### Install Dependencies

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/README.md

Install the required dependencies for running EmotiEffLib examples by executing this command.

```bash
pip install -r requirements.txt
```

--------------------------------

### Install EmotiEffLib with All Dependencies (from source)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from the source code with all possible dependencies.

```sh
pip install .[all]
```

--------------------------------

### Install EmotiEffLib with ONNX Support (from source)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from the source code with basic ONNX support.

```sh
pip install .
```

--------------------------------

### Install EmotiEffLib with All Dependencies (from PyPI)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from PyPI with all possible dependencies.

```sh
pip install emotiefflib[all]
```

--------------------------------

### Install EmotiEffLib with ONNX and Torch Support (from source)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from the source code with ONNX and Torch support.

```sh
pip install .[torch]
```

--------------------------------

### Install EmotiEffLib with ONNX Support (from PyPI)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from PyPI with basic ONNX support.

```sh
pip install emotiefflib
```

--------------------------------

### Install EmotiEffLib with ONNX and Torch Support (from PyPI)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from PyPI with ONNX and Torch support.

```sh
pip install emotiefflib[torch]
```

--------------------------------

### Set Up Output Directory

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW8/bah.ipynb

Defines the output directory path and creates it if it does not exist. This is a standard setup for saving results.

```python
OUTDIR=os.path.join(TEST_DATA_DIR,'outputs')
print(OUTDIR)
if not os.path.exists(OUTDIR):
    os.makedirs(OUTDIR)
```

--------------------------------

### Load Test Set Examples

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW6/cexpr.ipynb

Reads a list of test set examples from a specified file. Assumes the file path is correctly defined in `DATA_DIR` and the file is comma-separated.

```python
with open(DATA_DIR+'../test_set_examples/CVPR_6th_ABAW_CE_test_set_example.txt', 'r') as f:
    test_set_sample=f.read().splitlines()
print(len(test_set_sample),test_set_sample[:5])
```

--------------------------------

### Install EmotiEffLib with ONNX and Engagement Support (from source)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from the source code with ONNX and TensorFlow support for engagement prediction.

```sh
pip install .[engagement]
```

--------------------------------

### Set Up Device and Model

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/Predict engagement and emotions on video.ipynb

Configures the processing device (GPU or CPU) and selects the appropriate EmotiEffLib model. This setup is crucial for performance and compatibility.

```python
test_dir = "data"
device = "cuda" if torch.cuda.is_available() else "cpu"
model_name = get_model_list()[0]
```

--------------------------------

### Install EmotiEffLib with Torch Support

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/Predict engagement and emotions on video.ipynb

Installs the EmotiEffLib library with all optional dependencies, including Torch support. This is a prerequisite for using advanced features.

```bash
!pip install emotiefflib[all]
```

--------------------------------

### Install EmotiEffLib with ONNX and Engagement Support (from PyPI)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from PyPI with ONNX and TensorFlow support for engagement prediction.

```sh
pip install emotiefflib[engagement]
```

--------------------------------

### Setup DataLoader for Training and Testing

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw4_lsd_train.ipynb

Initializes ImageFolder datasets and DataLoaders for training and testing, configuring batch size, shuffling, and multi-processing workers.

```python
kwargs = {'num_workers': 4, 'pin_memory': True} if use_cuda else {}

print(train_dir,test_dir)
train_dataset = datasets.ImageFolder(root=train_dir, transform=train_transforms)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, **kwargs)
test_dataset = datasets.ImageFolder(root=test_dir, transform=test_transforms)
test_loader  = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle=False, **kwargs) 

print(len(train_dataset), len(test_dataset))
```

--------------------------------

### Install EmotiEffLib with Torch and Engagement Support (from PyPI)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from PyPI with Torch and TensorFlow support for engagement prediction.

```sh
pip install emotiefflib[torch,engagement]
```

--------------------------------

### Training Epoch Output Example

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw4_lsd_train.ipynb

Example output from a single training epoch, showing loss, accuracy, and validation metrics.

```text
Output:
(4670,) (4670,)
Epoch : 1 - loss : 0.9188 - acc: 0.9201 - val_acc: 0.5901 - val_f1: 0.5624
```

--------------------------------

### Install EmotiEffLib with Torch and Engagement Support (from source)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the library from the source code with Torch and TensorFlow support for engagement prediction.

```sh
pip install .[torch,engagement]
```

--------------------------------

### Calibration Output Example

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW10/abaw10_expr_va_au.ipynb

This is an example output from the calibration process, showing frame index, a score, and a list of predicted values for action units.

```text
0.5
0 0.5270383115248468 [0.5532955361645552, 0.46052072742248934, 0.5676132939288541, 0.6081603369668356, 0.7208157113816389, 0.7371809402910641, 0.7228156221616712, 0.35819799018707527, 0.19182435360392894, 0.2456166868198307, 0.7970498745901836, 0.36136866478003604]
3 0.538943238075162 [0.5690584951148332, 0.4699850459303568, 0.5841102492657579, 0.6207622439725776, 0.7289476466235185, 0.7455725571969808, 0.735656019746199, 0.38063335679099225, 0.2031634446397188, 0.2568733370958639, 0.8056904583103055, 0.36686600221483945]
5 0.5403137831120037 [0.5714308635240509, 0.47138584786933163, 0.5877006690872554, 0.6246094303110398, 0.7305194423201364, 0.7455661909601603, 0.7392031462491271, 0.3872960877724845, 0.19776048586069464, 0.2580460005739822, 0.8066067275823552, 0.3636405052334256]
7 0.5401521255813982 [0.5720722545928993, 0.4706574858660388, 0.5904985715602249, 0.6279213927099452, 0.7309950810798089, 0.7440205225404813, 0.741381689713229, 0.3921002210759027, 0.18746238716148445, 0.25812627460773296, 0.8065187404653372, 0.3600708856036935]
10 0.5392338695761005 [0.5714379605819867, 0.47288563424038166, 0.5928172039675216, 0.6307517830112129, 0.7295052535538746, 0.7402813106907217, 0.7434034205509352, 0.3982717989002357, 0.17161893503841813, 0.25804261098378745, 0.805332137721375, 0.356458385672756]
15 0.5355940790976784 [0.5684582649197467, 0.47110357496114175, 0.5926554604678942, 0.63513764393943, 0.726919522231314, 0.7339525729175862, 0.744027964804191, 0.39885861214762786, 0.14284049866014215, 0.2620973029680465, 0.8032846828196317, 0.3477928483353884]
20 0.5305522720828864 [0.5625802065185059, 0.46693905737469066, 0.5901151929024, 0.6355678733787142, 0.7230328771803738, 0.7302339567385947, 0.7428415975802064, 0.39961190168175936, 0.11417298563590407, 0.26163128491620113, 0.8015808295359401, 0.33831950155134577]
25 0.5262195674360082 [0.5550957775134054, 0.46192277913312785, 0.5904256172492901, 0.6364768708749123, 0.7197607436981599, 0.7275777713822245, 0.7407000521600251, 0.39768238082696866, 0.09233166677488475, 0.2617590647547721, 0.8001625358494947, 0.330739549014832]
50 0.5105572321867163 [0.5250627698199093, 0.43788278489995525, 0.5776929933440517, 0.6393326301658836, 0.7069389974672171, 0.7143039905525759, 0.7317167816176913, 0.3890498057223596, 0.052972422324092265, 0.2499877600979192, 0.7966396719594053, 0.30510617826953473]
```

--------------------------------

### TensorFlow/Keras Setup and Version Check

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW6/eri.ipynb

Imports necessary TensorFlow and Keras modules, checks the TensorFlow version, and configures GPU memory growth. This is a standard setup for deep learning tasks.

```python
import tensorflow as tf
from tensorflow.keras.models import Model,Sequential, load_model,model_from_json
from tensorflow.keras.applications import mobilenet,mobilenet_v2,densenet,inception_resnet_v2,inception_v3,vgg16,resnet_v2,resnet
import efficientnet.tfkeras as enet
from tensorflow.keras.layers import Flatten, Dense, Dropout,GlobalAveragePooling2D,Activation, Conv2D, Reshape,DepthwiseConv2D,Input
from tensorflow.keras.optimizers import SGD, Adam
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard, Callback, CSVLogger, EarlyStopping
from tensorflow.keras.metrics import top_k_categorical_accuracy
import numpy as np
from sklearn.metrics import confusion_matrix
from tensorflow.keras.preprocessing.image import ImageDataGenerator

import tensorflow.keras.backend as K 

print(tf.__version__)
from tensorflow.compat.v1.keras.backend import set_session 
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
sess=tf.compat.v1.Session(config=config)
set_session(sess)
```

--------------------------------

### Dependency and Option Setup

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotieffcpplib/CMakeLists.txt

Finds required packages like OpenCV and sets up build options for tests, Torch, ONNX Runtime, and shared libraries. Includes a fatal error if no inference engine is specified.

```cmake
find_package(OpenCV REQUIRED)

option(BUILD_TESTS "Build EmotiEffCppLib with tests" OFF)
option(WITH_TORCH "Path to the direcotry with Libtorch" OFF)
option(WITH_ONNX "Path to the direcotry with ONNXRuntime" OFF)
option(BUILD_SHARED_LIBS "Build shared library instead of static" OFF)

if(NOT WITH_TORCH AND NOT WITH_ONNX)
    message(FATAL_ERROR "At least one inference engine should be specified: -DWITH_TORCH or -DWITH_ONNX")
endif()
```

--------------------------------

### Get Completion from Messages

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/emotions_llm/emotions_experiments.ipynb

Helper function to get a completion from a list of messages using the Aura provider. Ensure g4f is installed and updated.

```python
def get_completion_from_messages(messages):
    response = g4f.ChatCompletion.create(
        model=model_name,
        messages=messages,
        temperature=0,
        provider=g4f.Provider.Aura
    )
    return response
```

--------------------------------

### Get Completion from Content

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/emotions_llm/emotions_experiments.ipynb

Helper function to get a completion from a given content string using the Aura provider. Ensure g4f is installed and updated.

```python
def get_completion(content):
    response = g4f.ChatCompletion.create(
        model=model_name,
        messages=[{"role": "user", "content": content}],
        temperature=0,
        provider=g4f.Provider.Aura
    )
    
    return response
```

--------------------------------

### Clone EmotiEffCppLib Repository

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotieffcpplib/README.md

Clone the EmotiEffCppLib repository and its submodules to get started.

```sh
git clone https://github.com/sb-ai-lab/EmotiEffLib.git
cd EmotiEffLib
git submodule update --init --recursive
```

--------------------------------

### Print PyTorch Version

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw4_lsd_train.ipynb

Displays the installed PyTorch version. Useful for verifying the environment setup.

```python
print(f"Torch: {torch.__version__}")
```

--------------------------------

### Install Test Dependencies

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/emotiefflib/README.md

Install the necessary dependencies for running tests using pip.

```sh
pip install -r tests/requirements.txt
```

--------------------------------

### Set up test directory, device, and model (ONNX)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/One image emotion recognition.ipynb

Defines the test directory, selects the CPU device, and retrieves the first available model name for ONNX inference.

```python
test_dir = os.path.join("..", "..", "..", "tests")
device = "cpu"
model_name = get_model_list()[0]
```

--------------------------------

### Initialize Video Path and Recognizer Variables

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/cpp/Predict emotions on video.ipynb

Sets the path to the input video file and declares variables for the recognition backend, model name, recognizer instance, and file extension.

```cpp
auto videoPath = emotiEffLibRootDir + "/tests/data/video_samples/emotions/Angry/Angry.mp4";
std::string backend;
std::string modelName;
std::unique_ptr<EmotiEffLib::EmotiEffLibRecognizer> fer;
std::string ext;
```

--------------------------------

### Model Instantiation and Optimizer/Scheduler Setup

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW10/abaw10_expr_va_au.ipynb

Instantiates a SimpleMLP3 model and configures an Adam optimizer with a ReduceLROnPlateau learning rate scheduler. Weight decay is also applied.

```python
# instantiate with dropout=0.3 or tune
#model = SimpleMLP(input_dim=input_shape[0], hidden_dim=128, out_dim=num_labels).to(device)
#model = SimpleMLP2(input_dim=input_shape[0], hidden_dim=256, out_dim=num_labels, dropout=0.3).to(device)
model = SimpleMLP3(input_dim=input_shape[0], out_dim=num_labels).to(device)
print(model)

# ---------- 4) Optimizer + scheduler + weight_decay ----------
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3, weight_decay=1e-5)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.5, patience=2)
```

--------------------------------

### Initialize Dataset and Retrieve Labels

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/02 - Engagement user-independent model training.ipynb

Configures the dataset for binary classification and retrieves video-to-label mappings. This step is crucial for preparing the data for model training.

```python
ENGAGE_WILD_DATASET, BIN_CLASSIFICATION, N_CLASSES, DATASET_NAME, BASE_DATASET_DIR, ext = define_dataset(False, True)
video2label, labels_list = get_video2label()
print(labels_list)
print(video2label)
```

--------------------------------

### Install EmotiEffLib with Engagement Support

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/Predict engagement and emotions on video.ipynb

Installs the EmotiEffLib library with support for engagement prediction. This command should be run in a terminal or notebook cell.

```bash
!pip install emotiefflib[engagement]
```

--------------------------------

### Initialize Video Path and Recognizer Variables

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/cpp/Predict engagement and emotions on video.ipynb

Sets the path to the video file and declares variables for the backend, model name, EmotiEffLib recognizer, and file extension.

```cpp
auto videoPath = emotiEffLibRootDir + "/tests/data/video_samples/engagement/engaged/1_video1.mp4";
std::string backend;
std::string modelName;
std::unique_ptr<EmotiEffLib::EmotiEffLibRecognizer> fer;
std::string ext;
```

--------------------------------

### Initialize PyTorch Environment and Load Model

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw3_train.ipynb

Sets up the PyTorch environment, including device selection (GPU if available) and imports necessary libraries. Loads a pre-trained feature extractor model.

```python
import glob
from itertools import chain
import os
import random
import zipfile

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from PIL import Image
from sklearn.model_selection import train_test_split
from torch.optim.lr_scheduler import StepLR
from torch.utils.data import DataLoader, Dataset
from torchvision import datasets, transforms
from tqdm.notebook import tqdm

print(f"Torch: {torch.__version__}")
device = 'cuda'
use_cuda = torch.cuda.is_available()
print(use_cuda)
```

```python
if False:
    PATH='enet_b2_8.pt'
    #PATH='enet_b2_7.pt'
    IMG_SIZE=260
else:
    #PATH='enet_b0_7.pt'
    #PATH='enet_b0_8_best_afew.pt'
    PATH='enet_b0_8_best_vgaf.pt'
    IMG_SIZE=224
    
test_transforms = transforms.Compose(
    [
        transforms.Resize((IMG_SIZE,IMG_SIZE)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                                     std=[0.229, 0.224, 0.225])
    ]
)

print(PATH)
feature_extractor_model = torch.load('../../models/affectnet_emotions/'+PATH)
```

--------------------------------

### Build Doxygen Documentation

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/README.md

Run these commands in the project's root directory to generate Doxygen documentation. The output will be in the 'html' folder.

```bash
doxygen
doxygen Doxyfile-cpp
doxygen Doxyfile-python
```

--------------------------------

### Load EngageWild Dataset and Mappings

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/02 - Engagement user-independent model training.ipynb

Initializes the EngageWild dataset configuration, including binary classification status, number of classes, dataset name, base directory, and file extension. It also retrieves video-to-label mappings and prints the available labels and the mapping dictionary.

```python
ENGAGE_WILD_DATASET, BIN_CLASSIFICATION, N_CLASSES, DATASET_NAME, BASE_DATASET_DIR, ext = define_dataset(True, False)
video2label, labels_list = get_video2label()
print(labels_list)
print(video2label)
```

--------------------------------

### Set up test directory, device, and model (PyTorch)

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/One image emotion recognition.ipynb

Defines the test directory, selects the CUDA device if available otherwise CPU, and retrieves the first available model name for PyTorch inference.

```python
test_dir = os.path.join("..", "..", "..", "tests")
device = "cuda" if torch.cuda.is_available() else "cpu"
model_name = get_model_list()[0]
```

--------------------------------

### Install Specific timm Library Version

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/README.md

Install timm version 0.4.5, which is required for EmotiEffLib v0.1 to ensure compatibility with EfficientNet models for PyTorch.

```bash
pip install timm==0.4.5
```

--------------------------------

### Install EmotiEffLib with ONNX Support

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/One image emotion recognition.ipynb

Installs the EmotiEffLib library using pip. This command ensures that the library is available for use in the current Python environment.

```bash
!pip install emotiefflib
```

--------------------------------

### Load EngageWild Dataset and Mappings

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/02 - Engagement user-independent model training.ipynb

Initializes the EngageWild dataset for binary classification, retrieves the list of labels, and prints the video-to-label mapping. This is essential for setting up the data pipeline.

```python
ENGAGE_WILD_DATASET, BIN_CLASSIFICATION, N_CLASSES, DATASET_NAME, BASE_DATASET_DIR, ext = define_dataset(True, True)
video2label, labels_list = get_video2label()
print(labels_list)
print(video2label)
```

--------------------------------

### Training Configuration Settings

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/train_faces_torch.ipynb

Sets up hyperparameters for the training process, including batch size, number of epochs, learning rate, and device selection (CPU/GPU).

```python
# Training settings
batch_size = 48#64 #48# 32 #8 #16 #
epochs = 40
lr = 3e-5
gamma = 0.7
seed = 42
device = 'cuda'
use_cuda = torch.cuda.is_available()
print(use_cuda)
```

--------------------------------

### Install Dependencies in Colab

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/Predict engagement and emotions on video.ipynb

Checks if the code is running in Google Colab and installs necessary dependencies from a requirements file. This ensures the environment is set up correctly for the tutorial.

```python
try:
    import google.colab
    import urllib.request
    IN_COLAB = True
    urllib.request.urlretrieve("https://github.com/sb-ai-lab/EmotiEffLib/blob/main/docs/tutorials/python/requirements.txt?raw=true", "requirements.txt")
    !pip install -r requirements.txt
except:
    IN_COLAB = False
```

--------------------------------

### DDAMNet Model Setup and Loading

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW8/emi.ipynb

Initializes and loads a DDAMNet model, specifying the number of classes and heads. It also loads pre-trained weights from a checkpoint file.

```python
import sys
DDAMFN_PATH='/home/avsavchenko/src/distr/DDAMFN'
sys.path.append(DDAMFN_PATH)
from networks.DDAM import DDAMNet

if True:
    feature_extractor_model = DDAMNet(num_class=8, num_head=2)
    model_name='affectnet8_epoch4_acc0.6462'
else:
    feature_extractor_model = DDAMNet(num_class=10, num_head=2)
    model_name='affectnet8_epoch9_acc0.642_mtl_2'
feature_extractor_model.load_state_dict(torch.load(DDAMNFN_PATH+'/checkpoints/'+model_name+'.pth')['model_state_dict'])
feature_extractor_model.bn=torch.nn.Identity()
print(model_name)
```

--------------------------------

### Deprecation Warning Example

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/02 - Engagement user-independent model training.ipynb

This is an example of a DeprecationWarning related to converting a NumPy array with ndim > 0 to a scalar. It indicates a potential issue in future NumPy versions.

```python
/media/deelvin_disk/echuraev/tmp_dir/ipykernel_4051215/1902217451.py:28: DeprecationWarning: Conversion of an array with ndim > 0 to a scalar is deprecated, and will error in future. Ensure you extract a single element from your array before performing this operation. (Deprecated NumPy 1.25.)
  y_pred.append(float(pred[0]))
```

--------------------------------

### Load and Inspect Test Set Sample

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW7/cexpr.ipynb

Reads a test set file, splits it into lines, and prints the count and first five entries. Assumes 'DATA_DIR' is defined.

```python
with open(DATA_DIR+'../test_set_examples/CER.txt', 'r') as f:
    test_set_sample=f.read().splitlines()
print(len(test_set_sample),test_set_sample[:5])
```

--------------------------------

### Install Required Libraries

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/03 - Custom dataset features extractor.ipynb

Installs necessary Python packages for deep learning tasks. This includes libraries like mtcnn, timm, tensorflow, numba, and specific versions of PyTorch.

```python
if False:
    !pip install mtcnn
    !pip install timm==0.4.5
    !pip install tensorflow
    !pip install numba
    #!pip install keras==2.4.3
    #!pip install -q torch==1.7.1 torchvision
    !pip install torch==2.1.0+cu118 torchvision==0.16.0+cu118 -f https://download.pytorch.org/whl/torch_stable.html
```

--------------------------------

### Install EmotiEffLib with Torch support

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/docs/tutorials/python/One image emotion recognition.ipynb

Installs the EmotiEffLib library with PyTorch support using pip. Note: The output indicates a potential issue with the package name or shell interpretation.

```bash
!pip install emotiefflib[torch]
```

--------------------------------

### Define Data and Model Parameters

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/personalized_models/04 - Engagement user-adaptation.ipynb

Sets up various configuration parameters including data directories, dataset paths, model parameters like batch size, number of classes, image set size, epochs, learning rate, and concatenation strategy.

```python
DATA_DIR = '/media/deelvin_disk/echuraev/Workspace/HSE/engagement/'
USERS_DATASET = "/media/deelvin_disk/echuraev/Workspace/HSE/datasets/Engagement_dataset/"
USERS_FEATURES = USERS_DATASET + "features_with_balanced/"
USERS_WEIGHTS = USERS_DATASET + 'weights_with_balanced/'
TABLE_NAME = "04_User-adaptation.xlsx"
BATCH_SIZE = 128
N_CLASSES = 2
IMAGE_SET_SIZE=70
EPOCH_NUM = 12
LR = 1e-4
TABLE_NAME = None
CONCATENATE_STAT = True
THRESHOLD = 0.4
REPEAT = 10
```

--------------------------------

### Emotion Classification Setup

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/video_summarizer.ipynb

Configures the mapping between numerical indices and emotion labels, and defines a list of interesting emotional states for analysis. This setup is conditional based on the number of classifier biases.

```python
if len(classifier_bias)==7:
    idx_to_class={0: 'Anger', 1: 'Disgust', 2: 'Fear', 3: 'Happiness', 4: 'Neutral', 5: 'Sadness', 6: 'Surprise'}
    INTERESTING_STATES=[0,1,2,3,6]
else:
    idx_to_class={0: 'Anger', 1: 'Contempt', 2: 'Disgust', 3: 'Fear', 4: 'Happiness', 5: 'Neutral', 6: 'Sadness', 7: 'Surprise'}
    INTERESTING_STATES=[0,2,3,4,7]
print(idx_to_class)
```

--------------------------------

### Initialize PyTorch and Device

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/video_summarizer.ipynb

This snippet initializes PyTorch, imports necessary libraries like PIL and torchvision, and sets the computation device to CUDA if available.

```python
import torch
from PIL import Image
from torchvision import datasets, transforms

print(f"Torch: {torch.__version__}")
device = 'cuda'
```

--------------------------------

### Data Processing and Output Example

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw5_expr_va_au_train.ipynb

This snippet shows an example of data processing and the resulting output format, likely representing predictions or intermediate values during training. It includes numerical values and arrays.

```python
0.1
0.5091455374196453
(0.5279939392250242, 0.6166666666666666, array([0.55987886, 0.43465787, 0.56224317, 0.62654832, 0.74539631,
       0.75222757, 0.7285629 , 0.33105139, 0.18511054, 0.20258621,
       0.8408712 , 0.36679294]), array([0.8, 0.7, 0.7, 0.6, 0.4, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.5]), array([0.89047318, 0.89392512, 0.86332642, 0.81484223, 0.77535237,
       0.81338429, 0.86023336, 0.9498089 , 0.94286913, 0.92862846,
       0.77820992, 0.8147884 ]))
0.2
0.5127973921074275
(0.533017397148989, 0.6249999999999999, array([0.56484658, 0.44359876, 0.56877713, 0.62935288, 0.74865063,
       0.754916  , 0.73146233, 0.34267134, 0.19023254, 0.20882875,
       0.84174832, 0.37112349]), array([0.7, 0.7, 0.7, 0.6, 0.5, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.6]), array([0.87937717, 0.89669968, 0.86721575, 0.81786801, 0.79496272,
       0.8154097 , 0.86255933, 0.95284365, 0.94462313, 0.93507702,
       0.77923721, 0.84530859]))
0.30000000000000004
0.5161879402744332
(0.5375026584034432, 0.6166666666666666, array([0.56976915, 0.45084511, 0.57360552, 0.63160508, 0.75165132,
       0.75679927, 0.73335565, 0.35297151, 0.19374649, 0.21774768,
       0.84249049, 0.37544463]), array([0.7, 0.7, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.6]), array([0.88203286, 0.89877668, 0.87046134, 0.80362286, 0.79684234,
       0.81687885, 0.8642281 , 0.95496999, 0.94522874, 0.94091998,
       0.78018599, 0.84601288]))
0.4
0.5187270490065781
(0.541014384144264, 0.6166666666666666, array([0.57316591, 0.45698558, 0.57729665, 0.63373932, 0.7542145 ,
       0.75784588, 0.73414719, 0.35982333, 0.19863126, 0.2245393 ,
       0.84299787, 0.37878582]), array([0.7, 0.7, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.6]), array([0.88380481, 0.89999013, 0.87300487, 0.80574471, 0.79829354,
       0.81769081, 0.8652352 , 0.95611166, 0.94510762, 0.94478463,
       0.78093739, 0.84583345]))
0.5
0.5199393283667891
(0.5430868951332039, 0.6166666666666666, array([0.57564271, 0.46087072, 0.57899923, 0.6347521 , 0.75550435,
       0.75756078, 0.73303063, 0.36535557, 0.20331393, 0.22754806,
       0.84300553, 0.38145913]), array([0.7, 0.7, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.6]), array([0.88474461, 0.90025705, 0.87444935, 0.80709947, 0.79867261,
       0.81742614, 0.86495483, 0.95664325, 0.94413641, 0.94737078,
       0.78120654, 0.84459757]))
0.6000000000000001
0.5198255232950532
(0.543975201831992, 0.6249999999999999, array([0.57638351, 0.46688322, 0.57987711, 0.6332615 , 0.7558231 ,
       0.755736  , 0.73109244, 0.36942135, 0.20410174, 0.23078234,
       0.84263162, 0.38170849]), array([0.7, 0.8, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.7, 0.8, 0.3, 0.6]), array([0.88472443, 0.91312501, 0.87501458, 0.80686172, 0.7981141 ,
       0.81579774, 0.86405764, 0.95661409, 0.94202801, 0.94938722,
       0.78110337, 0.84139011]))
0.7000000000000001
0.5178935122387572
(0.5436623812129876, 0.6333333333333332, array([0.57502695, 0.47127263, 0.58108291, 0.62987163, 0.75483367,
       0.75371778, 0.72896558, 0.36984061, 0.20351354, 0.23411769,
       0.84176021, 0.37994538]), array([0.7, 0.8, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.8, 0.8, 0.3, 0.6]), array([0.88329341, 0.91254407, 0.8750886 , 0.80534546, 0.79638926,
       0.81397195, 0.8629496 , 0.95592774, 0.95789932, 0.95086758,
       0.78052692, 0.83702976]))
0.8
0.5149169725953825
(0.5426215845931188, 0.6416666666666666, array([0.57209059, 0.47380567, 0.57948074, 0.6253632 , 0.75328333,
       0.75140504, 0.72562433, 0.36879743, 0.20502473, 0.23820447,
       0.84024417, 0.37813532]), array([0.7, 0.8, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.8, 0.8, 0.3, 0.7]), array([0.88070277, 0.91101212, 0.87379216, 0.80276604, 0.79408347,
       0.8119914 , 0.86091298, 0.9546627 , 0.95493186, 0.95158085,
       0.77910936, 0.86103186]))
0.9
0.511355118935657
(0.5398550602179175, 0.65, array([0.56802046, 0.47297454, 0.57562929, 0.62092852, 0.7507178 ,
       0.74875759, 0.72278971, 0.36296607, 0.20020751, 0.24060946,
       0.83867702, 0.37598276]), array([0.8, 0.8, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.8, 0.8, 0.3, 0.7]), array([0.89164177, 0.90838559, 0.87161871, 0.8001754 , 0.79085134,
       0.80972375, 0.85908496, 0.95232776, 0.94985824, 0.95170646,
       0.77771647, 0.85579899]))
1.0
0.5075692080786435
(0.5367145284583944, 0.65, array([0.56438654, 0.47163507, 0.57207889, 0.61677155, 0.74818918,
       0.74534595, 0.71942028, 0.35743029, 0.19360184, 0.24140214,
       0.83680711, 0.37350551]), array([0.8, 0.8, 0.7, 0.5, 0.5, 0.5, 0.6, 0.8, 0.8, 0.8, 0.3, 0.7]), array([0.8884859 , 0.90530374, 0.8692748 , 0.79757579, 0.78782557,
       0.80682583, 0.8569631 , 0.94949937, 0.94323473, 0.95116814,
       0.77594452, 0.85026333]))
```

--------------------------------

### Load and Parse Test Set Examples

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/ABAW10/abaw10_expr_va_au.ipynb

Loads sample data from a text file, parses it to extract image locations and action unit labels, and organizes it by video.

```python
with open(os.path.join(DATA_DIR,'test_set_examples/CVPR_6th_ABAW_AU_test_set_example.txt'),'r') as f:
    test_set_sample=f.read().splitlines()
print(len(test_set_sample),test_set_sample[:5])

test_set_videos={}
for s in test_set_sample[1:]:
    videoname,img_name=s.split(',')[0].split('/')
    if videoname not in test_set_videos:
        test_set_videos[videoname]=[]
    test_set_videos[videoname].append(img_name)
    
print(len(test_set_videos))
```

--------------------------------

### Print TensorFlow Version

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/ABAW/abaw3_train.ipynb

Prints the installed TensorFlow version. Useful for debugging and ensuring compatibility.

```python
print(tf.__version__)

```

--------------------------------

### Model Configuration and Loading

Source: https://github.com/sb-ai-lab/emotiefflib/blob/main/training_and_examples/train_ramas.ipynb

This snippet shows how to conditionally select and configure different base models like EfficientNet, MobileNet, or InceptionV3 based on a flag. It also demonstrates loading a pre-trained model from a file.

```python
USE_EFFICIENT_NET=False

if USE_EFFICIENT_NET:
    #import efficientnet.keras as enet

    base_model = enet.EfficientNetB3(weights=None)
    #base_model.load_weights('/home/avsavchenko/distr/efficientnet/efficientnet-b3-weights.h5')

    INPUT_SIZE = base_model.input_shape[1]
    INPUT_SIZE=(INPUT_SIZE,INPUT_SIZE)
    net_description='affectnet_enet3'
    net_model=enet
    BATCH_SIZE=16
    print(INPUT_SIZE)
elif False:
    net_model=mobilenet
    net_description='mobilenet_v1'
    INPUT_SIZE = (224, 224)
    BATCH_SIZE=80
    base_model = mobilenet.MobileNet(input_shape=(INPUT_SIZE[0],INPUT_SIZE[1],3), include_top=True, weights='imagenet')
elif False:
    net_model=inception_v3
    net_description='inception_v3'
    INPUT_SIZE = (299, 299)
    BATCH_SIZE=32
    base_model = inception_v3.InceptionV3(input_shape=(INPUT_SIZE[0],INPUT_SIZE[1],3), include_top=True, weights='imagenet')
else:
    base_model=load_model('age_gender_tf2_224_deep-03-0.13-0.97.h5')
    base_model.summary()
    net_description='my_mobilenet_8'
    if False:
        net_model=vgg16
    else:
        def mobilenet_preprocess_input(x,**kwargs):
            x[..., 0] -= 103.939
            x[..., 1] -= 116.779
            x[..., 2] -= 123.68
            return x
        preprocessing_function=mobilenet_preprocess_input
        
    BATCH_SIZE=80
    INPUT_SIZE = (224, 224)
```