https://github.com/msmathcomp/hyperbolic-tsne
Tip revision: bba9d0f089659fb170c7270aa90c796f91bfb2b1 authored by Martin Skrodzki on 02 May 2024, 12:34:19 UTC
Update README.md
Update README.md
Tip revision: bba9d0f
table_average_iteration_vs_data_set_size_per_exact_or_accelerated_tsne.py
"""
This script creates a table with statistical values (minimum, average, standard deviation, maximum) on the run times of
embedding iterations for the different data sets. Specifically, these data are compared for the accelerated iterations
that use the polar quad tree acceleration structure as well as for the non-accelerated, exact hyperbolic t-SNE
implementation.
"""
###########
# IMPORTS #
###########
from pathlib import Path
import pandas as pd
import numpy as np
####################
# READING THE DATA #
####################
results_path = Path("../results/samples_per_data_set/")
df = pd.read_csv(results_path.joinpath("overview.csv"))
timings_dfs = []
for record in df.to_records():
timing_df = pd.read_csv(record.run_directory.replace(".", str(results_path)) + "/timings.csv")
timing_df = timing_df[(timing_df.time_type == "tot_gradient")]
for cn in df.columns:
timing_df[cn] = record[cn]
timings_dfs.append(timing_df)
timings_df = pd.concat(timings_dfs, axis=0, ignore_index=True)
timings_df["early_exag"] = np.repeat(False, timings_df.shape[0])
timings_df.loc[timings_df.it_n <= 250, "early_exag"] = True
del timings_dfs
############################
# COMPUTING THE STATISTICS #
############################
# Work with the "equal length" data, as this splitting technique proved to be more efficient, filtering by
# "equal_length" contains both accelerated and exact data.
plot_times_df = timings_df.copy()
plot_times_df = plot_times_df[(plot_times_df.splitting_strategy == "equal_length")]
# Filter out only the exact, i.e., non-accelerated data
plot_times_df_exact = plot_times_df.copy()
plot_times_df_exact = plot_times_df_exact[(plot_times_df_exact.tsne_type == "exact")]
# Print Min, Avg, Std, Max of the timings per dataset per size
grouped = plot_times_df_exact.groupby(["dataset", "sample_size"])
print("Statistics exact:")
print(grouped["total_time"].min())
print(grouped["total_time"].mean())
print(grouped["total_time"].std())
print(grouped["total_time"].max())
# Filter out only the accelerated data, i.e., the data using the polar quad tree
plot_times_df_accelerated = plot_times_df.copy()
plot_times_df_accelerated = plot_times_df_accelerated[(plot_times_df_accelerated.tsne_type == "accelerated")]
# Print Min, Avg, Std, Max of the timings per dataset per size
grouped = plot_times_df_accelerated.groupby(["dataset", "sample_size"])
print("Statistics accelerated:")
print(grouped["total_time"].min())
print(grouped["total_time"].mean())
print(grouped["total_time"].std())
print(grouped["total_time"].max())