Update README.md

README.md

@@ -1,126 +1,42 @@
 ---
-pipeline_tag: sentence-similarity
+pipeline_tag: text-classification
 tags:
 - sentence-transformers
-- feature-extraction
-- sentence-similarity
 - transformers
-
+- SetFit
+- News
 ---
 
-# {MODEL_NAME}
-
-This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
-
-<!--- Describe your model here -->
-
-## Usage (Sentence-Transformers)
-
-Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
-
-```
-pip install -U sentence-transformers
-```
-
-Then you can use the model like this:
-
-```python
-from sentence_transformers import SentenceTransformer
-sentences = ["This is an example sentence", "Each sentence is converted"]
-
-model = SentenceTransformer('{MODEL_NAME}')
-embeddings = model.encode(sentences)
-print(embeddings)
-```
-
-
-
-## Usage (HuggingFace Transformers)
-Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
-
-```python
-from transformers import AutoTokenizer, AutoModel
-import torch
-
-
-#Mean Pooling - Take attention mask into account for correct averaging
-def mean_pooling(model_output, attention_mask):
-    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
-    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
-    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
-
-
-# Sentences we want sentence embeddings for
-sentences = ['This is an example sentence', 'Each sentence is converted']
-
-# Load model from HuggingFace Hub
-tokenizer = AutoTokenizer.from_pretrained('{MODEL_NAME}')
-model = AutoModel.from_pretrained('{MODEL_NAME}')
-
-# Tokenize sentences
-encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
-
-# Compute token embeddings
-with torch.no_grad():
-    model_output = model(**encoded_input)
-
-# Perform pooling. In this case, mean pooling.
-sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
-
-print("Sentence embeddings:")
-print(sentence_embeddings)
-```
-
-
-
-## Evaluation Results
 
-<!--- Describe how your model was evaluated -->
+# IPTC topic classifier (multilingual)
 
-For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name={MODEL_NAME})
+A SetFit model fit on downlsampled multilingual IPTC Subject labels (concatenated for the lowest hierarchy level into artificial sentences of keywords) to predict the mid level news categories. 
+The purpose of this classifier is to support exploring corpora as weak labeler, since the representations of these descriptions are only approximations of real documents from those topics. 
+Accuracy on highest level labels in eval: 
+0.9779412
+Accuracy/F1/mcc on mid level labels in eval: 
+0.6992481/0.6666667/0.6992617
 
+More interestingly, I used the kaggle dataset with headlines from huffington post and manually selected 15 overlapping high level categories to evaluate the performance. 
+https://www.kaggle.com/datasets/rmisra/news-category-dataset
 
-## Training
-The model was trained with the parameters:
+While mcc 0.1968043 on this dataset does not sound as good as before, the mistakes usually could also be seen as a re-interpretation. I.e. news on arrests where categorized as entertainment in the huffington post dataset, the classifier put it into the crime category. 
+My current impression is this system is useful for the aimed for purpose. 
 
-**DataLoader**:
 
-`torch.utils.data.dataloader.DataLoader` of length 47800 with parameters:
-```
-{'batch_size': 1, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
-```
 
-**Loss**:
+The numeric categories can be joined with the labels by using this table: 
+https://huggingface.co/datasets/KnutJaegersberg/IPTC-topic-classifier-labels
 
-`sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss` 
 
-Parameters of the fit()-Method:
-```
-{
-    "epochs": 2,
-    "evaluation_steps": 0,
-    "evaluator": "NoneType",
-    "max_grad_norm": 1,
-    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
-    "optimizer_params": {
-        "lr": 2e-05
-    },
-    "scheduler": "WarmupLinear",
-    "steps_per_epoch": 95600,
-    "warmup_steps": 9560,
-    "weight_decay": 0.01
-}
-```
+Looks like try out api box to the right by huggingface does not yet handle setfit models, can't do anything about that. 
 
 
-## Full Model Architecture
-```
-SentenceTransformer(
-  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
-  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
-)
-```
+Use like any other SetFit model 
 
-## Citing & Authors
+from setfit import SetFitModel
 
-<!--- Describe where people can find more information -->
+# Download from Hub and run inference
+model = SetFitModel.from_pretrained("KnutJaegersberg/IPTC-classifier-ml")
+# Run inference
+preds = model(["Rachel Dolezal Faces Felony Charges For Welfare Fraud", "Elon Musk just got lucky", "The hype on AI is different from the hype on other tech topics"])