add model files

README.md

@@ -1,3 +1,88 @@
 ---
-license: mit
+license: cc-by-4.0
+tags:
+  - multilabel-classification
+  - scientific-literature
+  - tensorflow
+  - mlp
+  - chemistry
+datasets:
+  - Bocklitz-Lab/lit2vec-subfield-classifier-dataset
+language:
+  - en
+library_name: tensorflow
+model-index:
+  - name: MLP Subfield Classifier (Lit2Vec)
+    results: []
 ---
+
+# 🧪 MLP Multi-Label Classifier for Scientific Subfields
+
+A lightweight **multi-layer perceptron (MLP)** model trained to predict **multiple chemistry subfields** from precomputed Lit2Vec embeddings of scientific abstracts.
+
+Built using the [Lit2Vec Subfield Classifier Dataset](https://huggingface.co/datasets/Bocklitz-Lab/lit2vec-subfield-classifier-dataset), this model supports **multi-label classification** across 18 chemistry-related subfields.
+
+---
+
+## 📊 Subfields Predicted (18 labels)
+
+- Catalysis
+- Organic Chemistry
+- Polymer Chemistry
+- Inorganic Chemistry
+- Materials Science
+- Analytical Chemistry
+- Physical Chemistry
+- Biochemistry
+- Environmental Chemistry
+- Energy Chemistry
+- Medicinal Chemistry
+- Chemical Engineering
+- Supramolecular Chemistry
+- Radiochemistry & Nuclear Chemistry
+- Forensic & Legal Chemistry
+- Food Chemistry
+- Chemical Education
+- Others
+
+Each prediction returns **probabilities** per class (using sigmoid), and labels with probabilities > 0.5 are returned as positives.
+
+---
+
+## 🧠 Model Architecture
+
+- Input size: `768` (Lit2Vec embedding)
+- Hidden layers: `[256, 256]`
+- Activation: ReLU
+- Dropout: `0.3`
+- Output: 18 sigmoid units (multi-label)
+- Loss: Weighted binary crossentropy
+- Optimizer: Adam
+- Framework: TensorFlow 2 / Keras
+
+---
+
+## 🔧 Usage Example
+
+```python
+from huggingface_hub import hf_hub_download
+import tensorflow as tf
+import numpy as np
+import json
+
+# Load model
+model_path = hf_hub_download("your-username/my-mlp-model", "mlp_model.h5")
+model = tf.keras.models.load_model(model_path, compile=False)
+
+# Load label mapping
+label_map_path = hf_hub_download("your-username/my-mlp-model", "label_mapping.json")
+with open(label_map_path) as f:
+    label_mapping = json.load(f)["index_to_label"]
+
+# Predict on new embedding
+embedding = np.random.rand(768).tolist()  # or use Lit2Vec
+pred = model.predict(np.array([embedding]))[0]
+
+# Get labels with probability > 0.5
+labels = [label_mapping[str(i)] for i, p in enumerate(pred) if p > 0.5]
+print("Predicted labels:", labels)

config.yaml

@@ -0,0 +1,82 @@
+seed: 42
+mixed_precision: false
+
+dataset:
+  name: Bocklitz-Lab/lit2vec-subfield-classifier-dataset
+  train_split: train
+  val_split: validation
+  test_split: test
+  cache_dir: null
+  sample_limit: null      # e.g., 5000 for quick runs
+  shuffle_seed: 42
+
+model:
+  hidden_layers: [256, 256]
+  activation: relu
+  dropout: 0.3
+  batch_norm: true
+  learning_rate: 0.001
+  optimizer: adam
+
+training:
+  batch_size: 32
+  epochs: 50
+  threshold: 0.5
+  val_split_fraction: null  # used only if dataset has no explicit validation split
+  early_stopping:
+    enabled: true
+    monitor: val_loss
+    patience: 5
+    min_delta: 0.0
+  reduce_lr:
+    enabled: true
+    monitor: val_loss
+    factor: 0.5
+    patience: 3
+
+cross_validation:
+  enabled: true
+  n_folds: 5
+  shuffle: true
+  seed: 42
+
+class_weights:
+  enabled: true
+
+wandb:
+  enabled: true
+  project: mlp-multilabel-example
+  entity: null           # or your team/org
+  run_name: final_model
+  tags: [mlp, multilabel, lit2vec]
+
+io:
+  save_dir: saved_data
+  model_path: mlp_model.h5
+  history_path: training_history.json
+  test_x_path: test_embeddings.npy
+  test_y_path: test_labels.npy
+  label_mapping_path: label_mapping.json
+  f1_plot_path: f1_vs_freq.png
+
+# Optional: provide human-readable label names; if omitted, names will be "Label 0", ...
+labels:
+  label_to_index:
+    Catalysis: 0
+    Organic Chemistry: 1
+    Polymer Chemistry: 2
+    Inorganic Chemistry: 3
+    Materials Science: 4
+    Analytical Chemistry: 5
+    Physical Chemistry: 6
+    Biochemistry: 7
+    Environmental Chemistry: 8
+    Energy Chemistry: 9
+    Medicinal Chemistry: 10
+    Chemical Engineering: 11
+    Supramolecular Chemistry: 12
+    Radiochemistry & Nuclear Chemistry: 13
+    Forensic & Legal Chemistry: 14
+    Food Chemistry: 15
+    Chemical Education: 16
+    Others: 17

label_mapping.json

@@ -0,0 +1 @@
+{"label_to_index": {"Catalysis": 0, "Organic Chemistry": 1, "Polymer Chemistry": 2, "Inorganic Chemistry": 3, "Materials Science": 4, "Analytical Chemistry": 5, "Physical Chemistry": 6, "Biochemistry": 7, "Environmental Chemistry": 8, "Energy Chemistry": 9, "Medicinal Chemistry": 10, "Chemical Engineering": 11, "Supramolecular Chemistry": 12, "Radiochemistry & Nuclear Chemistry": 13, "Forensic & Legal Chemistry": 14, "Food Chemistry": 15, "Chemical Education": 16, "Others": 17}, "index_to_label": {"0": "Catalysis", "1": "Organic Chemistry", "2": "Polymer Chemistry", "3": "Inorganic Chemistry", "4": "Materials Science", "5": "Analytical Chemistry", "6": "Physical Chemistry", "7": "Biochemistry", "8": "Environmental Chemistry", "9": "Energy Chemistry", "10": "Medicinal Chemistry", "11": "Chemical Engineering", "12": "Supramolecular Chemistry", "13": "Radiochemistry & Nuclear Chemistry", "14": "Forensic & Legal Chemistry", "15": "Food Chemistry", "16": "Chemical Education", "17": "Others"}}

mlp_model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9fa090a94bd995924cff439616bb2996e155372e5ed52214ef28f5d1f0e76d3
+size 4061328

requirements.txt

@@ -0,0 +1,7 @@
+numpy==1.26
+scikit-learn==1.2
+tensorflow==2.15,<2.16
+matplotlib==3.7
+wandb==0.16
+datasets==2.19
+pyarrow==12

training_history.json

@@ -0,0 +1 @@
+{"loss": [8.470060348510742, 1.7850418090820312, 1.5056606531143188, 1.3478214740753174, 1.2950140237808228, 1.1986502408981323, 1.1413774490356445, 1.151053786277771, 1.093340516090393, 1.0624538660049438, 1.0414057970046997, 1.0123454332351685, 1.0043306350708008, 1.0047874450683594, 0.9833027124404907, 0.9159265160560608, 0.8808308243751526], "binary_accuracy": [0.8970237970352173, 0.9494465589523315, 0.9531154036521912, 0.9546661972999573, 0.9551639556884766, 0.9562273621559143, 0.9565128087997437, 0.9568208456039429, 0.9571045637130737, 0.9575257301330566, 0.9576963186264038, 0.9576197266578674, 0.958272397518158, 0.958265483379364, 0.9586291909217834, 0.959234893321991, 0.9597082734107971], "ROC_AUC": [0.8316560387611389, 0.9453409910202026, 0.9560701847076416, 0.9598875045776367, 0.9619840979576111, 0.9646556377410889, 0.9665591716766357, 0.9664934873580933, 0.967963457107544, 0.9682399034500122, 0.9691439867019653, 0.9693467020988464, 0.9692378640174866, 0.9697430729866028, 0.9695724844932556, 0.9717855453491211, 0.9725377559661865], "PR_AUC": [0.2990269064903259, 0.671329140663147, 0.7382231950759888, 0.7759764194488525, 0.785939633846283, 0.8087033033370972, 0.8194859027862549, 0.818324089050293, 0.8289315700531006, 0.835086464881897, 0.8388688564300537, 0.845190703868866, 0.8463453054428101, 0.847578763961792, 0.8502194881439209, 0.8617655634880066, 0.8688318729400635], "val_loss": [1.5648810863494873, 1.3782635927200317, 1.209653377532959, 1.2201181650161743, 1.1328657865524292, 1.1017009019851685, 1.207519292831421, 1.1550023555755615, 1.0857306718826294, 1.1150307655334473, 1.1117751598358154, 1.0641820430755615, 1.0861045122146606, 1.1217414140701294, 1.1736891269683838, 1.0881339311599731, 1.111235499382019], "val_binary_accuracy": [0.9555695056915283, 0.9594541788101196, 0.9599693417549133, 0.9622806906700134, 0.9624477624893188, 0.9605123996734619, 0.9613338708877563, 0.9635616540908813, 0.9630882740020752, 0.9632553458213806, 0.9630464911460876, 0.963840126991272, 0.9638819098472595, 0.963687002658844, 0.9635756015777588, 0.9646338224411011, 0.9646338224411011], "val_ROC_AUC": [0.9607527852058411, 0.9613373279571533, 0.9686673283576965, 0.9714778065681458, 0.9708147048950195, 0.9694383144378662, 0.9686042070388794, 0.972355842590332, 0.9735656380653381, 0.9730936884880066, 0.9701546430587769, 0.9719104766845703, 0.9702484607696533, 0.9693979620933533, 0.9722999930381775, 0.9711511135101318, 0.9721179604530334], "val_PR_AUC": [0.7497867941856384, 0.7998697757720947, 0.8228471279144287, 0.8277607560157776, 0.8360421657562256, 0.8357876539230347, 0.8285924792289734, 0.8429878950119019, 0.8490172624588013, 0.8468226790428162, 0.8406122326850891, 0.853649377822876, 0.8513867259025574, 0.8475092649459839, 0.8441946506500244, 0.8563084602355957, 0.854196310043335], "lr": [0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0010000000474974513, 0.0005000000237487257, 0.0005000000237487257]}