Name		Name	Last commit message	Last commit date
parent directory ..
0.1_Parse_DEBS_UCR_BTC.ipynb		0.1_Parse_DEBS_UCR_BTC.ipynb
0.2_Create_agg_data.ipynb		0.2_Create_agg_data.ipynb
0.3_Fig_construction.ipynb		0.3_Fig_construction.ipynb
0.4_Fig_metrics.ipynb		0.4_Fig_metrics.ipynb
1.1_Visual_representativity.ipynb		1.1_Visual_representativity.ipynb
1.2_Performance.ipynb		1.2_Performance.ipynb
1.3_LTTB_datapoint_selection.ipynb		1.3_LTTB_datapoint_selection.ipynb
README.md		README.md
downsample_times.pkl		downsample_times.pkl

README.md

Main notebooks

note: More information regarding the OR-conv mask can be found in this repository - or conv mask section

Name	Description
🔧 Data parsing
parse DEBS UCR	This notebook parses the DEBS 2012, DEBS 2013, UCR time series archive CINECG, and a bitcoin dataset. The parsed data is also stored in a more convenient parquet format.
create agg data	This notebook creates and stores the (reference and) aggregated time series for the datasets parsed in the previous notebook.
Aggregated Figure Generation	This notebook performs an extensive grid search over the following parameters: - Aggregation algorithm (LTTB / MinMaxLTTB / MinMax) - MinMaxLTTB `MinMax preselection ratio` $r_{ps}$. - `n_out`: the number of aggregated datapoints - Template time series and its data size - Visualization toolkit: plotly - Visualization configuration: interpolation = 'linear' line-width = 2 For each combination of parameters, a figure is generated and stored in the path_conf's `figure_root_dir` folder.
Figure Metrics computation	This notebook computes the (D)SSIM and MSE metrics for the aggregated figures generated in the previous notebook. Moreover, an `or-conv` mask is utilized to mitigate the variable range that which template time-series span over the image
core 📷
Visual Representativeness	This notebook asesses the visual representativeness using the image metrics from the above rows. Specifically this notebook highlights: - The trend of the aggregator performance over `n_out` - Analyzing hte effect of line width and anti-aliasing - Showing toolkit robustness via the noise data - Distribution plots of the aggregator performances - Dynamic frames showing the performance curves for varying `n_out` / line-width / toolkits
MinMaxLTTB performance	This notebook assesses the performance improvement of MinMaxLTTB compared to LTBB. Specifically, the performanc of: - MinMaxLTTB is compared to LTBB for the float32 value datatype - The same comparison is performed but MinMaxLTTB now leverages its paralellization capabilities.
LTTB data point selection analysis	This notebook investigates the properties of the data points which LTTB eventually selects. In conclusion LTTB favors selecting: - extrema datapoints that reside near the left bin edge - data points that alternate with the previous extrema. These two properties can be attributed to LTTB using the largest triangular surface to select data points.