ConTSG-Bench: A Unified Benchmark for Conditional Time Series Generation

Shaocheng Lan; Shuqi Gu; Zhangzhi Xiong; Kan Ren

cs.LGMar 5, 2026

ConTSG-Bench: A Unified Benchmark for Conditional Time Series Generation

Authors: Shaocheng Lan, Shuqi Gu, Zhangzhi Xiong, Kan Ren

Organizations: School of Information Science and Technology, ShanghaiTech University, Shanghai, China.

Abstract

Conditional time series generation plays a critical role in addressing data scarcity and enabling causal analysis in real-world applications. Despite its increasing importance, the field lacks a standardized and systematic benchmarking framework for evaluating generative models across diverse conditions. To address this gap, we introduce the Conditional Time Series Generation Benchmark (ConTSG-Bench). ConTSG-Bench comprises a large-scale, well-aligned dataset spanning diverse conditioning modalities and levels of semantic abstraction, first enabling systematic evaluation of representative generation methods across these dimensions with a comprehensive suite of metrics for generation fidelity and condition adherence. Both the quantitative benchmarking and in-depth analyses of conditional generation behaviors have revealed the traits and limitations of the current approaches, highlighting critical challenges and promising research directions, particularly with respect to precise structural controllability and downstream task utility under complex conditions.

Figures & tables

Figure 1: Conditional time series generation with varying conditioning modalities (text, attribute, class label) and semantic abstraction levels (morphological vs. conceptual).

Existing Benchmark
	Condition Modality			Condition Semantic		Evaluation Dimensions
Method	Text	Attr	Label	Morph	Concept	Adh.	Fine-gr.	Comp. Gen.	Down. Util.
TSGBench Ang et al. (2023)	✗	✗	✓	✗	✗	✗	✗	✗	✓
Label-conditioned
TimeVQVAE Lee et al. (2023)	✗	✗	✓	✗	✗	✗	✗	✗	✓
TTS-CGAN Li et al. (2022)	✗	✗	✓	✗	✗	✗	✗	✗	✓

Table 1: Comparison of conditional time series generation methods and benchmarks along three dimensions: (1) supported condition modalities, (2) semantic abstraction levels, and (3) evaluation dimensions beyond fidelity, which is universally assessed. Abbreviations: Attr = Attribute; Morph = Morphological; Adh. = Condition Adherence; Fine-gr. = Fine-grained control; Comp. Gen. = Compositional generalization; Down. Util. = Downstream utility.

Figure 2: Model ranking under two metric groups: (left) generation fidelity that evaluates marginal distribution of generated time series; (right) condition adherence that evaluates joint/conditional alignment between time series and conditions.

Figure 3: Morphological vs. conceptual conditioning: absolute performance. DTW and CRPS on PTB-XL and Weather under the two condition types.

Figure 4: Fine-grained control evaluation. Left: Joint shapelet classification accuracy on Synth-U, where all three segment-level local patterns must be correctly generated. Middle: Segment retrieval accuracy (Acc@1) as a function of candidate pool size on TelecomTS-Segment. Right: Segment–text temporal order accuracy on TelecomTS-Segment.

Figure 5: Compositional generalization analysis. Left: normalized retrieval accuracy for head (closest 20% to training distribution) vs. tail (farthest 20%, novel attribute combinations) test samples; points below the diagonal indicate performance degradation on out-of-distribution combinations. Right: accuracy gap (tail

-

head) for each model, where negative values reflect sensitivity to novel attribute combinations.

Figure 6: Drop rate distribution across datasets and models. Lower drop rate indicates better substitutability of generated data.

Appendix figures & tables49 assets

Supplementary material from the paper’s appendix.

Appendix

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	41.09	42.0	60	8.90
Validation	41.21	42.0	60	8.94
Test	41.21	42.0	60	8.92

Appendix

Table 2: Summary of token number statistics for Synth-U dataset.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	62.23	63.0	83	8.97
Validation	62.27	63.0	83	9.10
Test	62.39	63.0	82	9.17

Appendix

Table 3: Summary of token number statistics for Synth-M dataset.

Table 10

Figure 7: Overall Pipeline of LLM Caption Generation

Attribute	Value Options	Definition
Particulate_Matter_Profile	Consistently Low, Low with a Significant Spike, Moderate and Fluctuating, High and Worsening, Consistently High, High with Significant Improvement	The overall 24-hour trend and severity of PM2.5 and PM10.
Ozone_Peak_Intensity	Suppressed or No Peak, Moderate Peak, Strong/Very High Peak	The strength of the afternoon ozone (O3) peak.
Inverse_Relationship_Strength	No Clear Relationship, Weak Relationship, Strong and Clear Relationship	The clarity of the inverse relationship between O3 and primary pollutants (NO2, CO).
Primary_Pollution_Event_Timing	No Specific Event, Morning Peak, Midday/Afternoon Peak, Evening/Nighttime Peak	The main time of day when primary pollutants (PM, CO, NO2) are highest.

Appendix

Table 6: Attribute Set for Air Quality Profile.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	220.00	212.0	611	54.85
Validation	221.61	213.0	508	55.64
Test	218.08	209.0	472	54.33

Appendix

Table 7: Summary of token number statistics for Airquality Beijing dataset.

Figure 8: Case Visualization of Airquality Beijing Data

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	168.14	165.0	386	31.06
Validation	167.30	164.0	302	30.07
Test	168.67	165.0	323	31.27

Appendix

Table 8: Summary of token number statistics for Telecomts-Segment dataset.

Figure 9: Case Visualization of TelecomTS Data

Attribute	Value Options	Definition
level_shift_presence	multiple_shifts, no_shifts, other, single_downward_shift, single_upward_shift	Indicates if there are abrupt changes in the average level of the series.
overall_trend	downward_trend, fluctuating_trend, mixed_trend, no_clear_trend, other, stable_level, upward_trend	Captures the long-term trajectory and persistent directional movement of the signal.
periodicity	absent, other, present, unclear	Assesses the existence of repetitive cycles or patterns occurring at fixed time intervals.
prominent_features	minor_fluctuations, other, peaks_and_troughs, peaks_only, troughs_only	Characterizes the most distinct visual landmarks or morphological events within the data.
volatility_level	decreasing_volatility, high_volatility, increasing_volatility, low_volatility, mixed_volatility, moderate_volatility, other	Evaluates the intensity and temporal evolution of variance and fluctuations in the series.

Appendix

Table 9: Attribute Set for ETTm1.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	43.37	42.0	86	7.54
Validation	43.84	43.0	81	8.03
Test	44.24	43.0	74	7.71

Appendix

Table 10: Summary of token number statistics for ETTm1 dataset.

Figure 10: Case Visualization of ETTm1 Data

Attribute	Value Options	Definition
extreme_points	absent, other, present	Denotes the existence of significant local maxima (peaks) or minima (troughs) within the signal.
level_shifts	absent, other, present_distinct, present_gradual, step_wise	Specifies the occurrence and specific characteristics of sudden transitions in the series’ baseline.
overall_trend	downward, mixed, other, stable, upward	Represents the predominant long-term trajectory of the data over the entire observation window.
periodicity	absent, other, present	Evaluates whether the sequence demonstrates rhythmic or recurring temporal structures.
volatility_change	decreasing, increasing, other, stable_volatility	Tracks the evolution of fluctuation intensity, indicating if the variance expands or contracts over time.
volatility_level	high, low, moderate, no_volatility, other	Quantifies the overall magnitude of oscillations and noise levels present in the time series.

Appendix

Table 11: Attribute Set for Istanbul Traffic.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	45.64	44.0	288	11.60
Validation	45.40	43.0	188	11.73
Test	44.41	43.0	122	11.22

Appendix

Table 12: Summary of token number statistics for Istanbul Traffic dataset.

Figure 11: Case Visualization of Istanbul Traffic Data

Attribute	Value Options	Definition
rhythm	SR, AFIB, AFLT, STACH, SBRAD, SARRH, PACE, SVARR, BIGU, TRIGU, SVTAC, PSVT, unknown	Primary rhythm code (highest confidence)
hr_cat	bradycardia, normal, tachycardia	HR $<$ 60 / 60-100 / $>$ 100 bpm
rr_regularity	regular, mild_irregular, irregular	RR CV $<$ 0.05 / 0.05-0.12 / $>$ 0.12
qrs_cat	normal, borderline, wide	QRS $<$ 100 / 100-120 / $>$ 120 ms
qtc_cat	normal, borderline, prolonged	QTc $<$ 450 / 450-480 / $>$ 480 ms
st_anterior	normal, mild_elevation, high_elevation, mild_depression, high_depression	ST deviation in V1-V4

Appendix

Table 13: Morphological Condition Attribute Set for PTB-XL

Attribute	Value Options	Definition
age_group	young, middle_aged, elderly	Age $<$ 40 / 40-65 / $>$ 65
sex	male, female	From metadata
diagnosis	43 diagnostic codes (e.g., NORM, IMI, ASMI, LVH, …)	Primary diagnosis (highest confidence)
heart_axis	normal, LAD, RAD, ALAD, ARAD, unknown	From metadata

Appendix

Table 14: Conceptual Condition Attribute Set for PTB-XL

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	25.94	24.0	51	5.07
Validation	26.17	24.0	50	5.17
Test	26.16	24.0	50	4.98

Appendix

Table 15: Summary of token number statistics for PTB-XL morphological text data.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	16.29	16.0	29	4.37
Validation	16.13	16.0	28	4.17
Test	16.19	16.0	27	4.19

Appendix

Table 16: Summary of token number statistics for PTB-XL conceptual text data.

Figure 12: Case Visualization of PTB-XL Data

Attribute	Value Options	Definition
humidity_level	dry, extremely_high, high, low, medium, moderately_dry, saturated, somewhat_humid, unknown, very_high	A qualitative representation of the ambient moisture concentration in the atmosphere.
pressure_level	average, high, low, unknown, very_high, very_low	Reflects the barometric status and atmospheric pressure fluctuations.
season	autumn, spring, summer, unknown, winter	Identifies the specific climatological period of the year for the given data.
sky_condition	broken_clouds, clear, cloudy, fog, other, partly_cloudy, passing_clouds, precipitation, scattered_clouds, sunny	Characterizes the visual appearance of the firmament and the density of cloud coverage.
temperature_level	chilly, cool, high, low, medium, unknown, warm	Provides a non-numeric evaluation of the thermal state of the environment.
time_of_day	afternoon, early_morning, evening, morning, night, unknown	Categorizes the observation into broad diurnal temporal segments.

Appendix

Table 17: Conceptual Condition Attribute Set for Weather.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	142.49	140.0	276	28.91
Validation	146.09	143.0	251	31.04
Test	144.45	142.0	269	31.14

Appendix

Table 18: Summary of token number statistics for Weather morphological text data.

Set	Average Tokens	Median Tokens	Max Tokens	Std. Dev.
Training	70.10	69.0	130	10.05
Validation	73.40	73.0	125	10.64
Test	72.87	73.0	112	10.58

Appendix

Table 19: Summary of token number statistics for Weather conceptual text data.

Figure 13: Case Visualization of Weather Data

Category	Statistical	Embedding-based
Generation Fidelity	• MDD (Marginal Distribution Difference) • ACD (Auto-Correlation Difference) • SD / KD (Skewness / Kurtosis Difference)	• FID (Fréchet Inception Distance) • Precision & Recall
Condition Adherence	• DTW Score (Alignment Distance) • CRPS (Distribution Calibration and Sharpness)	• CTTP Score (Cosine Similarity) • J-FTSD (Joint-Space FID) • Joint Precision & Recall

Appendix

Table 20: Taxonomy of Evaluation Metrics for ConTSG

Dataset	Acc (%)	Dataset	Acc (%)
TelecomTS	37.19	ETTm1	16.59
TelecomTS-Segment	16.09	Istanbul Traffic	17.45
PTB-XL (Morph.)	28.31	Synth-M	98.35
PTB-XL (Concept)	33.85	Synth-U	92.52
Weather (Morph.)	20.52	Air Quality	18.37
Weather (Concept)	20.90

Appendix

Table 21: CTTP validation accuracy across different datasets. The values represent the maximum validation accuracy achieved during training.

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$\bm{0.0314}_{\pm 0.0116}$	$0.0974_{\pm 0.0775}$	$0.6417_{\pm 0.0562}$	$0.0282_{\pm 0.0001}$	$50.1814_{\pm 3.8816}$	$55.1401_{\pm 2.8398}$	$0.5906_{\pm 0.0411}$	$0.0745_{\pm 0.0104}$	$0.8228_{\pm 0.0404}$	$0.4752_{\pm 0.0240}$	$22.9919_{\pm 0.7649}$
DiffuSETS	$0.0552_{\pm 0.0113}$	$0.1404_{\pm 0.1382}$	$0.8331_{\pm 0.8645}$	$0.0203_{\pm 0.0050}$	$41.5913_{\pm 8.8965}$	$46.5630_{\pm 10.9944}$	$0.5842_{\pm 0.0770}$	$\bm{0.2054}_{\pm 0.0257}$	$0.8427_{\pm 0.1608}$	$0.6600_{\pm 0.1285}$	$24.6628_{\pm 4.6115}$
T2S	$0.0489_{\pm 0.0098}$	$0.4960_{\pm 0.4106}$	$0.7833_{\pm 0.4301}$	$0.0358_{\pm 0.0165}$	$144.4827_{\pm 27.3339}$	$156.3209_{\pm 27.2652}$	$0.2330_{\pm 0.1987}$	$0.0000_{\pm 0.0000}$	$0.2862_{\pm 0.0746}$	$0.0140_{\pm 0.0109}$	$10.3251_{\pm 4.3402}$
TEdit	$0.0658_{\pm 0.0015}$	$0.1632_{\pm 0.1069}$	$0.2876_{\pm 0.1147}$	$0.0216_{\pm 0.0056}$	$49.1590_{\pm 7.0552}$	$59.9647_{\pm 6.4816}$	$0.5448_{\pm 0.0081}$	$0.1217_{\pm 0.0344}$	$0.7358_{\pm 0.0081}$	$0.3991_{\pm 0.0576}$	$20.3756_{\pm 0.7434}$
Text2Motion	$0.0821_{\pm 0.0067}$	$0.0715_{\pm 0.0413}$	$0.2046_{\pm 0.0673}$	$\bm{0.0150}_{\pm 0.0029}$	$58.5985_{\pm 12.4265}$	$66.6559_{\pm 11.6903}$	$0.5509_{\pm 0.0702}$	$0.0562_{\pm 0.0302}$	$0.7030_{\pm 0.0355}$	$0.3052_{\pm 0.0937}$	$17.6067_{\pm 1.7010}$
TimeVQVAE	$0.0800_{\pm 0.0016}$	$0.0457_{\pm 0.0244}$	$0.7758_{\pm 0.0165}$	$0.0245_{\pm 0.0012}$	$75.7424_{\pm 2.1271}$	$83.9404_{\pm 2.0995}$	$\bm{0.7228}_{\pm 0.0130}$	$0.0170_{\pm 0.0011}$	$0.7329_{\pm 0.0121}$	$0.2192_{\pm 0.0138}$	$16.0165_{\pm 0.2354}$

Appendix

Table 22: Evaluation results for Synth-U dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.0580_{\pm 0.0239}$	$0.1382_{\pm 0.1703}$	$0.4220_{\pm 0.2471}$	$0.0294_{\pm 0.0003}$	$46.1556_{\pm 12.6821}$	$53.5891_{\pm 12.6868}$	$0.6962_{\pm 0.0760}$	$0.1510_{\pm 0.0483}$	$0.4030_{\pm 0.0832}$	$0.2333_{\pm 0.0861}$	$18.8397_{\pm 0.5783}$
DiffuSETS	$\bm{0.0413}_{\pm 0.0190}$	$0.0847_{\pm 0.0714}$	$\bm{0.1596}_{\pm 0.1405}$	$0.0209_{\pm 0.0030}$	$35.6547_{\pm 6.2623}$	$42.0699_{\pm 9.1293}$	$0.6668_{\pm 0.1098}$	$0.2551_{\pm 0.0755}$	$0.5958_{\pm 0.2316}$	$0.4658_{\pm 0.1855}$	$21.2378_{\pm 4.6499}$
T2S	$0.0753_{\pm 0.0360}$	$0.3972_{\pm 0.1165}$	$1.4948_{\pm 0.1724}$	$0.0305_{\pm 0.0047}$	$113.6724_{\pm 7.3416}$	$123.0193_{\pm 7.4631}$	$0.4246_{\pm 0.0932}$	$0.0000_{\pm 0.0000}$	$0.1523_{\pm 0.0379}$	$0.0096_{\pm 0.0025}$	$1.6491_{\pm 3.4922}$
TEdit	$0.0556_{\pm 0.0061}$	$0.0314_{\pm 0.0065}$	$0.2552_{\pm 0.1086}$	$0.0155_{\pm 0.0016}$	$35.1944_{\pm 0.9255}$	$43.8138_{\pm 1.1236}$	$0.7398_{\pm 0.0177}$	$0.2298_{\pm 0.0203}$	$0.6117_{\pm 0.0138}$	$0.3923_{\pm 0.0090}$	$21.2387_{\pm 0.3449}$
Text2Motion	$0.0764_{\pm 0.0132}$	$0.0572_{\pm 0.0455}$	$0.2479_{\pm 0.0465}$	$0.0140_{\pm 0.0007}$	$60.4473_{\pm 16.6599}$	$65.1781_{\pm 16.1543}$	$0.8230_{\pm 0.0588}$	$0.0746_{\pm 0.0694}$	$0.6024_{\pm 0.0818}$	$0.2304_{\pm 0.1378}$	$19.9581_{\pm 1.7206}$
TimeVQVAE	$0.0733_{\pm 0.0028}$	$\bm{0.0282}_{\pm 0.0148}$	$0.6307_{\pm 0.1129}$	$0.0211_{\pm 0.0006}$	$60.6209_{\pm 0.4303}$	$66.8617_{\pm 0.3628}$	$\bm{0.8671}_{\pm 0.0062}$	$0.0531_{\pm 0.0037}$	$0.5932_{\pm 0.0054}$	$0.2115_{\pm 0.0043}$	$19.8635_{\pm 0.1381}$

Appendix

Table 23: Evaluation results for Synth-M dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$\bm{0.0337}_{\pm 0.0069}$	$0.2094_{\pm 0.0560}$	$0.7371_{\pm 0.0695}$	$0.0290_{\pm 0.0012}$	$159.6945_{\pm 20.3313}$	$160.6358_{\pm 20.4623}$	$0.8957_{\pm 0.0040}$	$0.0458_{\pm 0.0079}$	$0.8081_{\pm 0.0362}$	$0.2602_{\pm 0.0690}$	$5.2102_{\pm 0.7037}$
DiffuSETS	$0.0467_{\pm 0.0051}$	$0.1189_{\pm 0.0173}$	$\bm{0.3604}_{\pm 0.1177}$	$0.0249_{\pm 0.0016}$	$121.7182_{\pm 10.4377}$	$122.5601_{\pm 10.4545}$	$0.9183_{\pm 0.0111}$	$0.0466_{\pm 0.0323}$	$0.8985_{\pm 0.0229}$	$0.4681_{\pm 0.0790}$	$6.1099_{\pm 0.2019}$
T2S	$0.1079_{\pm 0.0264}$	$0.3482_{\pm 0.1443}$	$1.5353_{\pm 0.2622}$	$0.0382_{\pm 0.0045}$	$223.7563_{\pm 25.8692}$	$225.3871_{\pm 25.8658}$	$0.0029_{\pm 0.0040}$	$\bm{0.3919}_{\pm 0.2264}$	$0.1564_{\pm 0.0922}$	$0.0281_{\pm 0.0088}$	$-1.2634_{\pm 1.6210}$
TEdit	$0.0369_{\pm 0.0082}$	$\bm{0.0975}_{\pm 0.0171}$	$0.5143_{\pm 0.0824}$	$\bm{0.0213}_{\pm 0.0008}$	$109.7493_{\pm 5.9814}$	$110.7139_{\pm 5.9993}$	$0.8953_{\pm 0.0242}$	$0.1456_{\pm 0.0287}$	$0.8407_{\pm 0.0087}$	$0.5160_{\pm 0.0375}$	$4.8901_{\pm 0.5026}$
Text2Motion	$0.0518_{\pm 0.0023}$	$0.1444_{\pm 0.0659}$	$0.5000_{\pm 0.0190}$	$0.0246_{\pm 0.0003}$	$\bm{83.4466}_{\pm 5.3609}$	$\bm{84.2614}_{\pm 5.3351}$	$0.9126_{\pm 0.0092}$	$0.1368_{\pm 0.0201}$	$\bm{0.9610}_{\pm 0.0023}$	$\bm{0.7296}_{\pm 0.0086}$	$\bm{6.9590}_{\pm 0.1399}$
TimeVQVAE	—	—	—	$0.0469_{\pm 0.0199}$	—	—	—	—	—	—	—

Appendix

Table 24: Evaluation results for AirQuality Beijing dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.1069_{\pm 0.0210}$	$\bm{0.4925}_{\pm 0.0322}$	$3.3521_{\pm 0.0656}$	$0.0272_{\pm 0.0021}$	$42.0076_{\pm 9.4701}$	$42.6005_{\pm 9.5725}$	$0.6698_{\pm 0.0357}$	$0.4616_{\pm 0.0276}$	$0.6089_{\pm 0.0069}$	$0.4793_{\pm 0.0305}$	$1.8156_{\pm 0.3851}$
DiffuSETS	$0.1420_{\pm 0.0860}$	$8.5331_{\pm 11.3647}$	$258.7216_{\pm 428.3528}$	$0.0424_{\pm 0.0130}$	$56.4434_{\pm 37.3998}$	$56.9916_{\pm 37.5302}$	$0.6126_{\pm 0.4295}$	$0.3859_{\pm 0.2482}$	$0.5409_{\pm 0.1071}$	$0.4806_{\pm 0.2584}$	$1.9019_{\pm 1.1863}$
T2S	$0.1150_{\pm 0.1321}$	$1.1269_{\pm 1.5301}$	$3.1332_{\pm 1.8773}$	$0.0540_{\pm 0.0060}$	$95.0026_{\pm 49.3734}$	$95.6860_{\pm 49.4863}$	$0.1115_{\pm 0.0891}$	$0.3323_{\pm 0.2293}$	$0.2886_{\pm 0.1804}$	$0.2387_{\pm 0.1566}$	$\bm{3.4514}_{\pm 0.9474}$
TEdit	$0.1692_{\pm 0.0495}$	$0.7493_{\pm 0.1854}$	$3.8786_{\pm 0.3975}$	$\bm{0.0233}_{\pm 0.0058}$	$\bm{23.1472}_{\pm 1.4568}$	$\bm{23.8535}_{\pm 1.4339}$	$0.6089_{\pm 0.0985}$	$0.5883_{\pm 0.0525}$	$0.4828_{\pm 0.0635}$	$0.5748_{\pm 0.0152}$	$1.9972_{\pm 0.0725}$
Text2Motion	$0.1114_{\pm 0.0418}$	$0.8648_{\pm 1.1884}$	$8.8360_{\pm 13.6306}$	$0.0243_{\pm 0.0050}$	$48.5253_{\pm 14.8678}$	$48.9968_{\pm 14.9018}$	$0.6105_{\pm 0.1036}$	$0.2676_{\pm 0.1474}$	$\bm{0.6579}_{\pm 0.0434}$	$0.4560_{\pm 0.0629}$	$1.9365_{\pm 0.4914}$
TimeVQVAE	$0.0530_{\pm 0.0183}$	$1.9516_{\pm 0.1991}$	$9.3044_{\pm 5.1511}$	$0.0358_{\pm 0.0025}$	$28.2658_{\pm 6.1412}$	$28.7677_{\pm 6.1357}$	$\bm{0.7515}_{\pm 0.0630}$	$0.3961_{\pm 0.0843}$	$0.6195_{\pm 0.0233}$	$\bm{0.6089}_{\pm 0.0345}$	$2.1357_{\pm 0.1940}$

Appendix

Table 25: Evaluation results for ETTm1 dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.0294_{\pm 0.0019}$	$0.4812_{\pm 0.2556}$	$0.1500_{\pm 0.1943}$	$0.0363_{\pm 0.0051}$	$14.5301_{\pm 2.1501}$	$15.0547_{\pm 2.1321}$	$\bm{0.9347}_{\pm 0.0086}$	$0.2896_{\pm 0.0140}$	$0.6885_{\pm 0.0511}$	$0.5944_{\pm 0.0409}$	$4.7030_{\pm 0.0194}$
DiffuSETS	$0.1470_{\pm 0.0883}$	$2.2346_{\pm 1.5908}$	$12.9962_{\pm 11.1741}$	$0.0408_{\pm 0.0191}$	$28.6764_{\pm 21.1976}$	$29.3117_{\pm 21.3957}$	$0.5917_{\pm 0.3045}$	$0.3923_{\pm 0.3925}$	$0.6023_{\pm 0.1723}$	$0.6127_{\pm 0.1566}$	$4.7812_{\pm 0.0942}$
T2S	$0.2451_{\pm 0.0331}$	$0.3192_{\pm 0.3463}$	$1.5178_{\pm 1.0697}$	$0.0281_{\pm 0.0067}$	$142.4714_{\pm 16.1469}$	$143.6973_{\pm 16.1239}$	$0.0000_{\pm 0.0000}$	$0.1653_{\pm 0.1908}$	$0.0052_{\pm 0.0042}$	$0.0708_{\pm 0.0754}$	$\bm{5.4458}_{\pm 0.1697}$
TEdit	$0.0310_{\pm 0.0013}$	$0.2893_{\pm 0.1829}$	$0.2992_{\pm 0.2275}$	$0.0176_{\pm 0.0020}$	$7.4086_{\pm 1.4342}$	$8.1024_{\pm 1.4414}$	$0.8277_{\pm 0.0097}$	$0.8405_{\pm 0.0470}$	$0.6223_{\pm 0.0059}$	$0.7407_{\pm 0.0267}$	$4.7114_{\pm 0.0205}$
Text2Motion	$0.0573_{\pm 0.0095}$	$0.1450_{\pm 0.0801}$	$\bm{0.0779}_{\pm 0.0754}$	$\bm{0.0102}_{\pm 0.0015}$	$\bm{2.7859}_{\pm 1.9849}$	$\bm{3.1654}_{\pm 1.9895}$	$0.8548_{\pm 0.0071}$	$0.7271_{\pm 0.0547}$	$\bm{0.8559}_{\pm 0.0074}$	$0.8218_{\pm 0.0083}$	$4.8171_{\pm 0.0368}$
TimeVQVAE	$0.2015_{\pm 0.2004}$	$0.3223_{\pm 0.1256}$	$2.0618_{\pm 2.2354}$	$0.0401_{\pm 0.0049}$	$33.9157_{\pm 29.6866}$	$34.7945_{\pm 29.9070}$	$0.4548_{\pm 0.5252}$	$0.1102_{\pm 0.1273}$	$0.3241_{\pm 0.3565}$	$0.3084_{\pm 0.2928}$	$4.7204_{\pm 0.2820}$

Appendix

Table 26: Evaluation results for Istanbul Traffic dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.0487_{\pm 0.0105}$	$0.9491_{\pm 0.4001}$	$9.1130_{\pm 2.3490}$	$0.0328_{\pm 0.0092}$	$91.1473_{\pm 30.6606}$	$97.2862_{\pm 34.8033}$	$\bm{0.6494}_{\pm 0.2329}$	$0.1973_{\pm 0.1341}$	$0.1758_{\pm 0.0312}$	$0.2073_{\pm 0.1534}$	$1.6884_{\pm 1.3374}$
DiffuSETS	$0.0510_{\pm 0.0225}$	$0.7260_{\pm 0.4397}$	$5.6203_{\pm 2.3342}$	$0.0237_{\pm 0.0061}$	$65.9887_{\pm 43.5030}$	$71.4901_{\pm 48.4934}$	$0.5795_{\pm 0.0342}$	$0.2395_{\pm 0.1685}$	$0.2928_{\pm 0.0737}$	$0.3366_{\pm 0.1251}$	$4.5457_{\pm 3.3877}$
T2S	$0.1037_{\pm 0.0461}$	$\bm{0.5110}_{\pm 0.3631}$	$7.4168_{\pm 1.4742}$	$0.0352_{\pm 0.0099}$	$155.1567_{\pm 53.1827}$	$161.8594_{\pm 51.7074}$	$0.6012_{\pm 0.1847}$	$0.0556_{\pm 0.0582}$	$0.1321_{\pm 0.0769}$	$0.0390_{\pm 0.0366}$	$0.1068_{\pm 1.1882}$
TEdit	$0.0661_{\pm 0.0202}$	$0.5190_{\pm 0.2698}$	$8.3735_{\pm 1.4039}$	$0.0240_{\pm 0.0027}$	$58.6366_{\pm 7.6197}$	$62.8779_{\pm 7.5856}$	$0.5381_{\pm 0.0291}$	$0.4052_{\pm 0.1134}$	$0.1658_{\pm 0.0115}$	$0.4103_{\pm 0.0534}$	$2.1875_{\pm 0.3157}$
Text2Motion	$0.0791_{\pm 0.0194}$	$0.9901_{\pm 0.5188}$	$6.9028_{\pm 7.4454}$	$\bm{0.0134}_{\pm 0.0013}$	$53.5293_{\pm 49.7279}$	$57.1443_{\pm 55.5941}$	$0.6083_{\pm 0.0382}$	$0.3191_{\pm 0.0470}$	$\bm{0.6771}_{\pm 0.0478}$	$\bm{0.5239}_{\pm 0.0683}$	$\bm{9.3332}_{\pm 5.8628}$
TimeVQVAE	$0.0945_{\pm 0.0048}$	$0.6221_{\pm 0.5406}$	$9.2415_{\pm 5.1114}$	$0.0356_{\pm 0.0044}$	$123.6608_{\pm 29.9672}$	$127.4633_{\pm 29.4431}$	$0.4742_{\pm 0.0994}$	$0.1486_{\pm 0.0431}$	$0.1206_{\pm 0.0329}$	$0.1557_{\pm 0.0355}$	$1.1549_{\pm 0.2186}$

Appendix

Table 27: Evaluation results for TelecomTS dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.0343_{\pm 0.0026}$	$19.5672_{\pm 0.0892}$	$2320.6266_{\pm 0.5614}$	$0.0205_{\pm 0.0002}$	$45.0646_{\pm 12.7574}$	$47.7810_{\pm 13.0501}$	$0.7533_{\pm 0.0367}$	$0.0292_{\pm 0.0035}$	$0.6301_{\pm 0.0803}$	$0.2497_{\pm 0.0523}$	$27.3101_{\pm 2.2800}$
DiffuSETS	$0.1379_{\pm 0.0601}$	$19.1210_{\pm 0.2734}$	$2317.8549_{\pm 1.5389}$	$0.0395_{\pm 0.0004}$	$149.2328_{\pm 16.4051}$	$156.1811_{\pm 15.6110}$	$0.5943_{\pm 0.5162}$	$0.0025_{\pm 0.0027}$	$0.1057_{\pm 0.0414}$	$0.0368_{\pm 0.0335}$	$16.0544_{\pm 1.6389}$
T2S	$0.0932_{\pm 0.0099}$	$18.8126_{\pm 0.2219}$	$2315.7835_{\pm 3.1762}$	$0.0303_{\pm 0.0120}$	$110.0314_{\pm 70.9398}$	$114.0980_{\pm 71.4178}$	$0.1852_{\pm 0.2814}$	$0.2594_{\pm 0.2369}$	$0.2071_{\pm 0.2433}$	$0.2154_{\pm 0.2516}$	$20.2179_{\pm 6.5903}$
TEdit	$0.0341_{\pm 0.0079}$	$23.4401_{\pm 3.5566}$	$2362.1122_{\pm 232.1081}$	$0.0087_{\pm 0.0006}$	$12.0106_{\pm 0.3898}$	$14.0126_{\pm 0.3361}$	$0.8595_{\pm 0.0103}$	$0.4530_{\pm 0.0183}$	$0.8321_{\pm 0.0394}$	$0.7543_{\pm 0.0223}$	$30.2036_{\pm 0.6333}$
Text2Motion	$0.0500_{\pm 0.0067}$	$18.1200_{\pm 0.0912}$	$2298.9082_{\pm 3.6489}$	$0.0077_{\pm 0.0002}$	$\bm{5.5376}_{\pm 0.3020}$	$\bm{6.8965}_{\pm 0.3269}$	$\bm{0.9215}_{\pm 0.0068}$	$\bm{0.5861}_{\pm 0.0076}$	$\bm{0.9726}_{\pm 0.0013}$	$\bm{0.9174}_{\pm 0.0080}$	$\bm{32.8174}_{\pm 0.0093}$
TimeVQVAE	$0.0707_{\pm 0.0160}$	$19.0337_{\pm 0.1169}$	$2319.5249_{\pm 4.1065}$	$0.0263_{\pm 0.0006}$	$47.5908_{\pm 4.9055}$	$53.5766_{\pm 5.1236}$	$0.3864_{\pm 0.0614}$	$0.0030_{\pm 0.0033}$	$0.2642_{\pm 0.0303}$	$0.1413_{\pm 0.0256}$	$22.9355_{\pm 0.3264}$

Appendix

Table 28: Evaluation results for Weather Conceptual dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$0.0434_{\pm 0.0056}$	$20.2137_{\pm 1.0936}$	$2332.3187_{\pm 23.7888}$	$0.0247_{\pm 0.0019}$	$64.8944_{\pm 13.2709}$	$66.2769_{\pm 13.2006}$	$0.6966_{\pm 0.0382}$	$0.0536_{\pm 0.0298}$	$0.5948_{\pm 0.0314}$	$0.3425_{\pm 0.0798}$	$9.3509_{\pm 0.6411}$
DiffuSETS	$0.1140_{\pm 0.0194}$	$19.3972_{\pm 0.0958}$	$2316.9628_{\pm 0.8818}$	$0.0351_{\pm 0.0073}$	$157.8052_{\pm 30.8667}$	$159.6305_{\pm 30.7734}$	$0.6047_{\pm 0.0700}$	$0.0000_{\pm 0.0000}$	$0.3559_{\pm 0.1084}$	$0.0432_{\pm 0.0213}$	$6.6785_{\pm 0.4084}$
T2S	$0.1051_{\pm 0.0062}$	$18.9384_{\pm 0.1181}$	$2307.1773_{\pm 15.1844}$	$0.0379_{\pm 0.0060}$	$192.1624_{\pm 16.7269}$	$194.2186_{\pm 16.1008}$	$0.0262_{\pm 0.0197}$	$0.0000_{\pm 0.0000}$	$0.1502_{\pm 0.0008}$	$0.0246_{\pm 0.0032}$	$5.9304_{\pm 0.2041}$
TEdit	$0.0678_{\pm 0.0148}$	$22.1650_{\pm 2.8657}$	$2419.3456_{\pm 169.2376}$	$0.0218_{\pm 0.0031}$	$65.5004_{\pm 3.5415}$	$67.7105_{\pm 3.5083}$	$0.6801_{\pm 0.1054}$	$0.2309_{\pm 0.0457}$	$0.4573_{\pm 0.0650}$	$0.3496_{\pm 0.0237}$	$7.0731_{\pm 0.0988}$
Text2Motion	$0.0537_{\pm 0.0011}$	$\bm{16.9457}_{\pm 0.8890}$	$2202.2627_{\pm 81.5057}$	$\bm{0.0116}_{\pm 0.0008}$	$28.5415_{\pm 3.2102}$	$29.2164_{\pm 3.2280}$	$\bm{0.8674}_{\pm 0.0180}$	$0.2289_{\pm 0.0476}$	$\bm{0.8948}_{\pm 0.0168}$	$\bm{0.7569}_{\pm 0.0231}$	$\bm{12.2402}_{\pm 0.2928}$
TimeVQVAE	$0.0743_{\pm 0.0107}$	$19.5004_{\pm 0.0346}$	$2320.3193_{\pm 5.6500}$	$0.0276_{\pm 0.0012}$	$56.0410_{\pm 1.7473}$	$58.1840_{\pm 1.7424}$	$0.4939_{\pm 0.1206}$	$0.0224_{\pm 0.0178}$	$0.4527_{\pm 0.0118}$	$0.2782_{\pm 0.0165}$	$7.1866_{\pm 0.1688}$

Appendix

Table 29: Evaluation results for Weather Morphological dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$\bm{0.0623}_{\pm 0.0109}$	$2.8098_{\pm 0.0209}$	$184.4093_{\pm 0.2316}$	$0.0149_{\pm 0.0004}$	$324.8132_{\pm 8.8372}$	$336.5941_{\pm 8.4960}$	$0.9994_{\pm 0.0011}$	$0.0000_{\pm 0.0000}$	$\bm{0.5886}_{\pm 0.0157}$	$0.0002_{\pm 0.0003}$	$219.3064_{\pm 2.9385}$
DiffuSETS	$0.1051_{\pm 0.0163}$	$5.8130_{\pm 2.0835}$	$212.3178_{\pm 45.9144}$	$0.0187_{\pm 0.0109}$	$247.9444_{\pm 75.7033}$	$277.3275_{\pm 56.9501}$	$0.7419_{\pm 0.3434}$	$\bm{0.0027}_{\pm 0.0023}$	$0.4659_{\pm 0.2270}$	$\bm{0.0224}_{\pm 0.0357}$	$234.0506_{\pm 9.7990}$
T2S	$0.2110_{\pm 0.0467}$	$3.2119_{\pm 0.2035}$	$177.6389_{\pm 4.7735}$	$0.0296_{\pm 0.0113}$	$264.7998_{\pm 77.9925}$	$298.9812_{\pm 71.9978}$	$0.5946_{\pm 0.1745}$	$0.0005_{\pm 0.0008}$	$0.4369_{\pm 0.1171}$	$0.0030_{\pm 0.0041}$	$251.9576_{\pm 41.4880}$
TEdit	$0.0861_{\pm 0.0166}$	$2.7860_{\pm 0.0251}$	$182.9535_{\pm 0.4488}$	$0.0123_{\pm 0.0031}$	$250.0269_{\pm 17.9467}$	$283.1505_{\pm 12.3875}$	$0.7050_{\pm 0.1150}$	$0.0000_{\pm 0.0000}$	$0.5127_{\pm 0.0309}$	$0.0015_{\pm 0.0027}$	$247.6307_{\pm 9.3556}$
Text2Motion	$0.3232_{\pm 0.0300}$	$3.6515_{\pm 0.3254}$	$\bm{176.4981}_{\pm 6.7947}$	$0.0184_{\pm 0.0003}$	$375.3458_{\pm 11.3807}$	$384.0842_{\pm 10.1483}$	$\bm{1.0000}_{\pm 0.0000}$	$0.0000_{\pm 0.0000}$	$0.4729_{\pm 0.0241}$	$0.0000_{\pm 0.0000}$	$202.0106_{\pm 3.7835}$
TimeVQVAE	$0.1330_{\pm 0.0069}$	$3.0693_{\pm 0.0762}$	$178.4336_{\pm 5.5833}$	$0.0169_{\pm 0.0001}$	$336.9363_{\pm 5.3872}$	$348.3181_{\pm 4.9416}$	$\bm{1.0000}_{\pm 0.0000}$	$0.0000_{\pm 0.0000}$	$0.5535_{\pm 0.0279}$	$0.0000_{\pm 0.0000}$	$213.6566_{\pm 2.1786}$

Appendix

Table 30: Evaluation results for PTB-XL Conceptual dataset

Model	ACD	SD	KD	MDD	FID	J-FTSD	Precision	Recall	Joint Precision	Joint Recall	CTTP Score
Bridge	$\bm{0.0561}_{\pm 0.0069}$	$2.8160_{\pm 0.0264}$	$184.2775_{\pm 0.0876}$	$0.0158_{\pm 0.0004}$	$500.8181_{\pm 26.0678}$	$513.6664_{\pm 22.7653}$	$0.5895_{\pm 0.3693}$	$0.0000_{\pm 0.0000}$	$0.3423_{\pm 0.0101}$	$0.0011_{\pm 0.0003}$	$140.6902_{\pm 12.3004}$
DiffuSETS	$0.1072_{\pm 0.0214}$	$\bm{2.6434}_{\pm 0.4425}$	$\bm{157.9633}_{\pm 33.0249}$	$0.0134_{\pm 0.0048}$	$313.8154_{\pm 93.6001}$	$346.6596_{\pm 89.2739}$	$0.4795_{\pm 0.2270}$	$0.0073_{\pm 0.0043}$	$0.4268_{\pm 0.0776}$	$0.0173_{\pm 0.0132}$	$211.1689_{\pm 39.4551}$
T2S	$0.1938_{\pm 0.0339}$	$3.5991_{\pm 0.9877}$	$176.7059_{\pm 6.9755}$	$0.0295_{\pm 0.0149}$	$399.7197_{\pm 70.8724}$	$437.2591_{\pm 62.4090}$	$0.9020_{\pm 0.0396}$	$0.0021_{\pm 0.0021}$	$0.4022_{\pm 0.0374}$	$0.0026_{\pm 0.0026}$	$150.1159_{\pm 18.8086}$
TEdit	$0.0896_{\pm 0.0067}$	$2.8194_{\pm 0.0438}$	$182.7912_{\pm 0.4054}$	$0.0107_{\pm 0.0013}$	$\bm{258.1951}_{\pm 37.9238}$	$\bm{292.2583}_{\pm 34.9973}$	$0.8967_{\pm 0.0563}$	$0.0009_{\pm 0.0010}$	$0.5207_{\pm 0.0606}$	$0.0047_{\pm 0.0024}$	$\bm{256.6325}_{\pm 15.3421}$
Text2Motion	$0.2502_{\pm 0.0827}$	$4.7425_{\pm 0.1136}$	$165.2220_{\pm 3.7201}$	$0.0192_{\pm 0.0001}$	$558.6359_{\pm 0.9435}$	$564.2688_{\pm 0.3375}$	$\bm{0.9998}_{\pm 0.0003}$	$0.0002_{\pm 0.0003}$	$0.3270_{\pm 0.0009}$	$0.0003_{\pm 0.0005}$	$115.7349_{\pm 0.4308}$
TimeVQVAE	$0.1271_{\pm 0.0131}$	$3.8563_{\pm 0.7084}$	$204.6594_{\pm 31.9805}$	$0.0167_{\pm 0.0004}$	$541.0195_{\pm 10.8172}$	$554.5409_{\pm 9.8160}$	$0.9472_{\pm 0.0359}$	$\bm{0.2059}_{\pm 0.1926}$	$0.2991_{\pm 0.0128}$	$\bm{0.0231}_{\pm 0.0257}$	$120.2263_{\pm 3.5252}$

Appendix

Table 31: Evaluation results for PTB-XL Morphological dataset

Figure 14: Generation Quality Comparison: Synth-U. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 15: Generation Quality Comparison: Synth-M. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 16: Generation Quality Comparison: AirQuality Beijing. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 17: Generation Quality Comparison: TelecomTS. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 18: Generation Quality Comparison: ETTm1. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 19: Generation Quality Comparison: Istanbul Traffic. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 20: Generation Quality Comparison: PTB-XL Conceptual. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 21: Generation Quality Comparison: PTB-XL Morphological. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 22: Generation Quality Comparison: WEATHER Conceptual. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 23: Generation Quality Comparison: WEATHER Morphological. Each subplot shows one model’s performance on one variable. Blue line: ground truth; Red line: generated median; Red bands: 25%-75% (dark) and 10%-90% (light) quantile ranges of 10 generated samples.

Figure 24: Morphological vs. conceptual conditioning: rank stability. Each point is a model with ranks computed separately under morphological (x-axis) and conceptual (y-axis) conditions (lower is better). Points near the diagonal indicate stable rankings across condition types, while off-diagonal points indicate sensitivity to condition semantics.

Figure 25: Temporal order confusion matrices on TelecomTS-Segment. Rows denote generated segments (TS $i$ ) and columns denote within-series captions (Text $j$ ); diagonal dominance indicates correct segment–text alignment.

Dataset	TEdit	TimeWeaver	DiffuSETS	TimeVQVAE	VerbalTS	WaveStitch	Text2Motion	Bridge	T2S	TTSCGAN
AirQuality Beijing	0.301	0.336	0.574	0.509	0.631	0.505	0.696	0.787	1.106	0.863
ETTm1	0.290	0.382	0.365	0.560	0.279	0.507	1.086	0.598	0.426	0.573
Istanbul Traffic	1.232	0.705	0.918	1.210	1.020	1.046	0.741	0.953	1.290	0.747
PTB-XL (Conceptual)	0.672	0.722	0.532	0.841	0.582	0.827	1.383	0.568	1.003	1.083
PTB-XL (Morphological)	0.320	0.624	0.202	0.607	0.499	0.620	0.585	0.719	0.645	0.546
Synth-M	0.093	0.096	0.538	0.420	0.098	0.571	0.537	0.730	0.905	0.989

Appendix

Table 32: Drop Rate by Model and Dataset (lower is better)

Explore similar work

GenTS: A Comprehensive Benchmark Library for Generative Time Series Models
May 18, 2026Chenxi Wang, Xiaorong Wang, Peiyang Li +1Time Series Generation Forecasting Benchmarks
The University of Hong Kong Hong Kong SAR, China · Fudan University Shanghai, China
UPLOTS: A Unified Pretrained Language Model for Constrained Time-series Generation
Jun 9, 2026Du Yin, Hao Xue, Jinliang Deng +4Constrained Generative Modeling Time Series Generation
University of New South Wales Sydney, NSW, Australia · HKUST(GZ) China · BUAA Beijing, China
Assessing Sample Quality in Conditional Generation under Compositional Shift
Jun 8, 2026Berker Demirel, Valentino Maiorca, Marco Fumero +2Distribution Shift Conditional Generation
Institute of Science and Technology Austria (ISTA) · Pyramidal Inc, San Francisco, CA, USA · Achira Inc, San Francisco, CA, USA

May 18, 2026J/K move · Enter open · S save

GenTS: A Comprehensive Benchmark Library for Generative Time Series Models

Chenxi Wang, Xiaorong Wang, Peiyang Li, Yi Wang

The University of Hong Kong Hong Kong SAR, China · Fudan University Shanghai, China

Time Series Generation Forecasting Benchmarks

Open paper AskPDF arXiv

Generative models have demonstrated remarkable potential in time series analysis tasks, like synthesis, forecasting, imputation, etc. However, offering limited coverage for generative models, existing time series libraries are mainly engineered for discriminative models, with standardized workflows for specific tasks, such as optimizing Mean Squared Errors for time series forecasting. This rigid structure is fundamentally incompatible with the distinct and often complex paradigms of generative models (e.g., adversarial training, diffusion processes), which learn the underlying data distribution rather than a direct input-output mapping. To this end, we proposed GenTS, a comprehensive and extensible benchmark library designed for systematic assessment on generative time series models. GenTS features a unified data preprocessing pipeline, a collection of versatile models, and panoramic evaluation metrics. Its modular design also enables the researchers to flexibly customize beyond our built-in datasets and models. Based on GenTS, we conducted benchmarking experiments under diverse tasks, accordingly offering suggestions for model selection and identifying potential directions for future research. Our codes are open-source at https://github.com/WillWang1113/GenTS. The official tutorials and document are available at https://willwang1113.github.io/GenTS/.

Abstract

Explore similar work

GenTS: A Comprehensive Benchmark Library for Generative Time Series Models

UPLOTS: A Unified Pretrained Language Model for Constrained Time-series Generation

Assessing Sample Quality in Conditional Generation under Compositional Shift