source_code/post_test_episodic.py

from pathlib import Path
import random
from statistics import mean

import numpy as np
import torch
from torch import nn
from tqdm import tqdm
from torch.utils.data import DataLoader

import argparse
import os
import time
import torch.optim as optim
from glob import glob

import re
import utilities
import matplotlib.pyplot as plt
from itertools import chain, combinations, permutations
import argparse


def to_softmax(inarr):
    return np.exp(inarr)/np.sum(np.exp(inarr), axis=1)[:, None]
    
def powerset(iterable):
    "powerset([1,2,3]) --> () (1,) (2,) (3,) (1,2) (1,3) (2,3) (1,2,3)"
    s = list(iterable)
    return chain.from_iterable(combinations(s, r) for r in range(len(s)+1))

torch.autograd.set_detect_anomaly(True)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(device)

parser = argparse.ArgumentParser()
parser.add_argument('--run_num', help='Number for Identifying created folders', required=True)
parser.add_argument('--addlayer', help='layer(s) to add after resnet', required=False, default="dense")
parser.add_argument('--multi', help='If 0: Test on multiclass training. If 1: Test on binary training', required=False, default=0)

args = parser.parse_args()
training_num = int(args.run_num)
additional_layer = args.addlayer
use_multiclass_training = int(args.multi)

# Hyper Params
base_path_data = "../../Spectrum_Highway1/"
base_path = "./"
test_file = base_path_data + "GNSS_dataset/test_adap.txt"
norm_data = [-135.99690198, 0.]
batchSize = 40
time_size = 243
# ----------- #

# Data folders
# The test results will be saved in the already existing path base_path + "results_episodic + x" + "test"
train_output_folder = base_path + "results_episodic"
if additional_layer != "None":
    train_output_folder += "_" + additional_layer
if not use_multiclass_training:
    train_output_folder += "_binary"
folder = glob(train_output_folder + "/%06d_202*" % training_num)[0]
fewshot_method = folder.split("_")[-3]
new_classes = [3, 6, 8, 9]
old_classes = list(set(np.arange(11)) - set(new_classes))
print("New classes: ", new_classes)
class_permutations = list(permutations(new_classes))
# Test on all class permutations
class_permutations = np.concatenate([np.zeros((len(class_permutations), 1)) - 1, class_permutations], axis=1).astype(int)
# ---------- #

# Support Data --------------- #
with open(base_path_data + "GNSS_dataset/support_set.txt", "r") as f:
    support_info = f.readlines()
# support_list = {int(si.split("\n")[0].split("\t")[-1]): [base_path_data + ("GNSS_dataset/data/%02d/" % (int(si.split(".npy")[0].split("/")[1]) // 15000)) + si.split("\t")[0].split("/")[1]] for si in support_info}
support_list = {int(si.split("\n")[0].split("\t")[-1]): [base_path_data + "GNSS_dataset/data/" + si.split("\t")[0].split("/")[1]] for si in support_info}
# ---------------------------- #

wm = re.compile(r'.*/weights/(?P<num>\d{1,6}).h5')
wm_fs = re.compile(r'.*/weights/(?P<num>\d{1,6})_fs.h5')
if fewshot_method == "prototypical" or fewshot_method == "simpleshot":
    weights = glob(folder + "/weights/*")
    weights.sort(key=lambda x: int(wm.match(x).group(1)))
elif fewshot_method == "relation" or fewshot_method == "bdcspn" or fewshot_method == "FEAT" or fewshot_method == "laplacianshot" or fewshot_method == "matchingnetworks":
    weights = glob(folder + "/weights/*_fs.h5")
    weights.sort(key=lambda x: int(wm_fs.match(x).group(1)))
else:
    print("Invalid fewshot method: ", fewshot_method)
    exit()
if len(weights) == 0:
    print("Skipped %d: Training not finished." % training_num)
    exit()
if not os.path.exists(folder + "/post/test/"):
    os.makedirs(folder + "/post/test/")
else:
    print("Skipped %d: Already Tested." % training_num)
    exit()
# For getting best weights --- #
log_train_file = glob(folder + "/log.txt")[0]
with open(log_train_file, "r") as f:
    log_train_info = f.readlines()
log_accs = np.array([float(x.split("\n")[0].split(" ")[-1]) for x in log_train_info])
weights_ind_best = np.argmax(log_accs)
# ---------------------------- #
startweight = weights[weights_ind_best]
print(startweight)

# Create Test data lists -------------- #
with open(test_file, "r") as f:
    test_info = f.readlines()
test_info = np.array([x.split("\n")[0].split("\t") for x in test_info])
test_list_raw = test_info[:, 0]
test_classes = test_info[:, 1].astype(int)
test_list = []
for tlr in test_list_raw:
    # add_folder = ("data/%02d" % (int(tlr.split("/")[-1].split(".npy")[0]) // 15000)) + "/" + tlr.split("/")[-1]
    add_folder = ("data/" + tlr.split("/")[-1]
    test_list.append(base_path_data + "GNSS_dataset/" + add_folder)
# ------------------------------------- #

# Create Model and load weights --------------------------------------------------------------- #
class extract_tensor(nn.Module):
    # For extracting the tensor from an lstm output
    def forward(self,x):
        # Output shape (batch, features, hidden)
        tensor, _ = x
        return tensor

class TransposeB(torch.nn.Module):
    def forward(self, x):
        x = torch.transpose(x, 1, 0)
        return x

class TransposeT(torch.nn.Module):
    def forward(self, x):
        x = torch.transpose(x, 2, 1)
        return x

class multihead_Attention(nn.Module):
    def __init__(self, embeded_dim, num_heads):
        super().__init__()
        self.attention = nn.MultiheadAttention(embeded_dim, num_heads, batch_first=True)

    def forward(self,x):
        x, _ = self.attention(x, x, x)
        return x

# Create Model
from torchvision.models import resnet18
model = resnet18()
if additional_layer == "dense":
    model.fc = nn.Linear(512, 512)
elif additional_layer == "lstm":
    model.avgpool = nn.AdaptiveAvgPool2d((8, 1))
    model.use_flatten = False
    model.fc = nn.Sequential(nn.Flatten(start_dim=-2),
                             nn.LSTM(8, 8, batch_first=True),
                             extract_tensor(),
                             nn.AdaptiveAvgPool1d(1),
                             nn.Flatten())
elif additional_layer == "multiheadattention":
    model.avgpool = nn.AdaptiveAvgPool2d((8, 1))
    model.use_flatten = False
    model.fc = nn.Sequential(nn.Flatten(start_dim=-2),
                             multihead_Attention(8, 8),
                             nn.AdaptiveAvgPool2d(1),
                             nn.Flatten())
elif additional_layer == "conv":
    model.avgpool = nn.AdaptiveAvgPool2d((16, 1))  # Shape: (BS, 512, 16, 1)
    model.use_flatten = False
    model.fc = nn.Sequential(nn.Flatten(start_dim=-2),
                             nn.Conv1d(512, 512, 4),
                             nn.AdaptiveAvgPool1d(1),
                             nn.Flatten())
elif additional_layer == "TSiT":
    from tsai.models.TSiTPlus import TSiTPlus
    model.avgpool = nn.AdaptiveAvgPool2d((16, 1))  # -> (BS, 512, 16, 1)
    model.use_flatten = False
    model.fc = nn.Sequential(nn.Flatten(start_dim=-2),  # -> (BS, 512, 16)
                             TransposeT(),  # -> (BS, 16, 512)
                             TSiTPlus(16, 16, seq_len=512, d_model=32, depth=3))  # -> (BS, 16)
elif additional_layer == "gMLP":
    from tsai.models.gMLP import gMLP
    model.avgpool = nn.AdaptiveAvgPool2d((16, 1))  # -> (BS, 512, 16, 1)
    model.use_flatten = False
    model.fc = nn.Sequential(nn.Flatten(start_dim=-2),  # -> (BS, 512, 16)
                             TransposeT(),  # -> (BS, 16, 512)
                             gMLP(16, 16, seq_len=512, d_model=32, depth=3))  # -> (BS, 16)
elif additional_layer == "autoencoder":
    model.fc = nn.Sequential(nn.Linear(512, 128),
                             nn.ReLU(),
                             nn.Linear(128, 32),
                             nn.ReLU(),
                             nn.Linear(32, 8),
                             nn.Linear(8, 32),
                             nn.ReLU(),
                             nn.Linear(32, 128),
                             nn.ReLU(),
                             nn.Linear(128, 512))
else:
    model.use_fc = False
model.load_state_dict(torch.load(startweight, map_location=device))
model.eval()
from easyfsl.methods import PrototypicalNetworks
few_shot_classifier = PrototypicalNetworks(model).to(device)
# --------------------------------------------------------------------------------------------- #

addWeightspath = folder + "/post/weights/"
save_file_base = folder + "/post/test/summary_"
save_file_all_vals = folder + "/post/test/preds_"
save_file_all_vals_ind = folder + "/post/test/inds_"
for cpi, class_perm in enumerate(class_permutations):  # Test on every permutation of classes
    print("Dealing with classes permutation: ", cpi, class_perm)
    prev_classes = []
    results = []
    save_file = save_file_base + "_".join([str(pc) for pc in class_perm[1:]]) + ".txt"
    for cpj, cclass in enumerate(class_perm):  # Add the new classes on by one and test
        if cclass == -1:
            prev_weights = None
        else:
            prev_classes.append(str(cclass))
            prev_weights = addWeightspath + "_".join(prev_classes) + ".h5"
            few_shot_classifier.backbone.fc.load_state_dict(torch.load(prev_weights, map_location=device))
        print("Dealing with current class: ", cclass, " from perm: ", class_perm, " with prev classes: ", prev_classes)
        cur_old_classes = np.array(old_classes + [int(c) for c in prev_classes])
        cur_old_classes.sort()
        cur_new_classes = np.array([pc for pc in range(11) if (pc not in cur_old_classes)])
        additional_classes = [3 + np.array(list(x)).astype(int) for x in list(powerset("".join((cur_new_classes - 3).astype("str"))))]
        results.append("Additionally trained on: " + " ".join([str(pc) for pc in prev_classes]) + "\n")
        for ai, add_class in enumerate(additional_classes):
            if os.path.exists(save_file_all_vals + "_prev_classes_" + "_".join([str(pc) for pc in prev_classes]) + "_add_classes_" + "_".join([str(pc) for pc in add_class]) + ".npy"):
                print("Skipped: ", ai, add_class, " (Already evaluated previously)")
                continue
            else:
                print("Dealing with additional classes: ", ai, add_class)
            cur_classes = np.concatenate([cur_old_classes, add_class])
            cur_classes.sort()
            cur_support_list_files = []
            cur_support_list_classes = []
            for ci, cc in enumerate(cur_classes):
                cur_support_list_files.extend(support_list[cc])
                cur_support_list_classes.extend([ci] * len(support_list[cc]))
            cur_support_list = [cur_support_list_files, cur_support_list_classes]
            # Prepare support images #
            support_images, support_labels = utilities.get_support_data(cur_support_list, time_size, norm=norm_data, norm_type=1)
            support_images = torch.tensor(support_images).to(device)
            support_labels = torch.tensor(support_labels).to(device)
            few_shot_classifier.process_support_set(support_images, support_labels)
            # ---------------------- #
        
            # -------------------------- #
            datagen = utilities.gen_data_batch_test(test_list, test_classes, time_size, batchSize, norm=norm_data, norm_type=1)
            pred = []
            lab = []
            inds = []
            bcount = 0
            for X_batch, Y_batch, ind_batch in datagen:
                if bcount % 100 == 0:
                    print("Dealing with batch: %d" % bcount)
                bcount += 1
                inputs = torch.tensor(X_batch).to(device)
                out = few_shot_classifier(inputs).float()
                pred_x = out.detach().cpu().numpy()
                pred.extend(pred_x)
                lab.extend(Y_batch)
                inds.extend(ind_batch)
                del inputs, out
            pred = np.array(pred)
            lab = np.array(lab)
            inds = np.array(inds)
            np.save(save_file_all_vals + "_prev_classes_" + "_".join([str(pc) for pc in prev_classes]) + "_add_classes_" + "_".join([str(pc) for pc in add_class]) + ".npy", pred)
            np.save(save_file_all_vals_ind + "_prev_classes_" + "_".join([str(pc) for pc in prev_classes]) + "_add_classes_" + "_".join([str(pc) for pc in add_class]) + ".npy", inds)
            pred = to_softmax(pred)

            # reduce same inds -- #
            # Each sample is predicted on 4 times. The 4 predictions correspond to horizontal and vertical flipping of the sample
            pred_red_arg = []
            pred_red = []
            lab_red = []
            cpred_arg = np.zeros(len(cur_classes))
            cpred = np.zeros(len(cur_classes))
            clab = []
            cind = inds[0]
            for icind, ccind in enumerate(inds):
                if ccind == cind:
                    cpred += pred[icind]
                    cpred_arg[np.argmax(pred[icind])] += 1
                    clab.append(lab[icind])
                else:
                    cind = ccind
                    pred_red.append(cpred)
                    pred_red_arg.append(cpred_arg)
                    lab_red.append(np.sum(clab) / len(clab))
                    cpred = np.zeros(len(cur_classes))
                    cpred_arg = np.zeros(len(cur_classes))
                    cpred += pred[icind]
                    cpred_arg[np.argmax(pred[icind])] += 1
                    clab = [lab[icind]]
            pred_red.append(cpred)
            pred_red_arg.append(cpred_arg)
            lab_red.append(np.sum(clab) / len(clab))
            pred = np.array(pred_red)
            pred_arg = np.array(pred_red_arg)
            lab = np.array(lab_red)
            # ------------------- #
            
            # Calcualte the results for the binary case of jammer/no jammer #
            lab_bool = lab > 2
            pred_bool = np.argmax(pred, axis=1) > 2
            pred_arg_bool = np.sum(pred_arg[:, 3:], axis=1) > 0
            acc = np.mean(pred_bool == lab_bool)
            true_positives = np.sum(pred_bool * lab_bool)
            true_negatives = np.sum((~pred_bool) * (~lab_bool))
            false_positives = np.sum(pred_bool * (~lab_bool))
            false_negatives = np.sum((~pred_bool) * lab_bool)
            precision = true_positives/(true_positives + false_positives)
            recall = true_positives/(true_positives + false_negatives)
            f1_score = (2 * precision * recall) / (precision + recall)
            f2_score = ((1 + 2**2) * precision * recall) / (2**2 * precision + recall)
            print('Test Acc Prob: %.5f, F1: %.5f, F2: %.5f, Prec: %.5f, Rec: %.5f' % (acc, f1_score, f2_score, precision, recall))
            results.append([acc, f1_score, f2_score, precision, recall, 1])
            
            acc = np.mean(pred_arg_bool == lab_bool)
            true_positives = np.sum(pred_arg_bool * lab_bool)
            true_negatives = np.sum((~pred_arg_bool) * (~lab_bool))
            false_positives = np.sum(pred_arg_bool * (~lab_bool))
            false_negatives = np.sum((~pred_arg_bool) * lab_bool)
            precision = true_positives/(true_positives + false_positives)
            recall = true_positives/(true_positives + false_negatives)
            f1_score = (2 * precision * recall) / (precision + recall)
            f2_score = ((1 + 2**2) * precision * recall) / (2**2 * precision + recall)
            print('Test Acc Arg: %.5f, F1: %.5f, F2: %.5f, Prec: %.5f, Rec: %.5f' % (acc, f1_score, f2_score, precision, recall))
            results.append([acc, f1_score, f2_score, precision, recall, 0])
    with open(save_file, 'a') as f:
        f.write("Train\tAccuracy\tF1\tF2\tPrecision\tRecall\tUse Prob\n")
        for ir, res in enumerate(results):
            if len(res) > 6:
                f.write(res)
            else:
                f.write("%d\t%.5f\t\t%.5f\t%.5f\t%.5f\t\t%.5f\t\t%d\n" % (ir, res[0], res[1], res[2], res[3], res[4], res[5]))
print("Done")