Datasets:

chlange
/

SubstrateMixRaman

@@ -23,11 +23,14 @@ configs:
 - config_name: default
   data_files:
   - split: train
-    path: train_*.csv
   - split: validation
-    path: val_*.csv
   - split: test
-    path: test_*.csv
 size_categories:
 - 1K<n<10K
 ---
@@ -38,7 +41,6 @@ This dataset, designed for biotechnological applications, provides a valuable re
 The primary motivation behind this dataset is to offer a **calibration resource** with **uncorrelated metabolite concentrations**, addressing a common challenge in biotechnological applications. By ensuring the independence of substance concentrations, the dataset is ideal for training robust models and for integration with use-case specific labeled spectra.
----
 ## Substances and Concentration Ranges
 The dataset includes the following eight pure substances (metabolites) chosen for their relevance to fermentation processes:
@@ -54,14 +56,12 @@ The dataset includes the following eight pure substances (metabolites) chosen fo
 Concentrations for each substance were randomly sampled from a uniform distribution within ranges relevant for bacterial fermentations. This approach ensures that all possible combinations of concentrations have an equal chance of appearing in the dataset, effectively preventing correlations among the substances. Lower concentrations were deliberately oversampled to enhance predictive accuracy for models trained on data within that range.
----
 ## Additional Components and Their Impact
 To better mimic real-world conditions encountered during bacterial fermentations, varying concentrations of **mineral salt medium** and **antifoam** were also included in the samples.
 * **Antifoam** significantly impacts signal intensity; increased amounts lead to greater turbidity and a decrease in overall signal. By incorporating varying antifoam concentrations, the dataset helps ensure that models trained on this data are **agnostic to changing concentrations of antifoam or mineral salt medium**, improving their applicability in diverse experimental setups.
----
 ## Data Acquisition
 Samples were prepared using a Liquid Handler (LiHa) system. The process involved:
@@ -71,4 +71,4 @@ Samples were prepared using a Liquid Handler (LiHa) system. The process involved
 3.  **Recording Raman spectra** of the prepared sample.
 More details on the creation process can be found in the paper: https://dx.doi.org/10.2139/ssrn.5239248
-As soon as this paper is published in a peer-reviewed journal, we will add the link.

 - config_name: default
   data_files:
   - split: train
+    path: "train_spectra.csv"
+    path: "train_label.csv"
   - split: validation
+    path: "val_spectra.csv"
+    path: "val_label.csv"
   - split: test
+    path: "test_spectra.csv"
+    path: "test_label.csv"
 size_categories:
 - 1K<n<10K
 ---
 The primary motivation behind this dataset is to offer a **calibration resource** with **uncorrelated metabolite concentrations**, addressing a common challenge in biotechnological applications. By ensuring the independence of substance concentrations, the dataset is ideal for training robust models and for integration with use-case specific labeled spectra.
 ## Substances and Concentration Ranges
 The dataset includes the following eight pure substances (metabolites) chosen for their relevance to fermentation processes:
 Concentrations for each substance were randomly sampled from a uniform distribution within ranges relevant for bacterial fermentations. This approach ensures that all possible combinations of concentrations have an equal chance of appearing in the dataset, effectively preventing correlations among the substances. Lower concentrations were deliberately oversampled to enhance predictive accuracy for models trained on data within that range.
 ## Additional Components and Their Impact
 To better mimic real-world conditions encountered during bacterial fermentations, varying concentrations of **mineral salt medium** and **antifoam** were also included in the samples.
 * **Antifoam** significantly impacts signal intensity; increased amounts lead to greater turbidity and a decrease in overall signal. By incorporating varying antifoam concentrations, the dataset helps ensure that models trained on this data are **agnostic to changing concentrations of antifoam or mineral salt medium**, improving their applicability in diverse experimental setups.
 ## Data Acquisition
 Samples were prepared using a Liquid Handler (LiHa) system. The process involved:
 3.  **Recording Raman spectra** of the prepared sample.
 More details on the creation process can be found in the paper: https://dx.doi.org/10.2139/ssrn.5239248
+As soon as this paper is published in a peer-reviewed journal, we will add the link.