'
]),
caption='Linear Regression over Cluster Size, Workload {}'.format(
workload.capitalize()),
label=label)
elif comparison_description in ['rp_v_vm', 'wlan_v_eth']:
dd = {}
dd['slope'] = []
dd['intercept'] = []
dd['r_value'] = []
dd['p_value'] = []
dd['std_err'] = []
dd['cluster_size'] = []
for cluster_size in range(1, 6 + 1):
d = {
'nt': ['vm', 'rp'],
'wl': workload,
'nn': cluster_size,
'ram': '1GB',
't': range(10, 30 + 1)
}
# Tune the filter
if comparison_description in ['rp_v_vm']:
d['nm'] = ['eth', 'nodal']
d['nt'] = ['rp', 'vm']
elif comparison_description in ['wlan_v_eth']:
d['nm'] = ['eth', 'wlan']
d['nt'] = 'rp'
df_filtered = rfd(df=df, d=d)
if comparison_description in ['rp_v_vm']:
name_of_new_column = 'is_limited_hardware'
df_filtered[name_of_new_column] = df_filtered['nt'].map(
assign_contrast_rp_v_vm)
elif comparison_description in ['wlan_v_eth']:
name_of_new_column = 'is_wireless_lan'
df_filtered[name_of_new_column] = df_filtered['nm'].map(
assign_contrast_wired_v_wireless)
else:
name_of_new_column = 'error_occurred'
x = df_filtered[name_of_new_column]
y = df_filtered[measurement_of_interest]
slope, intercept, r_value, p_value, std_err = stats.linregress(
x, y)
dd['slope'].append(slope)
dd['intercept'].append(intercept)
dd['r_value'].append(r_value)
dd['p_value'].append(p_value)
dd['std_err'].append(std_err)
dd['cluster_size'].append(cluster_size)
cluster_size = [1, 2, 3, 4, 5, 6]
# -- Now append the Overall, over 1,2,3,4,5,6 -- #
d['nn'] = cluster_size
df_filtered = rfd(df=df, d=d)
if comparison_description in ['rp_v_vm']:
name_of_new_column = 'is_limited_hardware'
df_filtered[name_of_new_column] = df_filtered['nt'].map(
assign_contrast_rp_v_vm)
elif comparison_description in ['wlan_v_eth']:
name_of_new_column = 'is_wireless_lan'
df_filtered[name_of_new_column] = df_filtered['nm'].map(
assign_contrast_wired_v_wireless)
x = df_filtered[name_of_new_column]
y = df_filtered[measurement_of_interest]
slope, intercept, r_value, p_value, std_err = stats.linregress(x, y)
dd['slope'].append(slope)
dd['intercept'].append(intercept)
dd['r_value'].append(r_value)
dd['p_value'].append(p_value)
dd['std_err'].append(std_err)
dd['cluster_size'].append('OVERALL')
# --- Now convert to dataframe ---
dd = pd.DataFrame(dd)
if comparison_description in ['rp_v_vm']:
caption = 'Linear Regression over the effect of limited hardware, Workload {}. The designation (NaN) indicates that data was not collected for this cluster size. r- values in the high nineties indicate that there is a pronounced effect. Lower r-values indicate a less pronounced effect, likely attributed to high variance.'.format(
workload.capitalize())
elif comparison_description in ['wlan_v_eth']:
caption = 'Linear Regression over the effect of 802.11 links, Workload {}. The designation (NaN) indicates that data was not collected for this cluster size. r-values in the high nineties indicate that there is a pronounced effect. Lower r-values indicate a less pronounced effect, likely attributed to high variance.'.format(
workload.capitalize())
else:
caption = 'Something went wrong with the caption assignment.'
insert0 = return_embedded_latex_tables(
latex_table_as_string=dd.to_latex(index=False,
columns=[
'cluster_size', 'slope',
'intercept', 'r_value',
'p_value', 'std_err'
]),
caption=caption,
label=label)
# -- Get the speedup statistics --
if comparison_description in ['rp_v_vm']:
caption_for_speedup = 'Speedup over the effect of limited hardware, Workload {}'.format(
workload.capitalize())
elif comparison_description in ['wlan_v_eth']:
caption_for_speedup = 'Speedup over the effect of 802.11 links, Workload {}'.format(
workload.capitalize())
else:
caption_for_speedup = 'Something went wrong with the caption assignment.'
label_for_speedup = label + '_speedup'
entire_speedup_table = return_entire_speedup_table(
workload=workload,
comparison_description=comparison_description,
measurement_of_interest='[OVERALL] RunTime(ms)',
csv_file=csv_file)
reference_statement = 'See Tables \\ref{{{}}} and \\ref{{{}}}.'.format(
'table:{}'.format(label), 'table:{}'.format(label_for_speedup))
insert1 = return_embedded_latex_tables(
latex_table_as_string=entire_speedup_table,
caption=caption_for_speedup,
label=label_for_speedup)
s = '\n\n' + insert0 + '\n\n' + insert1 + '\n\n'
elif comparison_description in ['rp_v_ref', 'vm_v_ref']:
s = ''
s = reference_statement + '\n\n' + s
return s