Browse Source

Split data generation script

Jérôme BUISINE 1 year ago
parent
commit
9d3f09fdf0

File diff suppressed because it is too large
+ 179 - 37
analysis/svd_reconstruction_analysis.ipynb


File diff suppressed because it is too large
+ 125 - 14
analysis/svd_zones_analysis.ipynb


+ 0 - 1
deep_network_keras_svd.py

@@ -4,7 +4,6 @@ from keras.layers import Conv1D, MaxPooling1D
 from keras.layers import Activation, Dropout, Flatten, Dense, BatchNormalization
 from keras.wrappers.scikit_learn import KerasClassifier
 from keras import backend as K
-import matplotlib.pyplot as plt
 
 from sklearn.utils import shuffle
 from sklearn.metrics import roc_auc_score

+ 74 - 0
generateAndTrain_maxwell_custom_split.sh

@@ -0,0 +1,74 @@
+#! bin/bash
+
+if [ -z "$1" ]
+  then
+    echo "No argument supplied"
+    echo "Need of vector size"
+    exit 1
+fi
+
+if [ -z "$2" ]
+  then
+    echo "No argument supplied"
+    echo "Need of metric information"
+    exit 1
+fi
+
+result_filename="models_info/models_comparisons.csv"
+VECTOR_SIZE=200
+size=$1
+metric=$2
+
+# selection of four scenes (only maxwell)
+scenes="A, D, G, H"
+
+half=$(($size/2))
+start=-$half
+for counter in {0..4}; do
+    end=$(($start+$size))
+
+    if [ "$end" -gt "$VECTOR_SIZE" ]; then
+        start=$(($VECTOR_SIZE-$size))
+        end=$(($VECTOR_SIZE))
+    fi
+
+    if [ "$start" -lt "0" ]; then
+        start=$((0))
+        end=$(($size))
+    fi
+
+    for nb_zones in {4,6,8,10,12}; do
+
+        echo $start $end
+
+        for mode in {"svd","svdn","svdne"}; do
+            for model in {"svm_model","ensemble_model","ensemble_model_v2"}; do
+
+                FILENAME="data/${model}_N${size}_B${start}_E${end}_nb_zones_${nb_zones}_${metric}_${mode}"
+                MODEL_NAME="${model}_N${size}_B${start}_E${end}_nb_zones_${nb_zones}_${metric}_${mode}"
+                CUSTOM_MIN_MAX_FILENAME="N${size}_B${start}_E${end}_nb_zones_${nb_zones}_${metric}_${mode}_min_max"
+
+                echo $FILENAME
+
+                # only compute if necessary (perhaps server will fall.. Just in case)
+                if grep -q "${MODEL_NAME}" "${result_filename}"; then
+
+                    echo "${MODEL_NAME} results already generated..."
+                else
+                    python generate_data_model_random_split.py --output ${FILENAME} --interval "${start},${end}" --kind ${mode} --metric ${metric} --scenes "${scenes}" --nb_zones "${nb_zones}" --percent 1 --renderer "maxwell" --step 40 --random 1 --custom ${CUSTOM_MIN_MAX_FILENAME}
+                    python train_model.py --data ${FILENAME} --output ${MODEL_NAME} --choice ${model}
+
+                    #python predict_seuil_expe_maxwell.py --interval "${start},${end}" --model "saved_models/${MODEL_NAME}.joblib" --mode "${mode}" --metric ${metric} --limit_detection '2' --custom ${CUSTOM_MIN_MAX_FILENAME}
+                    python save_model_result_in_md_maxwell.py --interval "${start},${end}" --model "saved_models/${MODEL_NAME}.joblib" --mode "${mode}" --metric ${metric}
+                fi
+            done
+        done
+    done
+
+    if [ "$counter" -eq "0" ]; then
+        start=$(($start+50-$half))
+    else
+        start=$(($start+50))
+    fi
+
+done

+ 313 - 0
generate_data_model_random_split.py

@@ -0,0 +1,313 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Fri Sep 14 21:02:42 2018
+
+@author: jbuisine
+"""
+
+from __future__ import print_function
+import sys, os, argparse
+import numpy as np
+import random
+import time
+import json
+
+from PIL import Image
+from ipfml import processing, metrics, utils
+
+from modules.utils import config as cfg
+from modules.utils import data as dt
+
+# getting configuration information
+config_filename         = cfg.config_filename
+learned_folder          = cfg.learned_zones_folder
+min_max_filename        = cfg.min_max_filename_extension
+
+# define all scenes values
+all_scenes_list         = cfg.scenes_names
+all_scenes_indices      = cfg.scenes_indices
+
+normalization_choices   = cfg.normalization_choices
+path                    = cfg.dataset_path
+zones                   = cfg.zones_indices
+seuil_expe_filename     = cfg.seuil_expe_filename
+
+renderer_choices        = cfg.renderer_choices
+metric_choices          = cfg.metric_choices_labels
+output_data_folder      = cfg.output_data_folder
+custom_min_max_folder   = cfg.min_max_custom_folder
+min_max_ext             = cfg.min_max_filename_extension
+
+generic_output_file_svd = '_random.csv'
+
+min_value_interval      = sys.maxsize
+max_value_interval      = 0
+abs_gap_data            = 100
+
+
+def construct_new_line(seuil_learned, interval, line, choice, each, norm):
+    begin, end = interval
+
+    line_data = line.split(';')
+    seuil = line_data[0]
+    metrics = line_data[begin+1:end+1]
+
+    # keep only if modulo result is 0 (keep only each wanted values)
+    metrics = [float(m) for id, m in enumerate(metrics) if id % each == 0]
+
+    # TODO : check if it's always necessary to do that (loss of information for svd)
+    if norm:
+
+        if choice == 'svdne':
+            metrics = utils.normalize_arr_with_range(metrics, min_value_interval, max_value_interval)
+        if choice == 'svdn':
+            metrics = utils.normalize_arr(metrics)
+
+    if seuil_learned > int(seuil):
+        line = '1'
+    else:
+        line = '0'
+
+    for idx, val in enumerate(metrics):
+        line += ';'
+        line += str(val)
+    line += '\n'
+
+    return line
+
+def get_min_max_value_interval(_scenes_list, _interval, _metric):
+
+    global min_value_interval, max_value_interval
+
+    scenes = os.listdir(path)
+
+    # remove min max file from scenes folder
+    scenes = [s for s in scenes if min_max_filename not in s]
+
+    for id_scene, folder_scene in enumerate(scenes):
+
+        # only take care of maxwell scenes
+        if folder_scene in _scenes_list:
+
+            scene_path = os.path.join(path, folder_scene)
+
+            zones_folder = []
+            # create zones list
+            for index in zones:
+                index_str = str(index)
+                if len(index_str) < 2:
+                    index_str = "0" + index_str
+                zones_folder.append("zone"+index_str)
+
+            for id_zone, zone_folder in enumerate(zones_folder):
+
+                zone_path = os.path.join(scene_path, zone_folder)
+
+                # if custom normalization choices then we use svd values not already normalized
+                data_filename = _metric + "_svd"+ generic_output_file_svd
+
+                data_file_path = os.path.join(zone_path, data_filename)
+
+                # getting number of line and read randomly lines
+                f = open(data_file_path)
+                lines = f.readlines()
+
+                # check if user select current scene and zone to be part of training data set
+                for line in lines:
+
+                    begin, end = _interval
+
+                    line_data = line.split(';')
+
+                    metrics = line_data[begin+1:end+1]
+                    metrics = [float(m) for m in metrics]
+
+                    min_value = min(metrics)
+                    max_value = max(metrics)
+
+                    if min_value < min_value_interval:
+                        min_value_interval = min_value
+
+                    if max_value > max_value_interval:
+                        max_value_interval = max_value
+
+
+def generate_data_model(_scenes_list, _filename, _interval, _choice, _metric, _scenes, _nb_zones = 4, _percent = 1, _random=0, _step=1, _each=1, _custom = False):
+
+    output_train_filename = _filename + ".train"
+    output_test_filename = _filename + ".test"
+
+    if not '/' in output_train_filename:
+        raise Exception("Please select filename with directory path to save data. Example : data/dataset")
+
+    # create path if not exists
+    if not os.path.exists(output_data_folder):
+        os.makedirs(output_data_folder)
+
+    train_file_data = []
+    test_file_data  = []
+
+    for id_scene, folder_scene in enumerate(_scenes_list):
+
+        scene_path = os.path.join(path, folder_scene)
+
+        zones_indices = zones
+
+        # shuffle list of zones (=> randomly choose zones)
+        # only in random mode
+        if _random:
+            random.shuffle(zones_indices)
+
+        # store zones learned
+        learned_zones_indices = zones_indices[:_nb_zones]
+
+        # write into file
+        folder_learned_path = os.path.join(learned_folder, _filename.split('/')[1])
+
+        if not os.path.exists(folder_learned_path):
+            os.makedirs(folder_learned_path)
+
+        file_learned_path = os.path.join(folder_learned_path, folder_scene + '.csv')
+
+        with open(file_learned_path, 'w') as f:
+            for i in learned_zones_indices:
+                f.write(str(i) + ';')
+
+        for id_zone, index_folder in enumerate(zones_indices):
+
+            index_str = str(index_folder)
+            if len(index_str) < 2:
+                index_str = "0" + index_str
+            current_zone_folder = "zone" + index_str
+
+            zone_path = os.path.join(scene_path, current_zone_folder)
+
+            # if custom normalization choices then we use svd values not already normalized
+            if _custom:
+                data_filename = _metric + "_svd"+ generic_output_file_svd
+            else:
+                data_filename = _metric + "_" + _choice + generic_output_file_svd
+
+            data_file_path = os.path.join(zone_path, data_filename)
+
+            # getting number of line and read randomly lines
+            f = open(data_file_path)
+            lines = f.readlines()
+
+            num_lines = len(lines)
+
+            # randomly shuffle image
+            if _random:
+                random.shuffle(lines)
+
+            path_seuil = os.path.join(zone_path, seuil_expe_filename)
+
+            with open(path_seuil, "r") as seuil_file:
+                seuil_learned = int(seuil_file.readline().strip())
+
+            counter = 0
+            # check if user select current scene and zone to be part of training data set
+            for data in lines:
+
+                percent = counter / num_lines
+                image_index = int(data.split(';')[0])
+
+                if image_index % _step == 0:
+
+                    with open(path_seuil, "r") as seuil_file:
+                        seuil_learned = int(seuil_file.readline().strip())
+
+                    gap_threshold = abs(seuil_learned - image_index)
+
+                    if gap_threshold > abs_gap_data:
+
+                        line = construct_new_line(seuil_learned, _interval, data, _choice, _each, _custom)
+
+                        if id_zone < _nb_zones and folder_scene in _scenes and percent <= _percent:
+                            train_file_data.append(line)
+                        else:
+                            test_file_data.append(line)
+
+                counter += 1
+
+            f.close()
+
+    train_file = open(output_train_filename, 'w')
+    test_file = open(output_test_filename, 'w')
+
+    for line in train_file_data:
+        train_file.write(line)
+
+    for line in test_file_data:
+        test_file.write(line)
+
+    train_file.close()
+    test_file.close()
+
+
+def main():
+
+    # getting all params
+    parser = argparse.ArgumentParser(description="Generate data for model using correlation matrix information from data")
+
+    parser.add_argument('--output', type=str, help='output file name desired (.train and .test)')
+    parser.add_argument('--interval', type=str, help='Interval value to keep from svd', default='"0, 200"')
+    parser.add_argument('--kind', type=str, help='Kind of normalization level wished', choices=normalization_choices)
+    parser.add_argument('--metric', type=str, help='Metric data choice', choices=metric_choices)
+    parser.add_argument('--scenes', type=str, help='List of scenes to use for training data')
+    parser.add_argument('--nb_zones', type=int, help='Number of zones to use for training data set')
+    parser.add_argument('--random', type=int, help='Data will be randomly filled or not', choices=[0, 1])
+    parser.add_argument('--percent', type=float, help='Percent of data use for train and test dataset (by default 1)')
+    parser.add_argument('--step', type=int, help='Photo step to keep for build datasets', default=1)
+    parser.add_argument('--each', type=int, help='Each features to keep from interval', default=1)
+    parser.add_argument('--renderer', type=str, help='Renderer choice in order to limit scenes used', choices=renderer_choices, default='all')
+    parser.add_argument('--custom', type=str, help='Name of custom min max file if use of renormalization of data', default=False)
+
+    args = parser.parse_args()
+
+    p_filename = args.output
+    p_interval = list(map(int, args.interval.split(',')))
+    p_kind     = args.kind
+    p_metric   = args.metric
+    p_scenes   = args.scenes.split(',')
+    p_nb_zones = args.nb_zones
+    p_random   = args.random
+    p_percent  = args.percent
+    p_step     = args.step
+    p_each     = args.each
+    p_renderer = args.renderer
+    p_custom   = args.custom
+
+
+    # list all possibles choices of renderer
+    scenes_list = dt.get_renderer_scenes_names(p_renderer)
+    scenes_indices = dt.get_renderer_scenes_indices(p_renderer)
+
+    # getting scenes from indexes user selection
+    scenes_selected = []
+
+    for scene_id in p_scenes:
+        index = scenes_indices.index(scene_id.strip())
+        scenes_selected.append(scenes_list[index])
+
+    # find min max value if necessary to renormalize data
+    if p_custom:
+        get_min_max_value_interval(scenes_list, p_interval, p_metric)
+
+        # write new file to save
+        if not os.path.exists(custom_min_max_folder):
+            os.makedirs(custom_min_max_folder)
+
+        min_max_folder_path = os.path.join(os.path.dirname(__file__), custom_min_max_folder)
+        min_max_filename_path = os.path.join(min_max_folder_path, p_custom)
+
+        with open(min_max_filename_path, 'w') as f:
+            f.write(str(min_value_interval) + '\n')
+            f.write(str(max_value_interval) + '\n')
+
+    # create database using img folder (generate first time only)
+    generate_data_model(scenes_list, p_filename, p_interval, p_kind, p_metric, scenes_selected, p_nb_zones, p_percent, p_random, p_step, p_each, p_custom)
+
+if __name__== "__main__":
+    main()

+ 24 - 0
runAll_maxwell_custom_split.sh

@@ -0,0 +1,24 @@
+#! bin/bash
+
+# erase "models_info/models_comparisons.csv" file and write new header
+file_path='models_info/models_comparisons.csv'
+
+erased=$1
+
+if [ "${erased}" == "Y" ]; then
+    echo "Previous data file erased..."
+    rm ${file_path}
+    mkdir -p models_info
+    touch ${file_path}
+
+    # add of header
+    echo 'model_name; vector_size; start; end; nb_zones; metric; mode; tran_size; val_size; test_size; train_pct_size; val_pct_size; test_pct_size; train_acc; val_acc; test_acc; all_acc; F1_train; recall_train; roc_auc_train; F1_val; recall_val; roc_auc_val; F1_test; recall_test; roc_auc_test; F1_all; recall_all; roc_auc_all;' >> ${file_path}
+
+fi
+
+for size in {"4","8","16","26","32","40"}; do
+
+    for metric in {"lab","mscn","low_bits_2","low_bits_3","low_bits_4","low_bits_5","low_bits_6","low_bits_4_shifted_2"}; do
+        bash generateAndTrain_maxwell_custom_split.sh ${size} ${metric}
+    done
+done