Bug on custom dataset

[qingyuanxingsi]

Describe the bug
A clear and concise description of what the bug is.
I construct a dataset with size 50000*6(feat_dim)*48(context length), finetune meeting the following error:
To Reproduce
Please provide a code snippet of a minimal reproducible example for the error.

/root/miniconda3/envs/llm_ts/bin/python3 -m cli.train -cp conf/finetune run_name=demo_run model=moirai_1.0_R_small data=feed_train val_data=feed_eval

Expected behavior
A clear and concise description of what you expected to happen.

Error message or code output
Paste the complete error message, including stack trace, or the undesired output that the above snippet produces.

AssertionError: no valid patch size candidates for time series shape: (6, 48), freq: H, patch_sizes: [8, 16, 32, 64, 128], constraints: range(32, 33), min_time_patches: 2, patch_size_ceil: 24

Environment

• Operating system:
• Python version:
• PyTorch version:
• uni2ts version (have you pulled the latest version?)

[chenghaoliu89]

The length of each training sample is required to be longer than min_time_patches (2). Here the patch size range is (32, 33), which indicates the training sample should be longer than 64. Your data (48 length) does not satisfy this constraints.

One quick fix is to modify the DEFAULT_RANGES in patch.py to "H": (1, 64),