How to use fastprogress - 10 common examples

def fit(self, epochs, lr, validate=True, schedule_type="warmup_linear"): num_train_steps = int(len(self.data.train_dl) / self.grad_accumulation_steps * epochs) if self.optimizer is None: self.optimizer, self.schedule = self.get_optimizer(lr , num_train_steps) t_total = num_train_steps if self.multi_gpu == False: t_total = t_total // torch.distributed.get_world_size() global_step = 0  pbar = master_bar(range(epochs)) for epoch in pbar: self.model.train() tr_loss = 0 nb_tr_examples, nb_tr_steps = 0, 0 for step, batch in enumerate(progress_bar(self.data.train_dl, parent=pbar)): batch = tuple(t.to(self.device) for t in batch) input_ids, input_mask, segment_ids, label_ids = batch if self.is_fp16 and self.multi_label: label_ids = label_ids.half() loss = self.model(input_ids, segment_ids, input_mask, label_ids) if self.multi_gpu:

t_total = num_train_steps if self.multi_gpu == False: t_total = t_total // torch.distributed.get_world_size() global_step = 0 pbar = master_bar(range(epochs)) for epoch in pbar: self.model.train() tr_loss = 0 nb_tr_examples, nb_tr_steps = 0, 0  for step, batch in enumerate(progress_bar(self.data.train_dl, parent=pbar)): batch = tuple(t.to(self.device) for t in batch) input_ids, input_mask, segment_ids, label_ids = batch if self.is_fp16 and self.multi_label: label_ids = label_ids.half() loss = self.model(input_ids, segment_ids, input_mask, label_ids) if self.multi_gpu: loss = loss.mean() # mean() to average on multi-gpu. if self.grad_accumulation_steps > 1: loss = loss / self.grad_accumulation_steps if self.is_fp16: self.optimizer.backward(loss) else: loss.backward()

visual_activity_concept = visual_activity_concept.cpu().detach().numpy() # convert matrix to vector : (1,outdim) --> (outdim,) visual_feature = visual_feature.reshape(visual_feature.shape[1],) visual_activity_concept = visual_activity_concept.reshape(visual_activity_concept.shape[1],) # save feature np.save(os.path.join(visual_feature_out_dir, name), visual_feature) np.save(os.path.join(activity_concept_out_dir, activity_name(name)), visual_activity_concept) if args.verbose: print('save {}.npy'.format(name)) # get movie name movie_names = [] for movie in root_dir: movie_names.append(movie.split('/')[-1])  for movie_name in progress_bar(movie_names): pool_feature(movie_name, activity_concept_out_dir, pooled_activity_concept_out_dir)

if verbose: print("Optimising RF parameters") rfp, rf = get_opt_rf_params(train_df[train_feats], train_df[targ_name], val_df[train_feats], val_df[targ_name], objective, w_trn=w_trn, w_val=w_val, n_estimators=n_estimators, params=rf_params, verbose=False) else: rfp = rf_params rfp['n_estimators'] = n_estimators m = RandomForestClassifier if 'class' in objective.lower() else RandomForestRegressor rf = m(**rfp) rf.fit(X=train_df[train_feats], y=train_df[targ_name], sample_weight=w_trn) if verbose: print("Evalualting importances") fi = get_rf_feat_importance(rf, train_df[train_feats], train_df[targ_name], w_trn) orig_score = [rf.score(X=val_df[train_feats], y=val_df[targ_name], sample_weight=w_val)] if n_rfs > 1: m = RandomForestClassifier if 'class' in objective.lower() else RandomForestRegressor  for _ in progress_bar(range(n_rfs-1)): rf = m(**rfp) rf.fit(X=train_df[train_feats], y=train_df[targ_name], sample_weight=w_trn) fi = pd.merge(fi, get_rf_feat_importance(rf, train_df[train_feats], train_df[targ_name], w_trn), on='Feature', how='left') orig_score.append(rf.score(X=val_df[train_feats], y=val_df[targ_name], sample_weight=w_val)) fi['Importance'] = np.mean(fi[[f for f in fi.columns if 'Importance' in f]].values, axis=1) fi['Uncertainty'] = np.std(fi[[f for f in fi.columns if 'Importance' in f]].values, ddof=1, axis=1)/np.sqrt(n_rfs) fi.sort_values(by='Importance', ascending=False, inplace=True) orig_score = uncert_round(np.mean(orig_score), np.std(orig_score, ddof=1)) if verbose: print("Top ten most important features:\n", fi[['Feature', 'Importance']][:min(len(fi), 10)]) if plot_results: plot_importance(fi[:min(len(fi), n_max_display)], threshold=importance_cut, savename=savename, settings=plot_settings) top_feats = list(fi[fi.Importance >= importance_cut].Feature) if verbose: print(f"\n{len(top_feats)} features found with importance greater than {importance_cut}:\n", top_feats, '\n') if len(top_feats) == 0: if verbose: print(f"Model score: :\t{orig_score[0]}±{orig_score[1]}") print('No features found to be important, returning all training features. Good luck.')

def parallel(func, arr, max_workers=None): with ProcessPoolExecutor(max_workers=max_workers) as ex: futures = [ex.submit(func, o, i) for i, o in enumerate(arr)] results = []  for f in progress_bar(as_completed(futures), total=len(arr)): results.append(f.result()) return results

w_trn: training weights w_val: validation weights params: ordered dictionary mapping parameters to optimise to list of values to cosnider n_estimators: number of trees to use in each forest verbose: Print extra information and show a live plot of model performance Returns: params: dictionary mapping parameters to their optimised values rf: best performing Random Forest ''' if params is None: params = OrderedDict({'min_samples_leaf': [1,3,5,10,25,50,100], 'max_features': [0.3,0.5,0.7,0.9]}) rf = RandomForestClassifier if 'class' in objective.lower() else RandomForestRegressor best_params = {'n_estimators': n_estimators, 'n_jobs': -1, 'max_features':'sqrt'} best_scores = [] scores = []  mb = master_bar(params) mb.names = ['Best', 'Scores'] if verbose: mb.update_graph([[[],[]], [[], []]]) for param in mb: pb = progress_bar(params[param], parent=mb) pb.comment = f'{param} = {params[param][0]}' for i, value in enumerate(pb): pb.comment = f'{param} = {params[param][min(i+1, len(params[param])-1)]}' m = rf(**{**best_params, param: value}) m.fit(X=x_trn, y=y_trn, sample_weight=w_trn) scores.append(m.score(X=x_val, y=y_val, sample_weight=w_val)) if len(best_scores) == 0 or scores[-1] > best_scores[-1]: best_scores.append(scores[-1]) best_params[param] = value if verbose: print(f'Better score schieved: {param} @ {value} = {best_scores[-1]:.4f}') best_m = m else:

def begin_fit(self):  self.mbar = master_bar(range(self.epochs)) self.mbar.on_iter_begin() self.run.logger = partial(self.mbar.write, table=True)

def show_install(show_nvidia_smi:bool=False): "Print user's setup information" import platform, fastai.version rep = [] opt_mods = [] rep.append(["=== Software ===", None]) rep.append(["python", platform.python_version()]) rep.append(["fastai", fastai.__version__])  rep.append(["fastprogress", fastprogress.__version__]) rep.append(["torch", torch.__version__]) # nvidia-smi cmd = "nvidia-smi" have_nvidia_smi = False try: result = subprocess.run(cmd.split(), shell=False, check=False, stdout=subprocess.PIPE) except: pass else: if result.returncode == 0 and result.stdout: have_nvidia_smi = True # XXX: if nvidia-smi is not available, another check could be: # /proc/driver/nvidia/version on most systems, since it's the # currently active version if have_nvidia_smi: smi = result.stdout.decode('utf-8')

def show_install(show_nvidia_smi:bool=False): "Print user's setup information: python -c 'import fastai; fastai.show_install()'" import platform, fastai.version rep = [] opt_mods = [] rep.append(["=== Software ===", None]) rep.append(["python", platform.python_version()]) rep.append(["fastai", fastai.__version__])  rep.append(["fastprogress", fastprogress.__version__]) rep.append(["torch", torch.__version__]) # nvidia-smi cmd = "nvidia-smi" have_nvidia_smi = False try: result = subprocess.run(cmd.split(), shell=False, check=False, stdout=subprocess.PIPE) except: pass else: if result.returncode == 0 and result.stdout: have_nvidia_smi = True # XXX: if nvidia-smi is not available, another check could be: # /proc/driver/nvidia/version on most systems, since it's the # currently active version

def fit(self, epochs: int, lr: float, params_opt_dict: Optional[Dict] = None): "Main training loop" # Print logger at the start of the training loop self.logger.info(self.cfg) # Initialize the progress_bar  mb = master_bar(range(epochs)) # Initialize optimizer # Prepare Optimizer may need to be re-written as per use self.optimizer = self.prepare_optimizer(params_opt_dict) # Initialize scheduler # Prepare scheduler may need to re-written as per use self.lr_scheduler = self.prepare_scheduler(self.optimizer) # Write the top row display # mb.write(self.log_keys, table=True) self.master_bar_write(mb, line=self.log_keys, table=True) exception = False met_to_use = None # Keep record of time until exit st_time = time.time() try: # Loop over epochs