How to use run_group method in autotest

Best Python code snippet using autotest_python

opt_finetune.py

Source:opt_finetune.py

1import argparse2import logging3import numpy as np4import random5import torch6logger = logging.getLogger("SemEval")7def set_seed(args):8    random.seed(args.seed)9    np.random.seed(args.seed)10    torch.manual_seed(args.seed)11    if args.n_gpu > 0:12        torch.cuda.manual_seed_all(args.seed)13def bool_flag(v):14    if v.lower() in {"on", "true", "yes", "t", "y", "1"}:15        return True16    elif v.lower() in {"off", "false", "no", "f", "n", "0"}:17        return False18    else:19        raise argparse.ArgumentTypeError("Invalid value for a boolean flag!")20def set_args(parser, additional=False, **kwargs):21    path_config(parser)22    run_config(parser)23    ### add by func24    add_func = kwargs.get("add_func", None)25    if add_func is not None:26        for each in add_func:27            logger.info(f'Args add: [{each}]')28            eval(each)(parser)29    args = parser.parse_args()30    return args31def path_config(parser):32    path_group = parser.add_argument_group("Path information and required dirs")33    path_group.add_argument("--data_dir",34                            default=None,35                            type=str,36                            required=True,37                            help="The input data dir. Should contain the .csv files (or other data files) for the task.")38    path_group.add_argument("--train_file", default=None, type=str)39    path_group.add_argument("--dev_file",   default=None, type=str)40    path_group.add_argument("--test_file",  default=None, type=str)41    path_group.add_argument("--use_newd", default=False, type=bool_flag)42    path_group.add_argument("--split_dev", default=False, type=bool_flag)43    path_group.add_argument("--output_dir",44                            default=None,45                            type=str,46                            required=True,47                            help="The output directory where the model checkpoints will be written.")48    path_group.add_argument("--log_file", default="log.out", type=str)49    path_group.add_argument("--tfboard_log_dir", default="event.out", type=str)50    path_group.add_argument("--result_eval_file", default="result.eval.txt", type=str)51    path_group.add_argument("--result_test_file", default="result.test.txt", type=str)52    path_group.add_argument("--result_trial_file", default="result.trial.txt", type=str)53    path_group.add_argument("--model_type", default="bert", type=str)54    path_group.add_argument("--model_name_or_path",55                            default=None,56                            type=str,57                            required=True,58                            help="Bert pre-trained model selected in the list: bert-base-uncased, "59                                "bert-large-uncased, bert-base-cased, bert-large-cased, bert-base-multilingual-uncased, "60                                "bert-base-multilingual-cased, bert-base-chinese.")61    path_group.add_argument("--tokenizer_name_or_path",62                            default=None,63                            type=str,64                            required=True,65                            help="Bert pre-trained model selected in the list: bert-base-uncased, "66                                "bert-large-uncased, bert-base-cased, bert-large-cased, bert-base-multilingual-uncased, "67                                "bert-base-multilingual-cased, bert-base-chinese.")68    path_group.add_argument("--config_name",69                            default=None,70                            type=str,71                            help="Pretrained config name or path if not the same as model_name")72    path_group.add_argument('--overwrite_output_dir', action='store_true',73                            help="Overwrite the content of the output directory")74    path_group.add_argument('--overwrite_cache', action='store_true',75                            help="Overwrite the cached input feature sets.")76def run_config(parser):77    run_group = parser.add_argument_group("Run configs")78    run_group.add_argument("--task_name", default='wic_pair',79                           type=str,80                           required=True)81    run_group.add_argument("--network_name", default='basic',82                           type=str)83    ### Run parameters84    run_group.add_argument("--max_seq_length", default=128,85                           type=int,86                           help="The maximum total input sequence length after WordPiece tokenization. \n"87                                "Sequences longer than this will be truncated, and sequences shorter \n"88                                "than this will be padded.")89    run_group.add_argument("--do_lower_case",90                           action='store_true',91                           help="Set this flag if you are using an uncased model.")92    run_group.add_argument("--cls_segment_id", default=0,93                           type=int)94    run_group.add_argument("--from_tf", action='store_true')95    ### Run Mode96    run_group.add_argument("--do_train",97                           action='store_true',98                           help="Whether to run training.")99    run_group.add_argument("--do_eval",100                           action='store_true',101                           help="Whether to run eval on the dev set.")102    run_group.add_argument("--do_test",103                           action='store_true',104                           help="Whether to run test on the test set.")105    run_group.add_argument("--do_trial",106                           action='store_true',107                           help="Whether to run test on the unofficial dev set.")108    run_group.add_argument("--have_test_label",109                           action='store_true',110                           help="Used when testing")111    ### Train parameters112    run_group.add_argument("--train_batch_size",113                           default=32,114                           type=int,115                           help="Total batch size for training.\n"116                                "Discarded.")117    run_group.add_argument("--per_gpu_train_batch_size",118                           default=8,119                           type=int,120                           help="Batch size per GPU/CPU for training.")121    run_group.add_argument("--eval_batch_size",122                           default=8,123                           type=int,124                           help="Total batch size for eval. \n"125                                "Discarded.")126    run_group.add_argument("--per_gpu_eval_batch_size",127                           default=8,128                           type=int,129                           help="Batch size per GPU/CPU for evaluation.")130    run_group.add_argument("--learning_rate",131                           default=5e-5,132                           type=float,133                           help="The initial learning rate for Adam.")134    run_group.add_argument("--adam_epsilon",135                           # default=1e-8,136                           default=1e-6,137                           type=float,138                           help="Epsilon for Adam optimizer.")139    run_group.add_argument("--num_train_epochs",140                           default=3.0,141                           type=float,142                           help="Total number of training epochs to perform.")143    run_group.add_argument("--max_steps",144                           default=-1,145                           type=int,146                           help="If > 0: set total number of training steps to perform. Override num_train_epochs.")147    run_group.add_argument("--warmup_proportion",148                           default=0.1,149                           type=float,150                           help="Proportion of training to perform linear learning rate warmup for. "151                                "E.g., 0.1 = 10%% of training.")152    run_group.add_argument("--warmup_steps",153                           default=0, type=int,154                           help="Linear warmup over warmup_steps.")155    run_group.add_argument("--weight_decay",156                           # default=0.0,157                           default=0.01,158                           type=float,159                           help="Weight deay if we apply some.")160    run_group.add_argument('--gradient_accumulation_steps',161                           type=int,162                           default=1,163                           help="Number of updates steps to accumulate before performing a backward/update pass.")164    run_group.add_argument("--max_grad_norm",165                           default=1.0, # default is 1.0166                           type=float,167                           help="Max gradient norm.")168    run_group.add_argument('--loss_scale',169                           type=float, default=0,170                           help="Loss scaling to improve fp16 numeric stability. Only used when fp16 set to True.\n"171                                "0 (default value): dynamic loss scaling.\n"172                                "Positive power of 2: static loss scaling value.\n")173    run_group.add_argument('--lm_coef', type=float, default=0.9,174                           help="parameter to balance lm loss and task loss for GPT/GPT2")175    run_group.add_argument('--add_loss_coef', type=float, default=1.0,176                           help="parameter to balance main loss and additional loss for Task")177    ### Environment178    run_group.add_argument("--no_cuda",179                           action='store_true',180                           help="Whether not to use CUDA when available")181    run_group.add_argument("--local_rank",182                           default=-1,183                           type=int,184                           help="local_rank for distributed training on gpus")185    run_group.add_argument('--seed',186                           default=42,187                           type=int,188                           help="random seed for initialization")189    run_group.add_argument('--fp16',190                           action='store_true',191                           help="Whether to use 16-bit float precision instead of 32-bit")192    run_group.add_argument('--fp16_opt_level',193                           type=str, default='O1',194                           help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."195                                 "See details at https://nvidia.github.io/apex/amp.html")196    ### Others197    run_group.add_argument('--logging_steps', type=int, default=100,198                           help="Log every X updates steps.")199    run_group.add_argument('--save_steps', type=int, default=0,200                           help="Save checkpoint every X updates steps.")201    run_group.add_argument("--eval_all_checkpoints", action='store_true',202                           help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number")203    run_group.add_argument("--evaluate_during_training", action='store_true',204                           help="Rul evaluation during training at each logging step.")205    run_group.add_argument("--evaluate_epoch", action='store_true',206                           help="Rul evaluation during training at each logging step.")207    ### Task specific208    run_group.add_argument("--output_mode",209                           default="classification",210                           type=str)211    run_group.add_argument("--num_choices", default=2,212                           type=int)213    run_group.add_argument("--have_passage",214                           action='store_true',215                           help="if example have context passage")216def add_args(now_args, additional_args):217    now_args, additional_args = vars(now_args), vars(additional_args)218    for k,v in additional_args.items():219        if k not in now_args:220            now_args[k] = v221            logger.info("Update additional config {}: {}".format(k,v))222        else:223            if v != now_args[k]:224                logger.info("Warn: additional config {}: {}/{} exist.".format(k, now_args[k], v))225    return argparse.Namespace(**now_args)226def check_args_version(load_args, now_args):227    load_args, now_args = vars(load_args), vars(now_args)228    for k, v in now_args.items():229        if k not in load_args:230            load_args[k] = v231            logger.info("Update load checkpoint config {}: {}".format(k,v))232    return argparse.Namespace(**load_args)233def override_args(old_args, new_args):234    KEEP_CONFIG = {}235    old_args, new_args = vars(old_args), vars(new_args)236    for k in old_args.keys():237        if k in new_args and old_args[k] != new_args[k]:238            if k in KEEP_CONFIG:239                logger.info("Overriding saved {}: {} --> {}".format(k, old_args[k], new_args[k]))240                old_args[k] = new_args[k]241            else:242                logger.info("Keeping saved {}: {}".format(k, old_args[k]))243    return  argparse.Namespace(**old_args)244def task_lm_finetune(parser):245    lm_task_group = parser.add_argument_group('Task configs: lm fine-tune')246    ### for lm finetuning247    lm_task_group.add_argument("--mlm",248                               action='store_true',249                               help="Train with masked-language modeling loss instead of language modeling.")250    lm_task_group.add_argument("--mlm_probability",251                               type=float,252                               default=0.15,253                               help="Ratio of tokens to mask for masked language modeling loss")254    lm_task_group.add_argument("--block_size",255                               default=-1,256                               type=int,257                               help="Optional input sequence length after tokenization."258                                    "The training dataset will be truncated in block of this size for training."259                                    "Default to the model max input length for single sentence inputs "260                                    "(take into account special tokens).")261    lm_task_group.add_argument('--save_total_limit',262                               type=int,263                               default=None,264                               help='Limit the total amount of checkpoints, delete the older checkpoints in the output_dir, '265                                    'does not delete by default')266    lm_task_group.add_argument("--cache_dir", default="", type=str,267                               help="Optional directory to store the pre-trained models downloaded from s3 "...

plot_network_strain.py

Source:plot_network_strain.py

1import argparse2from copy import deepcopy3import pandas as pd4import h5py5import matplotlib.pyplot as plt6import numpy as np7import seaborn as sns8from scipy import stats9import matplotlib.pylab as pylab10params = {'axes.titlesize':'x-large',11        'axes.labelsize': 'x-large'}12pylab.rcParams.update(params)13DIRECTORY = 'final/plots/' # TODO change this as needed14def plot_failures(failure_points, lowest, highest):15  16    print(len(failure_points))17 18    plt.figure()19    20    plt.xlim(0, 18.5)21    plt.hist(failure_points, bins=np.arange(0.5, 19.5, 1), cumulative=True, color='orange', edgecolor='k')22    23    plt.xticks(np.arange(1, 19))24    plt.ylabel('% Networks Failed')25    plt.xlabel('Total Task Count')26    plt.savefig('{}failures.pdf'.format(DIRECTORY), dpi=300, format='pdf')27def plot_strain(run_groups, metric):28    29    plt.figure()30    for run_group in run_groups:31        plt.plot(run_group[1], label=run_group[0])32    33    plt.legend(loc='upper left', fancybox=True, shadow=True)34    # if metric == 'Final Training Iteration Loss':35    #     plt.ylim(0, 10)36    plt.ylabel(metric)37    plt.xlabel('Task')38    plt.savefig('{}{}.pdf'.format(DIRECTORY, metric), dpi=300, format='pdf')39def parse_h5(filename):40    41    f = h5py.File(filename, 'r')                                                                                            42    failure = f['failure'][0]43    total = []44    st_dev = []45    avg = []46    maximum = []47    loss = []48    fisher_information = []49    for data in f['fisher_total']:50        total.append(data)51    for data in f['post_training_loss']:52        loss.append(data)53    for data in f['fisher_average']:54        avg.append(data)55    for data in f['fisher_st_dev']:56        st_dev.append(data)57    for data in f['fisher_max']:58        maximum.append(data)59    # for data in f['fisher_information']:60    #     fisher_information.append([])61    #     for task in data:62    #         fisher_information[len(fisher_information) - 1].append(task)63    f.close()64    return (failure, total), (failure, st_dev), (failure, avg), (failure, maximum), (failure, loss)#, (failure, fisher_information)65def plot_fisher_dist(run_group):66    plt.figure()67    tasks = []68    for fi_data in run_group[1]:69        tasks.append(fi_data)70    # bins = np.arange(0, 2.5, 0.01)71    plt.hist(tasks, bins=[15, 16, 17, 18], label=np.arange(0, run_group[0] + 1))72    plt.xticks([15, 16, 17, 18])73    # for i, task in enumerate(tasks):74    #     sns.distplot(task)75    plt.legend(loc='upper right', fancybox=True, shadow=True)76    plt.savefig('{}fisher_distribution_failure_at_{}.eps'.format(DIRECTORY, run_group[0]), dpi=300, format='eps')77def main():78    #sns.set(color_codes=True)79    parser = argparse.ArgumentParser(description='Plotting Tool')80    parser.add_argument('--filenames',81            nargs='+', type=str, default=['NONE'], metavar='FILENAMES',82            help='names of .h5 files containing experimental result data')83    args = parser.parse_args()84    runs = []85    for filename in args.filenames:86        runs.append([])87        #total, st_dev, avg, maximum, loss, fisher_information = parse_h5(filename)88        total, st_dev, avg, maximum, loss = parse_h5(filename)89        runs[len(runs) - 1].append(total)90        runs[len(runs) - 1].append(st_dev)91        runs[len(runs) - 1].append(avg)92        runs[len(runs) - 1].append(maximum)93        runs[len(runs) - 1].append(loss)94        #runs[len(runs) - 1].append(fisher_information)95    failure_points = []96    for data in runs:97        failure_points.append(data[0][0])98    highest = np.amax(failure_points)99    100    lowest = np.amin(failure_points)101    102    plot_failures(failure_points, lowest, highest)103   #  104   #  metrics = ['Sum of Fisher Information','Standard Deviation of Fisher Information','Average of Fisher Information',105   #             'Maximum Fisher Information Value','Final Training Iteration Loss']106   #  for strain_metric in range(5):107   #      # strain_per_task is an array organized like so:108   #      # [109   #      # [0, []]               row 0110   #      # [c1, [x1, x2, x3]]    row 1: [# of runs failed at task 1,111   #      #                               average network strain per task (index) for runs ending at task 1]112   #      # [c2, [y1, y2, y3]]    row 2: [# of runs failed at task 2,113   #      #                               average network strain per task (index) for runs ending at task 2]114   #      # ...115   #      # ]116   #      strain_per_task = []117   #      for i in np.arange(0, highest+1):118   #          strain_per_task.append([0, np.zeros(i)])119   #      for data in runs:120   #          for t in range(len(data[strain_metric][1])):121   #              strain_per_task[data[strain_metric][0]][1][t] += data[strain_metric][1][t]122   #          strain_per_task[data[strain_metric][0]][0] += 1123   #      for row in range(len(strain_per_task)):124   #          for i in range(len(strain_per_task[row][1])):125   #              if strain_per_task[row][0] != 0:126   #                  strain_per_task[row][1][i] /= strain_per_task[row][0]127   #      print(strain_per_task)128   #      run_groups = []129   #      for row in range(len(strain_per_task)):130   #          if strain_per_task[row][0] > 0:131   #              run_groups.append((row, strain_per_task[row][1]))132   #      print(run_groups)133   #      metric = metrics[strain_metric]134   #      plot_strain(run_groups, metric)135    # plot summed fisher info distribution136    # fisher_summed = []137    #138    # print(len(runs[0][5][1][1]))139    #140    #141    # for i in np.arange(0, highest + 1):142    #     fisher_summed.append([0, np.zeros((i, len(runs[0][5][1][1])))])143    #144    # for data in runs:145    #     for t in range(len(data[5][1])):146    #         for fi in range(len(data[5][1][t])):147    #             fisher_summed[data[5][0]][1][t][fi] += data[5][1][t][fi]148    #149    #     fisher_summed[data[5][0]][0] += 1150    #151    # # average fisher summed info for each task for each run group152    # for row in range(len(fisher_summed)):153    #     for task in range(len(fisher_summed[row][1])):154    #         for fisher_info in range(len(fisher_summed[row][1][task])):155    #             if fisher_summed[row][0] != 0:156    #                 fisher_summed[row][1][task][fisher_info] /= fisher_summed[row][0]157    #158    #159    # run_groups = []160    #161    # for row in range(len(fisher_summed)):162    #     if fisher_summed[row][0] > 0:163    #         run_groups.append((row, fisher_summed[row][1]))164    #165    # print(run_groups)166    #167    # # run groups is now [...(failure_point, [[fisher info task 0][fi t1][fi t2]...])...]168    # for group in run_groups:169    #     plot_fisher_dist(group)170if __name__ == '__main__':...

Automation Testing Tutorials

Learn to execute automation testing from scratch with LambdaTest Learning Hub. Right from setting up the prerequisites to run your first automation test, to following best practices and diving deeper into advanced test scenarios. LambdaTest Learning Hubs compile a list of step-by-step guides to help you be proficient with different test automation frameworks i.e. Selenium, Cypress, TestNG etc.