initial commit

.gitignore

@@ -0,0 +1 @@
+venv

Dockerfile

@@ -0,0 +1,22 @@
+# Use Python image
+FROM python:3.10-slim
+
+# Set working directory
+WORKDIR /code
+
+# Install system dependencies (optional)
+RUN apt-get update && apt-get install -y git
+
+# Install Python dependencies
+COPY requirements.txt .
+RUN pip install --upgrade pip
+RUN pip install -r requirements.txt
+
+# Copy model and app
+COPY . .
+
+# Expose the default HF Spaces port
+EXPOSE 7860
+
+# Run the FastAPI app using Uvicorn
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

app.py

@@ -0,0 +1,90 @@
+from fastapi import FastAPI, HTTPException
+from pydantic import BaseModel
+import pandas as pd
+from autogluon.multimodal import MultiModalPredictor
+
+app = FastAPI()
+
+# Load the model
+predictor = MultiModalPredictor.load("model_ml_dart")
+
+# Input schema
+class PredictionInput(BaseModel):
+    anchor_age: int
+    dbp: int
+    heart_rate: int
+    sbp: int
+    pH: float
+    PaCO2: float
+    PaO2: float
+    HCO3: float
+    SaO2: float
+    Compliance: float
+    Flow_Rate_L_min: float
+    Inspired_O2_Fraction: float
+    Minute_Volume: float
+    Peak_Insp_Pressure: float
+    Plateau_Pressure: float
+    Resistance_Exp: float
+    Resistance_Insp: float
+    Respiratory_Rate_Total: float
+    Tidal_Volume_observed: float
+    Tidal_Volume_set: float
+    Total_PEEP_Level: float
+    respiratory_diagnoses: str
+
+# Column renaming to match training data
+rename_map = {
+    "anchor_age": "anchor_age",
+    "dbp": "dbp",
+    "heart_rate": "heart_rate",
+    "sbp": "sbp",
+    "pH": "pH",
+    "PaCO2": "PaCO2",
+    "PaO2": "PaO2",
+    "HCO3": "HCO3",
+    "SaO2": "SaO2",
+    "Compliance": "Compliance",
+    "Flow_Rate_L_min": "Flow Rate (L/min)",
+    "Inspired_O2_Fraction": "Inspired O2 Fraction",
+    "Minute_Volume": "Minute Volume",
+    "Peak_Insp_Pressure": "Peak Insp. Pressure",
+    "Plateau_Pressure": "Plateau Pressure",
+    "Resistance_Exp": "Resistance Exp",
+    "Resistance_Insp": "Resistance Insp",
+    "Respiratory_Rate_Total": "Respiratory Rate (Total)",
+    "Tidal_Volume_observed": "Tidal Volume (observed)",
+    "Tidal_Volume_set": "Tidal Volume (set)",
+    "Total_PEEP_Level": "Total PEEP Level",
+    "respiratory_diagnoses": "respiratory_diagnoses"
+}
+
+# Mapping from predicted class index to readable label
+label_map = {
+    0: "APRV",
+    1: "CMV",
+    2: "NIV",
+    3: "SPECIAL",
+    4: "PSV",
+    5: "SIMV",
+    6: "SPONT"
+}
+
+@app.post("/predict")
+def predict(input_data: PredictionInput):
+    try:
+        # Rename input fields to match training data
+        input_dict = input_data.dict()
+        renamed_input = {rename_map[k]: v for k, v in input_dict.items()}
+        df = pd.DataFrame([renamed_input])
+
+        # Run prediction
+        raw_prediction = predictor.predict(df)[0]
+
+        # Convert numeric label to string label
+        ventilation_mode = label_map.get(int(raw_prediction), "Unknown")
+
+        return {"ventilation_mode": ventilation_mode}
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Prediction failed: {str(e)}")

model_ml_dart/assets.json

@@ -0,0 +1 @@
+{"learner_class": "BaseLearner", "column_types": {"dbp": "numerical", "heart_rate": "numerical", "sbp": "numerical", "pH": "numerical", "PaCO2": "numerical", "PaO2": "numerical", "HCO3": "numerical", "SaO2": "numerical", "Compliance": "numerical", "Flow Rate (L/min)": "numerical", "Inspired O2 Fraction": "numerical", "Minute Volume": "numerical", "Peak Insp. Pressure": "numerical", "Plateau Pressure": "numerical", "Resistance Exp": "numerical", "Resistance Insp": "numerical", "Respiratory Rate (Total)": "numerical", "Tidal Volume (observed)": "numerical", "Tidal Volume (set)": "numerical", "Total PEEP Level": "numerical", "respiratory_diagnoses": "text", "vent_mode_group_enc": "categorical"}, "label_column": "vent_mode_group_enc", "problem_type": "multiclass", "presets": null, "eval_metric_name": "accuracy", "validation_metric_name": "accuracy", "minmax_mode": "max", "output_shape": 7, "save_path": "/kaggle/working/AutogluonModels/ag-20250803_114221", "pretrained": true, "pretrained_path": null, "fit_called": true, "best_score": null, "total_train_time": null, "version": "1.4.0"}

model_ml_dart/config.yaml

@@ -0,0 +1,174 @@
+model:
+  names:
+  - ft_transformer
+  - fusion_mlp
+  - hf_text
+  hf_text:
+    checkpoint_name: local://hf_text
+    gradient_checkpointing: false
+    pooling_mode: cls
+    data_types:
+    - text
+    tokenizer_name: hf_auto
+    use_fast: true
+    max_text_len: 512
+    insert_sep: true
+    low_cpu_mem_usage: false
+    text_segment_num: 2
+    stochastic_chunk: false
+    text_aug_detect_length: 10
+    text_trivial_aug_maxscale: 0.1
+    text_train_augment_types: null
+  fusion_mlp:
+    aux_loss_weight: null
+    adapt_in_features: max
+    hidden_sizes:
+    - 128
+    activation: leaky_relu
+    dropout: 0.1
+    normalization: layer_norm
+    data_types: null
+  ft_transformer:
+    data_types:
+    - numerical
+    embedding_arch:
+    - linear
+    token_dim: 192
+    hidden_size: 192
+    num_blocks: 3
+    attention_num_heads: 8
+    attention_dropout: 0.2
+    residual_dropout: 0.0
+    ffn_dropout: 0.1
+    ffn_hidden_size: 192
+    ffn_activation: geglu
+    head_activation: relu
+    normalization: layer_norm
+    merge: concat
+    requires_all_dtypes: false
+    additive_attention: false
+    share_qv_weights: false
+    pooling_mode: cls
+    checkpoint_name: null
+data:
+  image:
+    missing_value_strategy: zero
+  text:
+    normalize_text: false
+  categorical:
+    minimum_cat_count: 100
+    maximum_num_cat: 20
+    convert_to_text: false
+    convert_to_text_template: latex
+  numerical:
+    convert_to_text: false
+    scaler_with_mean: true
+    scaler_with_std: true
+  document:
+    missing_value_strategy: zero
+  label:
+    numerical_preprocessing: standardscaler
+  pos_label: null
+  column_features_pooling_mode: concat
+  mixup:
+    turn_on: false
+    mixup_alpha: 0.8
+    cutmix_alpha: 1.0
+    cutmix_minmax: null
+    prob: 1.0
+    switch_prob: 0.5
+    mode: batch
+    turn_off_epoch: 5
+    label_smoothing: 0.1
+  modality_dropout: 0
+  templates:
+    turn_on: false
+    num_templates: 30
+    template_length: 2048
+    preset_templates:
+    - super_glue
+    - rte
+    custom_templates: null
+optim:
+  optim_type: adamw
+  lr: 0.0001
+  weight_decay: 0.001
+  lr_choice: layerwise_decay
+  lr_decay: 0.9
+  lr_schedule: cosine_decay
+  max_epochs: 20
+  max_steps: -1
+  warmup_steps: 0.1
+  end_lr: 0
+  lr_mult: 1
+  patience: 10
+  val_check_interval: 0.5
+  check_val_every_n_epoch: 1
+  skip_final_val: false
+  gradient_clip_val: 1
+  gradient_clip_algorithm: norm
+  track_grad_norm: -1
+  log_every_n_steps: 10
+  label_smoothing: 0
+  top_k: 3
+  top_k_average_method: greedy_soup
+  peft: null
+  lora:
+    module_filter: null
+    filter:
+    - query
+    - value
+    - ^q$
+    - ^v$
+    - ^k$
+    - ^o$
+    r: 8
+    alpha: 8
+    conv_lora_expert_num: 8
+  loss_func: auto
+  focal_loss:
+    alpha: null
+    gamma: 2.0
+    reduction: mean
+  mask2former_loss:
+    loss_cross_entropy_weight: 10.0
+    loss_mask_weight: 5.0
+    loss_dice_weight: 5.0
+  extra_trainable_params: []
+  cross_modal_align: null
+  cross_modal_align_weight: 0
+  automatic_optimization: true
+  lemda:
+    turn_on: false
+    arch_type: mlp_vae
+    z_dim: 8
+    num_layers: 6
+    kld_weight: 0.1
+    mse_weight: 0.1
+    adv_weight: 0.0001
+    consist_weight: 0.01
+    consist_threshold: 0.5
+    lr: 0.0001
+    optim_type: adamw
+    weight_decay: 1.0e-05
+env:
+  num_gpus: 2
+  num_nodes: 1
+  batch_size: 128
+  per_gpu_batch_size: 8
+  inference_batch_size_ratio: 4
+  precision: 16-mixed
+  num_workers: 2
+  num_workers_inference: 2
+  accelerator: auto
+  fast_dev_run: false
+  deterministic: false
+  auto_select_gpus: true
+  strategy: ddp_fork_find_unused_parameters_true
+  deepspeed_allgather_size: 1000000000.0
+  deepspeed_allreduce_size: 1000000000.0
+  compile:
+    turn_on: false
+    mode: default
+    dynamic: true
+    backend: inductor

model_ml_dart/data_processors.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b70e3e39d4220b666ac74a39cada87914f1d48bb366aea9f1855c3e9757e9fa7
+size 17353

model_ml_dart/df_preprocessor.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a130ad40ba45c32c89ec0e9b5b1d61b8b9755141091e39091b569a3ef034f345
+size 26152

model_ml_dart/eval_metric.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0024ce25367aac2c6595f96f9b18415e8483aa9fb4771dff28ea934a6ce1bca
+size 220

model_ml_dart/events.out.tfevents.1754221349.40c7a73c014d.261.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8bebd84c9e172b3f3648f1542ab55f5804af249af25971ee1c76af2d1e21fba
+size 10332957

model_ml_dart/hf_text/config.json

@@ -0,0 +1,29 @@
+{
+  "_name_or_path": "google/electra-base-discriminator",
+  "architectures": [
+    "ElectraForPreTraining"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "embedding_size": 768,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "electra",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "summary_activation": "gelu",
+  "summary_last_dropout": 0.1,
+  "summary_type": "first",
+  "summary_use_proj": true,
+  "transformers_version": "4.49.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

model_ml_dart/hf_text/special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

model_ml_dart/hf_text/tokenizer_config.json

@@ -0,0 +1,56 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "ElectraTokenizer",
+  "unk_token": "[UNK]"
+}

model_ml_dart/hparams.yaml

@@ -0,0 +1,24 @@
+optim_type: adamw
+lr_choice: layerwise_decay
+lr_schedule: cosine_decay
+lr: 0.0001
+lr_decay: 0.9
+end_lr: 0
+lr_mult: 1
+weight_decay: 0.001
+warmup_steps: 0.1
+validation_metric_name: accuracy
+peft: null
+mixup_off_epoch: 5
+skip_final_val: false
+track_grad_norm: -1
+cross_modal_align: null
+cross_modal_align_weight: 0
+automatic_optimization: true
+accumulate_grad_batches: 8
+gradient_clip_val: 1
+gradient_clip_algorithm: norm
+use_aug_optim: false
+aug_lr: 0.0001
+aug_weight_decay: 1.0e-05
+aug_optim_type: adamw

model_ml_dart/model.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a12e8b83bbaaf5f217e9abacc3ceec69dfef901a5da5f553c2f0ff9559e1520
+size 442724170

requirement.txt

@@ -0,0 +1,5 @@
+fastapi
+uvicorn
+pydantic
+pandas
+autogluon.multimodal