Move main.predict_file to predict.predict_file and uses trainer.predict() for predict_file(). Speeds up main.evaluate!

[bw4sz]

1. Migrate the code in main.predict_file to predict.predict_file with just a wrapper in main. main.py is too long and shouldn't have any complex logic in it.
2. Used the trainer logic instead of manually moving batches to GPU, this causes significant speed up in evaluate, which should close main.evaluate is too slow #538.

Before was around 37 seconds.

After

Tested on hpc with 1 gpu

(base) [b.weinstein@login12 ~]$ cat tunnel.sh
#!/bin/bash
#SBATCH --job-name=tunnel   # Job name
#SBATCH --mail-type=END               # Mail events
#SBATCH [email protected]  # Where to send mail
#SBATCH --account=ewhite
#SBATCH --nodes=1                 # Number of MPI ran
#SBATCH --cpus-per-task=1
#SBATCH --mem=70GB
#SBATCH --time=12:00:00       #Time limit hrs:min:sec
#SBATCH --output=/home/b.weinstein/logs/tunnel.out   # Standard output and error log
#SBATCH --error=/home/b.weinstein/logs/tunnel.err
#SBATCH --partition=gpu
#SBATCH --gpus=1

(base) [b.weinstein@login12 tests]$ cat profile_predict_file.py
#Profile the dataset class on gpu
from deepforest import main
from deepforest import get_data
import os
import pandas as pd
import numpy as np
import cProfile, pstats
import tempfile
from PIL import Image
import cv2

def run(m, csv_file, root_dir):
    predictions = m.predict_file(csv_file=csv_file, root_dir=root_dir)

if __name__ == "__main__":
    profiler = cProfile.Profile()
    profiler.enable()
    m = main.deepforest()
    m.use_release()
    m.config["workers"] = 0
    m.config["batch_size"] = 24

    csv_file = get_data("OSBS_029.csv")
    image_path = get_data("OSBS_029.png")
    tmpdir = tempfile.gettempdir()
    df = pd.read_csv(csv_file)

    big_frame = []
    for x in range(100):
        img = Image.open("{}/{}".format(os.path.dirname(csv_file), df.image_path.unique()[0]))
        cv2.imwrite("{}/{}.png".format(tmpdir, x), np.array(img))
        new_df = df.copy()
        new_df.image_path = "{}.png".format(x)
        big_frame.append(new_df)

    big_frame = pd.concat(big_frame)
    big_frame.to_csv("{}/annotations.csv".format(tmpdir))


    run(m, csv_file = "{}/annotations.csv".format(tmpdir), root_dir = tmpdir)
    profiler.disable()
    stats = pstats.Stats(profiler).sort_stats('cumtime')
    stats.print_stats()
    stats.dump_stats('predict_file.prof')

[ethanwhite]

This all looks good. My one question is whether or not we could just import predict_file() directly from predict.py instead of having what is basically just a pass through function, but I'm guessing there's something special about the lightening module class that means we need to put a copy there, so I'm going to go ahead and merge.