def benchmark_share_get_plot(b_data_share_get):
"""
This function plots the graph for various protocols benchmarks for additive
shared tensors
Args:
b_data_share_get (list): the sample data to approximate
Returns:
benchmark_share_get.png (png): plotted graph in graph/ast_benchmarks directory
"""
# initializing workers
worker = workers(hook())
# available protocols
protocols = ["snn", "fss"]
# initializing graph plot
fig, ax = plt.subplots()
for protocol in protocols:
# list for handling graph data
x_data = []
y_data = []
for data in b_data_share_get:
# getting value from b_data_share_get
dtype, n_workers = data
# temporary list for calculating average execution time and error
temp_time_taken = []
for i in range(10):
start_time = timeit.default_timer()
benchmark_share_get(worker, protocol, dtype, n_workers)
time_taken = timeit.default_timer() - start_time
temp_time_taken.append(time_taken)
final_time_taken = sum(temp_time_taken) / len(temp_time_taken)
final_time_taken *= 1000
x_data.append(dtype + str(" / ") + str(n_workers))
y_data.append(final_time_taken)
ax.plot(x_data, y_data, label=protocol, linestyle="-")
x_data.clear()
y_data.clear()
ax.set_xlabel("dtype / n_workers")
ax.set_ylabel("Execution Time (ms)")
ax.legend(bbox_to_anchor=(1, 1.22),
loc="upper right",
title="Protocols",
fontsize="small")
plt.tight_layout()
plt.savefig("../graphs/ast_benchmarks/benchmark_share_get.png")
def benchmark_batch_norm_plot(b_data_batch_norm):
"""
This function plots the graph for various protocols benchmarks for
batch_norm.
Args:
b_data_batch_norm (list): list of protocols to approximate
Returns:
benchmark_batch_norm.png (png): plotted graph in graph/ast_benchmarks directory
"""
# initializing workers
worker = workers(hook())
# getting data (protocols)
protocols = b_data_batch_norm
# initializing graph plot
fig, ax = plt.subplots()
for protocol in protocols:
# list for handling graph data
x_data = []
y_data = []
for prec_frac in range(1, 5):
temp_time_taken = []
for i in range(10):
start_time = timeit.default_timer()
benchmark_batch_norm(worker, protocol, True, prec_frac)
time_taken = timeit.default_timer() - start_time
temp_time_taken.append(time_taken)
final_time_taken = sum(temp_time_taken) / len(temp_time_taken)
final_time_taken *= 1000
y_data.append(final_time_taken)
x_data.append(prec_frac)
ax.plot(x_data, y_data, label=protocol, linestyle="-")
x_data.clear()
y_data.clear()
# plotting of the data
plt.title("benchmark_batch_norm")
ax.set_xlabel("Precision Value")
ax.set_ylabel("Execution Time (ms)")
ax.legend(bbox_to_anchor=(1, 1.3),
loc="upper right",
title="Protocol",
fontsize="small")
plt.tight_layout()
plt.savefig("../graphs/ast_benchmarks/benchmark_batch_norm.png")
def sigmoid_approximation_plot(benchmark_data_sigmoid):
"""
This function plots the graph for various sigmoidal approximation benchmarks namely
'chebyshev', 'maclaurin', 'exp'.
Args:
benchmark_data_sigmoid (list): the sample data to approximate
Returns:
sigmoid_function_approximations_benchmark (png): plotted graph in graph directory
"""
# initializing workers
worker = workers(hook())
# initializing graph plot
fig, ax1 = plt.subplots()
ax2 = ax1.twinx()
# list for handling graph data
x_data = []
y_data = []
y2_data = []
for data in benchmark_data_sigmoid:
# getting value from benchmark_data_sigmoid
method, prec_frac = data
for precision_value in range(1, (prec_frac + 1)):
# temporary list for calculating average execution time and error
temp_time_taken = []
temp_error = []
for i in range(10):
start_time = timeit.default_timer()
error = benchmark_sigmoid(method, precision_value, worker)
time_taken = timeit.default_timer() - start_time
temp_time_taken.append(time_taken)
temp_error.append(error)
final_time_taken = sum(temp_time_taken) / len(temp_time_taken)
final_time_taken *= 1000
final_error = sum(temp_error) / len(temp_error)
x_data.append(precision_value)
y_data.append(final_time_taken)
y2_data.append(final_error)
ax1.plot(x_data, y_data, label=method, linestyle="-")
ax2.plot(x_data, y2_data, label=method, linestyle="--")
x_data.clear()
y_data.clear()
y2_data.clear()
# plotting of the data
ax1.set_xlabel("Precision Value")
ax1.set_ylabel("Execution Time (ms)")
ax2.set_ylabel("Error")
ax1.legend(bbox_to_anchor=(1, 1.3), loc="upper right", title="Method", fontsize="small")
ax2.legend(bbox_to_anchor=(0, 1.3), loc="upper left", title="Error", fontsize="small")
plt.tight_layout()
plt.savefig("../graphs/sigmoid_function_approximations_benchmark.png")
