Kansallinen Liiga Qualification Group stats & predictions

Storia Recente: Performance Precedenti nelle Qualificazioni

Ecco un breve riassunto delle performance delle squadre nelle qualificazioni passate:

• Helsinki United: Hanno raggiunto la finale nelle ultime due stagioni ma non sono riusciti a qualificarsi. Quest'anno vogliono cambiare rotta.
• Tampere Wolves: Hanno sempre dimostrato solidità difensiva ma spesso hanno faticato ad andare oltre le semifinali. Quest'anno mirano al titolo.
• Porvoo FC: Nelle ultime stagioni hanno faticato ma quest'anno hanno mostrato grande miglioramento e vogliono approfittarne al massimo.narendran-k/estimator<|file_sep|>/src/estimator/sampler.py # Copyright (c) Microsoft Corporation. # Licensed under the MIT License. import numpy as np import torch from torch.distributions import Categorical from estimator import utils def sample(model: torch.nn.Module, n_samples: int, *, device: torch.device = None, sample_from_prior: bool = False, **kwargs) -> dict: """Sample from the model using either the prior or posterior.""" if device is None: device = utils.get_default_device() # Initialize dictionary to hold samples. samples = {} # Get latent distributions. latent_dist = model.latent_dist # Sample from the prior if requested. if sample_from_prior: if hasattr(latent_dist, 'prior'): latent_dist = latent_dist.prior else: raise ValueError("Model has no prior.") # Sample from the prior. z = latent_dist.sample((n_samples,)) z = z.to(device) # Update samples. samples['z'] = z # If the model uses inference inputs to condition the prior, # then we need to sample from the prior with an empty batch. if model.inference_inputs is not None: samples.update(sample(model=model, n_samples=1, device=device, sample_from_prior=True)) # Remove single-sample inference input samples from dictionary. for k in model.inference_inputs: del samples[k] # Add indices to keep track of which inference inputs were used. indices = Categorical(probs=torch.ones(n_samples)).sample() samples['inference_indices'] = indices # Replicate inference inputs according to indices. for k in model.inference_inputs: n_inference_samples = len(samples[k]) replicated_inference_input = [] for idx in indices.tolist(): replicated_inference_input.append(samples[k][idx]) samples[k] = torch.stack(replicated_inference_input).to(device) del samples['inference_indices'] return samples # Get observation distribution. obs_dist = model.observation_dist # Get observation inputs. obs_inputs = kwargs.get('obs_inputs') # Check if observation inputs were passed and raise error if they were not. if obs_inputs is None: raise ValueError("Observation inputs must be provided.") # Add observation inputs to sample dictionary. for k in obs_inputs: if isinstance(obs_inputs[k], np.ndarray): obs_inputs[k] = torch.from_numpy(obs_inputs[k]).to(device) else: obs_inputs[k] = obs_inputs[k].to(device) samples[k] = obs_inputs[k] n_obs_samples = len(samples[obs_inputs.keys()[0]]) # Sample latent variables and add them to the sample dictionary. z = latent_dist.sample((n_obs_samples,)) z = z.to(device) samples['z'] = z # Generate observations and add them to the sample dictionary. x_params = obs_dist.sample(z=z, **samples) x = obs_dist.rsample(**x_params) x_reshape_fn = getattr(obs_dist.distribution, 'reshape_sample', lambda x: x) x_reshape_fn(x) samples['x'] = x return samples <|repo_name|>narendran-k/estimator<|file_sep|>/src/estimator/models/vae.py # Copyright (c) Microsoft Corporation. # Licensed under the MIT License. import logging from abc import ABCMeta import numpy as np import torch import torch.nn as nn import torch.nn.functional as F from pytorch_lightning.core.lightning import LightningModule from pytorch_lightning.metrics.functional import accuracy from estimator import utils from estimator.models.base import EstimatorBase logger = logging.getLogger(__name__) class VAE(EstimatorBase): def __init__(self, encoder, decoder, encoder_kwargs=None, decoder_kwargs=None, latent_dim=32, activation='relu', **kwargs): super().__init__(**kwargs) self.encoder_kwargs = encoder_kwargs or {} self.decoder_kwargs = decoder_kwargs or {} self.latent_dim = latent_dim self.activation_fn = utils.get_activation(activation) self.encoder_nn: nn.ModuleList = nn.ModuleList([encoder(**self.encoder_kwargs)]) self.decoder_nn: nn.ModuleList = nn.ModuleList([decoder(latent_dim=self.latent_dim, **self.decoder_kwargs)]) self._log_var_transform_fn = lambda logvar: logvar * 0.5 self.register_buffer('_logvar_eps', torch.Tensor([utils.EPS])) def encode(self, *args, **kwargs): x_in_shape_origs = [x.shape for x in args] x_in_shape_origs += [kwargs[k].shape for k in kwargs] kwargs.update(dict(zip(self.encoder_nn[0].input_keys, args))) kwargs.update({k:v for k,v in zip(self.encoder_nn[0].input_keys[len(args):], kwargs.values())}) x_in_shapes_flatten_len_origs = [np.prod(x.shape) for x in args] x_in_shapes_flatten_len_origs += [np.prod(x.shape) for x in kwargs.values()] encoded_flatten_len_origs_list_flat_list_of_lists_list_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_of_lists_list= [] encoded_flatten_len_origs_list_flat_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list_of_lists_list= [] encoded_flatten_len_origs_list_flat_list= [] encoded_flatten_len_origs= [] encoded_shapes_origs= [] encoded_flatten_len_origs_list_flat= [] encoded_flatten_len_origs_list= [] encoded_shapes_origs_list= [] encoded_shapes_flatten_len_origs= [] for enc_i,(enc_layer_i,x_in_shape,x_in_shapes_flatten_len) in enumerate(zip(self.encoder_nn[0].layers,x_in_shape_origs,x_in_shapes_flatten_len_origs)): encoded_flatten_len_new= enc_layer_i.get_encoded_flatten_len(x_in_shapes_flatten_len) encoded_shapes_new= enc_layer_i.get_encoded_shapes(x_in_shape) encoded_shapes_flatten_len_new= [np.prod(s) for s in encoded_shapes_new] encoded_flatten_len_origs.append(encoded_flatten_len_new) encoded_shapes_origs.append(encoded_shapes_new) encoded_shapes_flatten_len