README.md

metadata

language:
  - en
tags:
  - generated_from_trainer
datasets:
  - imagefolder
metrics:
  - accuracy
  - f1
  - recall
  - precision
pipeline_tag: image-classification
base_model: microsoft/dit-base
model-index:
  - name: dit-base-Business_Documents_Classified_v2
    results:
      - task:
          type: image-classification
          name: Image Classification
        dataset:
          name: imagefolder
          type: imagefolder
          config: data
          split: train
          args: data
        metrics:
          - type: accuracy
            value: 0.826
            name: Accuracy

dit-base-Business_Documents_Classified_v2

This model is a fine-tuned version of microsoft/dit-base on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.6715
• Accuracy: 0.826
• Weighted f1: 0.8272
• Micro f1: 0.826
• Macro f1: 0.8242
• Weighted recall: 0.826
• Micro recall: 0.826
• Macro recall: 0.8237
• Weighted precision: 0.8327
• Micro precision: 0.826
• Macro precision: 0.8293

Model description

This is a classification model of 16 different types of documents.

For more information on how it was created, check out the following link: https://github.com/DunnBC22/Vision_Audio_and_Multimodal_Projects/blob/main/Document%20AI/Multiclass%20Classification/Real%20World%20Documents%20Collections/Real%20World%20Documents%20Collections_v2.ipynb

Intended uses & limitations

This model is intended to demonstrate my ability to solve a complex problem using technology.

Training and evaluation data

Dataset Source: https://www.kaggle.com/datasets/shaz13/real-world-documents-collections

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 18

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Weighted f1	Micro f1	Macro f1	Weighted recall	Micro recall	Macro recall	Weighted precision	Micro precision	Macro precision
2.7266	0.99	31	2.4738	0.208	0.1811	0.208	0.1827	0.208	0.208	0.2101	0.2143	0.208	0.2246
2.171	1.98	62	1.8510	0.423	0.3936	0.4230	0.3925	0.423	0.423	0.4243	0.4503	0.423	0.4446
1.6525	2.98	93	1.2633	0.61	0.5884	0.61	0.5855	0.61	0.61	0.6124	0.6377	0.61	0.6283
1.346	4.0	125	1.0259	0.706	0.7023	0.706	0.6992	0.706	0.706	0.7058	0.7095	0.706	0.7034
1.253	4.99	156	0.9180	0.729	0.7277	0.729	0.7239	0.729	0.729	0.7291	0.7340	0.729	0.7261
1.0975	5.98	187	0.8859	0.747	0.7480	0.747	0.7437	0.747	0.747	0.7472	0.7609	0.747	0.7526
1.1122	6.98	218	0.8270	0.76	0.7606	0.76	0.7578	0.76	0.76	0.7594	0.7772	0.76	0.7727
1.0365	8.0	250	0.7806	0.775	0.7759	0.775	0.7730	0.775	0.775	0.7735	0.7957	0.775	0.7920
1.004	8.99	281	0.7472	0.796	0.7977	0.796	0.7957	0.796	0.796	0.7956	0.8193	0.796	0.8151
0.9278	9.98	312	0.7296	0.795	0.7974	0.795	0.7957	0.795	0.795	0.7953	0.8157	0.795	0.8115
0.8767	10.98	343	0.7257	0.809	0.8101	0.809	0.8078	0.809	0.809	0.8091	0.8182	0.809	0.8136
0.8656	12.0	375	0.6875	0.814	0.8137	0.8140	0.8106	0.814	0.814	0.8122	0.8207	0.814	0.8164
0.7905	12.99	406	0.7060	0.808	0.8093	0.808	0.8071	0.808	0.808	0.8068	0.8182	0.808	0.8145
0.8804	13.98	437	0.6849	0.82	0.8214	0.82	0.8183	0.82	0.82	0.8183	0.8260	0.82	0.8215
0.8265	14.98	468	0.6821	0.816	0.8171	0.816	0.8143	0.816	0.816	0.8142	0.8242	0.816	0.8206
0.7929	16.0	500	0.6877	0.818	0.8184	0.818	0.8152	0.818	0.818	0.8167	0.8240	0.818	0.8186
0.7993	16.99	531	0.6718	0.825	0.8259	0.825	0.8234	0.825	0.825	0.8227	0.8306	0.825	0.8282
0.7954	17.86	558	0.6715	0.826	0.8272	0.826	0.8242	0.826	0.826	0.8237	0.8327	0.826	0.8293

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.0+cu118
• Datasets 2.11.0
• Tokenizers 0.13.3