zaidbhat1234

from autovideo.utils import set_log_path

set_log_path('log.txt') #Setup logger
data_dir = 'datasets/hmdb6/' #Directory containing dataset (hmdb-6)

train_table_path = os.path.join(data_dir, 'train.csv')
train_media_dir = os.path.join(data_dir, 'media')
target_index = 2 #Index of column containing label information

# Read the CSV file

zaidbhat1234

from autovideo.utils import set_log_path
import pandas as pd
import os

set_log_path('log.txt') #Setup logger
data_dir = 'datasets/hmdb6/' #Directory containing dataset (hmdb-6)

train_table_path = os.path.join(data_dir, 'train.csv')
train_media_dir = os.path.join(data_dir, 'media')
target_index = 2 #Index of column containing label information

zaidbhat1234

from autovideo import extract_frames

# Extract frames from the video
video_ext = train_dataset.iloc[0, 1].split('.')[-1]
extract_frames(train_media_dir, video_ext)

zaidbhat1234

# Build pipeline based on configs
# Here we can specify the hyperparameters defined in each primitive
# The default hyperparameters will be used if not specified
from autovideo import build_pipeline

config = {
    "algorithm": 'tsn', #Specify the Action Recognition algorithm to use. In this example we use TSN
    "load_pretrained": True, #To fine-tune from pretrained weights
    "learning_rate": 0.001, 
    "epochs": 20, 

zaidbhat1234

class Hyperparams(SupervisedHyperparamsBase):
    num_workers = hyperparams.Hyperparameter[int](
        semantic_types=['https://metadata.datadrivendiscovery.org/types/ResourcesUseParameter'],
        default=2,
        description='The number of subprocesses to use for data loading. 0 means that the data will be loaded in the '
                    'main process.'
    )
    batch_size = hyperparams.Hyperparameter[int](
        default=2,
        semantic_types=['https://metadata.datadrivendiscovery.org/types/TuningParameter'],

zaidbhat1234

from autovideo import fit
import torch

# Fit
_, fitted_pipeline = fit(train_dataset=train_dataset,
                        train_media_dir=train_media_dir,
                        target_index=target_index,
                        pipeline=pipeline)

# Save the fitted pipeline

zaidbhat1234

from autovideo.utils import set_log_path, logger
set_log_path('log.txt')

# Load fitted pipeline
import torch
if torch.cuda.is_available():
    fitted_pipeline = torch.load('fitted_pipeline', map_location="cuda:0")
else:
    fitted_pipeline = torch.load('fitted_pipeline', map_location="cpu")

zaidbhat1234

import os
import pandas as pd

data_dir = 'datasets/hmdb6'
train_table_path = os.path.join(data_dir, 'train.csv')
valid_table_path = os.path.join(data_dir, 'test.csv')
train_media_dir = os.path.join(data_dir, 'media')
valid_media_dir = train_media_dir

train_dataset = pd.read_csv(train_table_path)

zaidbhat1234

from autovideo.searcher import RaySearcher

import ray
from ray import tune

#Initialise the searcher
searcher = RaySearcher(
        train_dataset=train_dataset,
        train_media_dir=train_media_dir,
        valid_dataset=valid_dataset,

zaidbhat1234

best_config = searcher.search(
        search_space=search_space,
        config=config
)

print("Best config: ", best_config)