metadata
dataset_info:
- config_name: 30V_Jan24
features:
- name: config
dtype: string
- name: traj_id
dtype: string
- name: shape
list: int64
- name: data
list:
list:
list: uint8
- name: left_right
list:
list: int64
- name: barycenter
list:
list: float64
splits:
- name: train
num_bytes: 1011799274
num_examples: 220
download_size: 32061722
dataset_size: 1011799274
- config_name: 60V_Dec24
features:
- name: config
dtype: string
- name: traj_id
dtype: string
- name: shape
list: int64
- name: data
list:
list:
list: uint8
- name: left_right
list:
list: int64
- name: barycenter
list:
list: float64
splits:
- name: train
num_bytes: 1107240702
num_examples: 605
download_size: 34858112
dataset_size: 1107240702
- config_name: default
features:
- name: config
dtype: string
- name: traj_id
dtype: string
- name: shape
list: int64
- name: data
list:
list:
list: uint8
- name: left_right
list:
list: int64
- name: barycenter
list:
list: float64
splits:
- name: train
num_bytes: 2119039976
num_examples: 825
download_size: 66925984
dataset_size: 2119039976
configs:
- config_name: 30V_Jan24
data_files:
- split: train
path: 30V_Jan24/train-*
- config_name: 60V_Dec24
data_files:
- split: train
path: 60V_Dec24/train-*
- config_name: default
data_files:
- split: train
path: data/train-*
Descriptions
Converting script
import pickle
from pathlib import Path
import numpy as np
from datasets import Dataset
DATA_DIR = Path("/path/to/cached/hugging_face/datasets/for/MLDS-NUS/Experimental_Images")
# should end with something like "snapshots/fd299418e9435f8fd98956a3f0a7344d208cc142"
def calc_left_right(data: np.ndarray):
left_rights = []
for im in data:
nonzero_columns = (im != 0).any(axis=-2)
left = nonzero_columns.argmax() if nonzero_columns.any() else -1
# Find the rightmost non-zero column
right = len(nonzero_columns) - 1 - nonzero_columns[::-1].argmax() if nonzero_columns.any() else -1
left_right = np.array([left, right])
left_rights.append(left_right)
left_rights = np.stack(left_rights, axis=0) # shape: (seq_len, 2)
return left_rights
def calc_barycenter(data: np.ndarray) -> np.ndarray:
"""
Calculate the barycenter of the polymer from the snapshot.
Assumes snapshot shape is (100, 500).
"""
xx = np.arange(data.shape[-2]).reshape(-1, 1)
bary_x = (data * xx).sum(axis=(-2, -1)) / data.sum(axis=(-2, -1))
yy = np.arange(data.shape[-1]).reshape(1, -1)
bary_y = (data * yy).sum(axis=(-2, -1)) / data.sum(axis=(-2, -1))
barycenter = np.stack([bary_x, bary_y], axis=-1) # (seq_len, 2)
return barycenter
def gen():
for folder in ["30V_Jan24", "60V_Dec24"]:
with open(DATA_DIR / f"{folder}.pkl", "rb") as f:
data = pickle.load(f)
for k, v in data.items():
frame = np.clip(v, 0, 255).astype(np.uint8) # save memory
left_rights = calc_left_right(255 - frame)
barycenters = calc_barycenter(255 - frame)
yield {
"config": folder,
"traj_id": k,
"shape": list(frame.shape),
"data": frame,
"left_right": left_rights,
"barycenter": barycenters,
}
ds = Dataset.from_generator(gen)
ds = ds.with_format("numpy")
ds.push_to_hub("MLDS-NUS/polymer-dynamics_experimental-data")
# upload by configs
def gen(folder: str):
with open(DATA_DIR / f"{folder}.pkl", "rb") as f:
data = pickle.load(f)
for k, v in data.items():
frame = np.clip(v, 0, 255).astype(np.uint8)
left_rights = calc_left_right(255 - frame)
barycenters = calc_barycenter(255 - frame)
yield {
"config": folder,
"traj_id": k,
"shape": list(frame.shape),
"data": frame,
"left_right": left_rights,
"barycenter": barycenters,
}
for config_name in ["30V_Jan24", "60V_Dec24"]:
ds = Dataset.from_generator(lambda cn=config_name: gen(cn))
ds = ds.with_format("numpy")
ds.push_to_hub(
"MLDS-NUS/polymer-dynamics_experimental-data",
config_name=config_name,
data_dir=f"{config_name}",
)
How to use
Directly loading by datasets is supported now!
from datasets import load_dataset
import numpy as np
hf_dataset_30V = load_dataset("MLDS-NUS/polymer-dynamics_experimental-data", config_name="30V_Jan24")
hf_dataset_60V = load_dataset("MLDS-NUS/polymer-dynamics_experimental-data", config_name="60V_Jan24")
hf_dataset_30V = hf_dataset_30V.with_format("numpy")["train"]
hf_dataset_60V = hf_dataset_60V.with_format("numpy")["train"]
for sample in hf_dataset_30V:
for k, v in sample.items():
if isinstance(v, np.ndarray):
print(f"{k}: {type(v)}, shape={v.shape}, dtype={v.dtype}")
else:
print(f"{k}: {v}")
output:
config: 30V_Jan24
traj_id: 30V_Tra_0
shape: <class 'numpy.ndarray'>, shape=(3,), dtype=int64
data: <class 'numpy.ndarray'>, shape=(160, 100, 350), dtype=int64
left_right: <class 'numpy.ndarray'>, shape=(160, 2), dtype=int64
barycenter: <class 'numpy.ndarray'>, shape=(160, 2), dtype=float32
How to contribute
import numpy as np
from datasets import Dataset
def gen(config_name: str):
for data in your_database_retriever(config_name):
frame = ...
traj_id = ...
shape = ...
left_rights = ...
barycenters = ...
yield {
"config": config_name,
"traj_id": traj_id,
"shape": shape,
"data": data, # a np.ndarray object of shape `shape`
"left_right": left_rights,
"barycenter": barycenters,
}
config_name = ...
ds = Dataset.from_generator(lambda cn=config_name: gen(cn))
ds = ds.with_format("numpy")
ds.push_to_hub(
"MLDS-NUS/polymer-dynamics_experimental-data",
config_name=config_name,
data_dir=f"{config_name}",
)